相关申请的交叉引用CROSS-REFERENCE TO RELATED APPLICATIONS
本申请要求于2020年12月21日提交的美国申请63/128764和于2021年3月10日提交的美国申请63/159,389的优先权，这些申请通过引用整体并入本文。This application claims priority to U.S. Application No. 63/128,764, filed on December 21, 2020, and U.S. Application No. 63/159,389, filed on March 10, 2021, which are incorporated herein by reference in their entirety.
技术领域Technical Field
本文中提供的实施例涉及半导体制造，并且更具体地涉及半导体晶片的失败分析。Embodiments provided herein relate to semiconductor manufacturing, and more particularly, to failure analysis of semiconductor wafers.
背景技术Background Art
在集成电路(IC)的制造过程中，对未完成或已完成的电路组件进行检查以确保它们是根据设计而制造的并且没有缺陷。可以采用利用光学显微镜或带电粒子(例如，电子)束显微镜的检查系统。例如，带电粒子(例如，电子)束显微镜(诸如扫描电子显微镜(SEM)或透射电子显微镜(TEM))可以用作检查IC组件的实用工具。During the manufacture of integrated circuits (ICs), unfinished or completed circuit components are inspected to ensure that they are manufactured according to design and are free of defects. Inspection systems utilizing optical microscopes or charged particle (e.g., electron) beam microscopes may be employed. For example, charged particle (e.g., electron) beam microscopes such as scanning electron microscopes (SEMs) or transmission electron microscopes (TEMs) may be used as practical tools for inspecting IC components.
从SEM或TEM图像测量的图案或结构的临界尺寸可以用于检测所制造的IC的缺陷。例如，图案之间的偏移或边缘放置变化可以有助于控制制造过程以及确定缺陷。随着IC组件的物理尺寸不断缩小，缺陷检测的准确性和产率变得更加重要。The critical dimensions of patterns or structures measured from SEM or TEM images can be used to detect defects in manufactured ICs. For example, offset or edge placement variations between patterns can help control the manufacturing process as well as identify defects. As the physical size of IC components continues to shrink, the accuracy and yield of defect detection become more important.
半导体微芯片制造包括数百个步骤。根本原因分析很重要，因为它有助于通过标识重要步骤和研究缺陷原因来优化制造过程。Semiconductor microchip manufacturing includes hundreds of steps. Root cause analysis is important because it helps optimize the manufacturing process by identifying important steps and studying the causes of defects.
发明内容Summary of the invention
本公开的一个方面涉及一种分析第一晶片上的第一区域的输入电子显微镜图像的方法。该方法包括从输入电子显微镜图像获取与多个可解释模式相对应的多个模式图像。该方法还包括评估多个模式图像，以及基于评估结果确定多个可解释模式对输入电子显微镜图像的贡献。该方法还包括基于所确定的贡献来预测第一晶片上的第一区域中的一个或多个特性。One aspect of the present disclosure relates to a method for analyzing an input electron microscope image of a first region on a first wafer. The method includes acquiring a plurality of pattern images corresponding to a plurality of interpretable patterns from the input electron microscope image. The method also includes evaluating the plurality of pattern images, and determining contributions of the plurality of interpretable patterns to the input electron microscope image based on the evaluation results. The method also includes predicting one or more characteristics in the first region on the first wafer based on the determined contributions.
本公开的另一方面涉及一种用于分析第一晶片上的第一区域的输入电子显微镜图像的装置。该装置包括存储指令集的存储器；以及至少一个处理器，该至少一个处理器被配置为执行该指令集以使得该装置执行：从输入电子显微镜图像获取与多个可解释模式相对应的多个模式图像，评估多个模式图，基于评估结果确定多个可解释模式对输入电子显微镜图像的贡献，以及基于所确定的贡献来预测第一晶片上的第一区域中的一个或多个特性。Another aspect of the present disclosure relates to an apparatus for analyzing an input electron microscope image of a first region on a first wafer. The apparatus includes a memory storing an instruction set; and at least one processor configured to execute the instruction set so that the apparatus performs: acquiring a plurality of pattern images corresponding to a plurality of interpretable patterns from the input electron microscope image, evaluating a plurality of pattern images, determining contributions of the plurality of interpretable patterns to the input electron microscope image based on the evaluation results, and predicting one or more characteristics in the first region on the first wafer based on the determined contributions.
本公开的另一方面涉及一种非暂态计算机可读介质，该介质存储指令集，该指令集由计算设备的至少一个处理器可执行以使得计算设备执行用于促进晶片检查的方法。该方法包括从输入电子显微镜图像获取与多个可解释模式相对应的多个模式图像；评估多个模式图像；基于评估结果确定多个可解释模式对输入电子显微镜图像的贡献；以及基于所确定的贡献来预测第一晶片上的第一区域中的一个或多个特性。Another aspect of the present disclosure relates to a non-transitory computer-readable medium storing an instruction set executable by at least one processor of a computing device to cause the computing device to perform a method for facilitating wafer inspection. The method includes acquiring a plurality of pattern images corresponding to a plurality of interpretable patterns from an input electron microscope image; evaluating the plurality of pattern images; determining contributions of the plurality of interpretable patterns to the input electron microscope image based on the evaluation results; and predicting one or more characteristics in a first region on a first wafer based on the determined contributions.
本公开的另一方面涉及一种训练用于对电子显微镜图像进行分类的分类器模型的方法。该方法包括获取多个晶片的训练电子显微镜图像；获取训练电子显微镜图像的标记数据，标记数据指示与训练电子显微镜图像中的每个训练电子显微镜图像相关联的多个可解释模式；以及基于训练电子显微镜图像和标记数据来训练分类器模型。Another aspect of the present disclosure relates to a method for training a classifier model for classifying electron microscope images. The method includes acquiring training electron microscope images of a plurality of wafers; acquiring labeled data for the training electron microscope images, the labeled data indicating a plurality of interpretable patterns associated with each of the training electron microscope images; and training a classifier model based on the training electron microscope images and the labeled data.
本公开的另一方面涉及一种用于训练用于对电子显微镜图像进行分类的分类器模型的装置。该装置包括存储指令集的存储器；以及至少一个处理器，该至少一个处理器被配置为执行该指令集以使得该装置执行：获取多个晶片的训练电子显微镜图像；获取训练电子显微镜图像的标记数据，标记数据指示与训练电子显微镜图像中的每个训练电子显微镜图像相关联的多个可解释模式；以及基于训练电子显微镜图像和标记数据来训练分类器模型。Another aspect of the present disclosure relates to an apparatus for training a classifier model for classifying electron microscope images. The apparatus includes a memory storing an instruction set; and at least one processor configured to execute the instruction set so that the apparatus performs: acquiring training electron microscope images of a plurality of wafers; acquiring label data of the training electron microscope images, the label data indicating a plurality of interpretable patterns associated with each of the training electron microscope images; and training a classifier model based on the training electron microscope images and the label data.
本公开的另一方面涉及一种非暂态计算机可读介质，该介质存储指令集，该指令集由计算设备的至少一个处理器可执行以使得计算设备执行训练用于对电子显微镜图像进行分类的分类器模型的方法。该方法包括获取多个晶片的训练电子显微镜图像；获取训练电子显微镜图像的标记数据，标记数据指示与训练电子显微镜图像中的每个训练电子显微镜图像相关联的多个可解释模式；以及基于训练电子显微镜图像和标记数据来训练分类器模型。Another aspect of the present disclosure relates to a non-transitory computer-readable medium storing an instruction set executable by at least one processor of a computing device to cause the computing device to perform a method of training a classifier model for classifying electron microscope images. The method includes acquiring training electron microscope images of a plurality of wafers; acquiring labeled data for the training electron microscope images, the labeled data indicating a plurality of interpretable patterns associated with each of the training electron microscope images; and training a classifier model based on the training electron microscope images and the labeled data.
本公开的另一方面涉及一种用于基于晶片的输入电子显微镜图像进行自动根本原因分析的方法。该方法包括：获取与输入电子显微镜图像相关联的输入数据，输入数据包括晶片的多个工艺特征；通过将多个预训练的决策树模型应用于多个工艺特征来从多个工艺特征中标识一组工艺特征；以及输出该组工艺特征的排名结果。Another aspect of the present disclosure relates to a method for automatic root cause analysis based on an input electron microscope image of a wafer. The method includes: obtaining input data associated with the input electron microscope image, the input data including a plurality of process features of the wafer; identifying a group of process features from the plurality of process features by applying a plurality of pre-trained decision tree models to the plurality of process features; and outputting a ranking result of the group of process features.
本公开的另一方面涉及一种用于基于晶片的输入电子显微镜图像进行自动根本原因分析的装置。该装置包括存储指令集的存储器；以及至少一个处理器，该至少一个处理器被配置为执行该指令集以使得该装置执行：获取与输入电子显微镜图像相关联的输入数据，输入数据包括晶片的多个工艺特征；通过将多个预训练的决策树模型应用于多个工艺特征来从多个工艺特征中标识一组工艺特征；以及输出该组工艺特征的排名结果。Another aspect of the present disclosure relates to an apparatus for performing automatic root cause analysis based on an input electron microscope image of a wafer. The apparatus includes a memory storing an instruction set; and at least one processor configured to execute the instruction set so that the apparatus performs: obtaining input data associated with the input electron microscope image, the input data including a plurality of process features of the wafer; identifying a group of process features from the plurality of process features by applying a plurality of pre-trained decision tree models to the plurality of process features; and outputting a ranking result of the group of process features.
本公开的另一方面涉及一种非暂态计算机可读介质，该介质存储指令集，该指令集由计算设备的至少一个处理器可执行以使得计算设备执行训练用于对电子显微镜图像进行分类的分类器模型的方法。该方法包括：获取与输入电子显微镜图像相关联的输入数据，输入数据包括晶片的多个工艺特征；通过将多个预训练的决策树模型应用于多个工艺特征来从多个工艺特征中标识一组工艺特征；以及输出该组工艺特征的排名结果。Another aspect of the present disclosure relates to a non-transitory computer-readable medium storing an instruction set executable by at least one processor of a computing device to cause the computing device to perform a method of training a classifier model for classifying an electron microscope image. The method includes: obtaining input data associated with an input electron microscope image, the input data including a plurality of process features of a wafer; identifying a group of process features from the plurality of process features by applying a plurality of pre-trained decision tree models to the plurality of process features; and outputting a ranking result of the group of process features.
通过以下结合附图的描述，本公开的实施例的其他优点将变得很清楚，在附图中通过说明和示例方式阐述了本公开的某些实施例。Other advantages of embodiments of the present disclosure will become apparent from the following description taken in conjunction with the accompanying drawings, in which certain embodiments of the present disclosure are illustrated and described by way of example.
附图说明BRIEF DESCRIPTION OF THE DRAWINGS
图1是示出与本公开的一些实施例一致的示例电子束检查(EBI)系统的示意图。1 is a schematic diagram illustrating an example electron beam inspection (EBI) system consistent with some embodiments of the present disclosure.
图2是示出与本公开的一些实施例一致的可以是图1的电子束检查系统的一部分的示例电子束工具的示意图。2 is a schematic diagram illustrating an example electron beam tool that may be part of the electron beam inspection system of FIG. 1 , consistent with some embodiments of the present disclosure.
图3是与本公开的一些实施例一致的与晶片分析和缺陷预测相关联的示例晶片分析系统的框图。3 is a block diagram of an example wafer analysis system associated with wafer analysis and defect prediction consistent with some embodiments of the present disclosure.
图4A是根据本公开的一些实施例的由训练图像获取器获取的一组训练图像的示例。FIG. 4A is an example of a set of training images acquired by a training image acquirer according to some embodiments of the present disclosure.
图4B是根据本公开的一些实施例的使用自动方法而处理的与图4A的该组训练图像相对应的一组标记图像的示例。4B is an example of a set of labeled images corresponding to the set of training images of FIG. 4A processed using an automatic method according to some embodiments of the present disclosure.
图5是与本公开的一些实施例一致的表示用于训练分类器模型的示例方法的过程流程图。5 is a process flow diagram representing an example method for training a classifier model, consistent with some embodiments of the present disclosure.
图6示出了与本公开的一些实施例一致的表示用于预测晶片上的缺陷类别的示例过程的流程图。6 shows a flow chart representing an example process for predicting defect classes on a wafer, consistent with some embodiments of the present disclosure.
图7示出了与本公开的一些实施例一致的可视化了将非线性分类器模型和线性化模型应用于被分解成两个可解释模式的多个扫描电子显微镜(SEM)图像的图。7 shows a diagram visualizing the application of a nonlinear classifier model and a linearized model to a plurality of scanning electron microscope (SEM) images decomposed into two interpretable modes, consistent with some embodiments of the present disclosure.
图8示出了根据本公开的一些实施例的通过对输入SEM图像执行处理而获取的可视化预测结果的示例。FIG. 8 shows an example of a visualized prediction result obtained by performing processing on an input SEM image according to some embodiments of the present disclosure.
图9A-图9E进一步示出了根据本公开的一些实施例的通过对各种输入SEM图像执行如图6中讨论的过程而获取的可视化预测结果(例如，如条形图)的示例。9A-9E further illustrate examples of visualized prediction results (eg, as bar graphs) obtained by performing the process as discussed in FIG. 6 on various input SEM images according to some embodiments of the present disclosure.
图10示出了根据本公开的一些实施例的使用逻辑分类器模型而获取的可视化预测结果的示例。FIG. 10 shows an example of a visualized prediction result obtained using a logical classifier model according to some embodiments of the present disclosure.
图11A是与本公开的一些实施例一致的表示用于基于输入SEM图像来预测失败模式的示例方法的过程流程图。11A is a process flow diagram representing an example method for predicting a failure mode based on an input SEM image, consistent with some embodiments of the present disclosure.
图11B示出了与本公开的一些实施例一致的可视化了基于多个SEM图像的可解释模式而进行的聚类和聚类结果在晶片上的映射的图。11B shows a diagram that visualizes clustering based on interpretable patterns of multiple SEM images and mapping of clustering results on a wafer, consistent with some embodiments of the present disclosure.
图11C是与本公开的一些实施例一致的表示用于基于晶片上的多个区域的输入SEM图像来分析失败模式的示例方法的过程流程图。11C is a process flow diagram representing an example method for analyzing failure modes based on input SEM images of multiple regions on a wafer, consistent with some embodiments of the present disclosure.
图12是与本公开的一些实施例一致的被配置为基于从预测模型获取的特征排名结果来执行根本原因分析的示例系统的框图。12 is a block diagram of an example system configured to perform root cause analysis based on feature ranking results obtained from a predictive model, consistent with some embodiments of the present disclosure.
图13示出了根据本公开的一些实施例的根据特征排名结果而进行的特征重要性的可视化的示例。FIG. 13 shows an example of visualization of feature importance based on feature ranking results according to some embodiments of the present disclosure.
图14是与本公开的一些实施例一致的表示用于执行自动根本原因分析的示例方法的过程流程图。14 is a process flow diagram representing an example method for performing automated root cause analysis, consistent with some embodiments of the present disclosure.
具体实施方式DETAILED DESCRIPTION
现在将详细参考示例性实施例，其示例如附图所示。以下描述引用附图，其中不同附图中的相同数字表示相同或相似的元素，除非另有说明。在示例性实施例的以下描述中阐述的实现并不表示所有实现。相反，它们仅仅是与所附权利要求中所述的公开实施例相关的方面一致的装置和方法的示例。例如，尽管在利用电子束的上下文中描述了一些实施例，但是本公开不限于此。可以类似地应用其他类型的带电粒子束。此外，可以使用其他成像系统，诸如光学成像、光检测、x射线检测等。Reference will now be made in detail to exemplary embodiments, examples of which are illustrated in the accompanying drawings. The following description refers to the accompanying drawings, in which the same numbers in different drawings represent the same or similar elements, unless otherwise specified. The implementations set forth in the following description of the exemplary embodiments do not represent all implementations. Instead, they are merely examples of apparatus and methods consistent with aspects related to the disclosed embodiments described in the appended claims. For example, although some embodiments are described in the context of utilizing electron beams, the present disclosure is not limited thereto. Other types of charged particle beams may be similarly applied. In addition, other imaging systems such as optical imaging, light detection, x-ray detection, etc. may be used.
制造极小的IC是一个复杂、耗时并且昂贵的过程，通常需要数百个个体步骤。即使是一个步骤中的错误也有可能导致成品IC出现缺陷，使其变得无用。因此，制造过程的一个目标是避免这样的缺陷，以使在工艺中制造的功能IC的数目最大化，即提高工艺的总产率。Manufacturing extremely small ICs is a complex, time-consuming, and expensive process that typically requires hundreds of individual steps. An error in even one step can cause a finished IC to be defective, rendering it useless. Therefore, one goal of the manufacturing process is to avoid such defects in order to maximize the number of functional ICs manufactured in the process, i.e., to increase the overall yield of the process.
提高产率的一个组成部分是监测芯片制造过程，以确保其生产足够数目的功能集成电路。监测该过程的一种方法是在芯片电路结构形成的各个阶段对其进行检查。可以使用扫描电子显微镜(SEM)进行检查。An integral part of improving yield is monitoring the chip manufacturing process to ensure that it is producing a sufficient number of functional integrated circuits. One way to monitor the process is to inspect the chip circuit structure at various stages of its formation. This inspection can be done using a scanning electron microscope (SEM).
在EUV光刻中，随机印刷失败成为工艺窗口的主要限制因素，这些失败是随机的、不重复的、孤立的缺陷，诸如微桥、局部断线、以及缺失或合并触点。这些缺陷可以在制造后使用蚀刻后SEM图像来检测。此外，预测模型也可以用于实现基于光刻后晶片的SEM图像对蚀刻后缺陷的早期预测，这可以进一步允许基于预测进行早期校正。现有缺陷预测基于：基于参数的方法(例如，诸如MetroLER或Stochalis)或机器学习模型。然而，现有方法要么预测精度低，要么不能提供与(多个)缺陷原因相关的信息。In EUV lithography, random printing failures become the main limiting factor of the process window. These failures are random, non-repeating, isolated defects such as microbridges, localized wire breaks, and missing or merged contacts. These defects can be detected after manufacturing using post-etch SEM images. In addition, the prediction model can also be used to achieve early prediction of post-etch defects based on SEM images of post-lithography wafers, which can further allow early correction based on prediction. Existing defect prediction is based on: parameter-based methods (e.g., such as MetroLER or Stochalis) or machine learning models. However, existing methods either have low prediction accuracy or cannot provide information related to the cause of (multiple) defects.
本公开的实施例包括训练和应用用于失败分类的机器学习模型，该模型可以有利地实现所生成的预测的可解释性。在训练阶段，由专家提出或自动计算一组可解释模式(例如，感兴趣模式)，诸如CD、偏移、椭圆度等，并且在这些模式上训练(例如，深度学习)分类器。在预测阶段，将给定SEM图像分解为可解释模式，并且应用分类器。分类器的数学近似(例如，加权多项式近似)被用于将预测分解为来自个体可解释模式的贡献。根据对模式的各个贡献的分析，可以标识出失败原因，例如，可能的原因是小CD、强y偏移、椭圆度、边缘模糊等中的一种或多种。Embodiments of the present disclosure include training and applying a machine learning model for failure classification, which can advantageously achieve interpretability of the generated predictions. In the training phase, a set of interpretable patterns (e.g., patterns of interest), such as CD, offset, ellipticity, etc., are proposed by experts or automatically calculated, and a classifier is trained (e.g., deep learning) on these patterns. In the prediction phase, a given SEM image is decomposed into interpretable patterns and a classifier is applied. A mathematical approximation of the classifier (e.g., a weighted polynomial approximation) is used to decompose the prediction into contributions from individual interpretable patterns. Based on the analysis of the individual contributions of the pattern, the cause of the failure can be identified, for example, the possible cause is one or more of small CD, strong y offset, ellipticity, edge blur, etc.
根据本公开的实施例，首先，基于蚀刻后晶片上的特定类型的特征(例如，接触孔区域或任何其他类型的特征)的SEM图像、以及相关联的标记数据，来训练机器学习模型，诸如神经网络模型，标记数据指示可以用于表征晶片上的各种缺陷类别的多个可解释模式的各个系数。在预测阶段期间，可以将显影后晶片的输入SEM图像分解为与多个预定义可解释模式(例如，对应于缺陷的不同原因、类型或类别)相关联的多个模式图像。已训练机器学习模型可以用于评估分解后的模式图像。例如，与可解释模式相关联的各个系数可以被用作机器学习模型的输入，并且机器学习模式的输出可以包括评估结果，例如，评估结果指示在输入SEM图像中存在对应可解释模式的可能性。然后可以使用回归模型来确定来自各个可解释模式的贡献。贡献可以用于解释缺陷的可能原因。尽管本公开描述了基于光刻后晶片的SEM图像的蚀刻后缺陷预测，但是本领域技术人员将理解，类似的预测过程也可以应用于半导体制造过程期间的其他阶段。例如，也可以应用沉积之后或执行化学机械抛光(CMP)层之后的晶片的SEM图像来预测金属触点上的缺陷。According to an embodiment of the present disclosure, first, a machine learning model, such as a neural network model, is trained based on an SEM image of a specific type of feature (e.g., a contact hole area or any other type of feature) on a post-etch wafer, and associated labeling data, the labeling data indicating individual coefficients of multiple interpretable patterns that can be used to characterize various defect categories on the wafer. During the prediction stage, the input SEM image of the post-development wafer can be decomposed into multiple pattern images associated with multiple predefined interpretable patterns (e.g., corresponding to different causes, types, or categories of defects). The trained machine learning model can be used to evaluate the decomposed pattern image. For example, the individual coefficients associated with the interpretable pattern can be used as input to the machine learning model, and the output of the machine learning model can include an evaluation result, for example, the evaluation result indicates the possibility of the corresponding interpretable pattern in the input SEM image. A regression model can then be used to determine the contribution from each interpretable pattern. The contribution can be used to explain the possible cause of the defect. Although the present disclosure describes post-etch defect prediction based on the SEM image of the post-lithography wafer, it will be understood by those skilled in the art that a similar prediction process can also be applied to other stages during the semiconductor manufacturing process. For example, SEM images of wafers after deposition or after performing a chemical mechanical polishing (CMP) layer can also be used to predict defects on metal contacts.
为了清晰起见，附图中组件的相对尺寸可能会被夸大。在附图的以下描述中，相同或相似的附图标记指代相同或相似组件或实体，并且仅描述关于个体实施例的差异。如本文中使用的，除非另有特别说明，否则术语“或”包括所有可能的组合，除非不可行。例如，如果声明组件可以包括A或B，则除非另有特别声明或不可行，否则该组件可以包括A、或B、或A和B。作为第二示例，如果声明组件可以包括A、B或C，则除非另有特别声明或不可行，否则该组件可以包括A、或B、或C、或A和B、或A和C、或B和C、或A和B和C。For the sake of clarity, the relative size of the components in the drawings may be exaggerated. In the following description of the drawings, the same or similar reference numerals refer to the same or similar components or entities, and only describe the differences about individual embodiments. As used herein, unless otherwise specifically stated, the term "or" includes all possible combinations unless not feasible. For example, if it is stated that a component can include A or B, then unless otherwise specifically stated or not feasible, the component can include A, or B, or A and B. As a second example, if it is stated that a component can include A, B or C, then unless otherwise specifically stated or not feasible, the component can include A, or B, or C, or A and B, or A and C, or B and C, or A and B and C.
本公开不限于可以用于获取图像的任何特定类型的SEM设备。图1示出了与本公开的一些实施例一致的示例性电子束检查(EBI)系统100。EBI系统100可以用于成像。如图1所示，EBI系统100包括主腔室101、装载/锁定腔室102、电子束工具104和设备前端模块(EFEM)106。电子束工具104位于主腔室101内。EFEM 106包括第一装载端口106a和第二装载端口106b。EFEM 106可以包括(多个)附加装载端口。第一装载端口106a和第二装载端口106b接收晶片前开式传送盒(FOUP)，FOUP容纳待检查的晶片(例如，半导体晶片或由其他材料制成的晶片)或样品(晶片和样品可以互换使用)。“批次”是指可以批量装载以进行处理的多个晶片。The present disclosure is not limited to any particular type of SEM equipment that can be used to acquire images. FIG. 1 shows an exemplary electron beam inspection (EBI) system 100 consistent with some embodiments of the present disclosure. The EBI system 100 can be used for imaging. As shown in FIG. 1, the EBI system 100 includes a main chamber 101, a load/lock chamber 102, an electron beam tool 104, and an equipment front end module (EFEM) 106. The electron beam tool 104 is located within the main chamber 101. The EFEM 106 includes a first loading port 106a and a second loading port 106b. The EFEM 106 may include (multiple) additional loading ports. The first loading port 106a and the second loading port 106b receive a wafer front opening transport box (FOUP), which contains wafers to be inspected (e.g., semiconductor wafers or wafers made of other materials) or samples (wafers and samples can be used interchangeably). "Batch" refers to multiple wafers that can be loaded in batches for processing.
EFEM 106中的一个或多个机械臂(未示出)可以将晶片运输到装载/锁定腔室102。装载/锁定腔室102连接到装载/锁定真空泵系统(未示出)，该系统去除装载/锁定腔室102中的气体分子以达到低于大气压的第一压力。在达到第一压力之后，一个或多个机械臂(未示出)可以将晶片从装载/锁定腔室102运输到主腔室101。主腔室101连接到主腔室真空泵系统(未示出)，该系统去除主腔室101中的气体分子以达到低于第一压力的第二压力。在达到第二压力之后，通过电子束工具104对晶片进行检查。电子束工具104可以是单束系统或多束系统。应当理解，本文中讨论的系统和方法可以应用于单束系统和多束系统两者。One or more robots (not shown) in the EFEM 106 can transport the wafer to the load/lock chamber 102. The load/lock chamber 102 is connected to a load/lock vacuum pump system (not shown), which removes gas molecules in the load/lock chamber 102 to reach a first pressure below atmospheric pressure. After reaching the first pressure, one or more robots (not shown) can transport the wafer from the load/lock chamber 102 to the main chamber 101. The main chamber 101 is connected to a main chamber vacuum pump system (not shown), which removes gas molecules in the main chamber 101 to reach a second pressure below the first pressure. After reaching the second pressure, the wafer is inspected by the electron beam tool 104. The electron beam tool 104 can be a single beam system or a multi-beam system. It should be understood that the systems and methods discussed herein can be applied to both single beam systems and multi-beam systems.
控制器109电连接到电子束工具104。控制器109可以是被配置为执行EBI系统100的各种控制的计算机。控制器109可以包括被配置为执行各种信号和图像处理功能的处理电路系统。虽然图1所示的控制器109位于包括主腔室101、装载/锁定腔室102和EFEM 106的结构之外，但可以理解，控制器109可以是该结构的一部分。在一些实施例中，控制器109可以包括一个或多个处理器，该一个或多个处理器耦合到存储支持控制器109的各种功能的指令的一个或多个存储器。The controller 109 is electrically connected to the electron beam tool 104. The controller 109 may be a computer configured to perform various controls of the EBI system 100. The controller 109 may include processing circuitry configured to perform various signal and image processing functions. Although the controller 109 shown in FIG. 1 is located outside of the structure including the main chamber 101, the load/lock chamber 102, and the EFEM 106, it is understood that the controller 109 may be part of the structure. In some embodiments, the controller 109 may include one or more processors coupled to one or more memories storing instructions that support the various functions of the controller 109.
现在参考图2，图2是与本公开的一些实施例一致的示例性电子束工具104的示意图，电子束工具104包括作为图1的EBI系统100的一部分的多束检查工具。应当理解，多束电子束工具旨在仅是说明性的，而不是限制性的。本公开也可以与单带电粒子束成像系统一起工作。如图2所示，电子束工具104(在本文中也称为装置104)包括被配置为生成主电子束的电子源201、被配置为减小库仑效应的库仑孔径板(或“枪孔径板”)271、被配置为聚焦主电子束的聚光透镜210、被配置为形成主子束(例如，主子束211、212和213)的源转换单元220、主投影系统230、机动载物台209、以及由机动载物台209支撑以保持待检查晶片208的样品保持器207。电子束工具104还可以包括次级投影系统250和电子检测设备240。主投影系统230可以包括物镜231。电子检测设备240可以包括多个检测元件241、242和243。束分离器233和偏转扫描单元232可以定位在主投影系统230内部。Now refer to Figure 2, which is a schematic diagram of an exemplary electron beam tool 104 consistent with some embodiments of the present disclosure, and the electron beam tool 104 includes a multi-beam inspection tool as part of the EBI system 100 of Figure 1. It should be understood that the multi-beam electron beam tool is intended to be illustrative only and not restrictive. The present disclosure can also work with a single charged particle beam imaging system. As shown in Figure 2, the electron beam tool 104 (also referred to as the device 104 in this article) includes an electron source 201 configured to generate a main electron beam, a Coulomb aperture plate (or "gun aperture plate") 271 configured to reduce the Coulomb effect, a condenser lens 210 configured to focus the main electron beam, a source conversion unit 220 configured to form a main beamlet (e.g., main beamlets 211, 212 and 213), a main projection system 230, a motorized stage 209, and a sample holder 207 supported by the motorized stage 209 to hold a wafer 208 to be inspected. The electron beam tool 104 may further include a secondary projection system 250 and an electron detection device 240. The main projection system 230 may include an objective lens 231. The electron detection device 240 may include a plurality of detection elements 241, 242, and 243. A beam splitter 233 and a deflection scanning unit 232 may be positioned inside the main projection system 230.
电子源201、库仑孔径板271、聚光透镜210、源转换单元220、束分离器233、偏转扫描单元232和主投影系统230可以与装置104的主光轴204对准。次级投影系统250和电子检测设备240可以与装置104的次级光轴251对准。The electron source 201, the Coulomb aperture plate 271, the condenser lens 210, the source conversion unit 220, the beam splitter 233, the deflection scanning unit 232 and the primary projection system 230 may be aligned with the primary optical axis 204 of the apparatus 104. The secondary projection system 250 and the electron detection device 240 may be aligned with the secondary optical axis 251 of the apparatus 104.
控制器109可以连接到图1的EBI系统100的各个部分，诸如源转换单元220、电子检测设备240、主投影系统230或机动载物台209。在一些实施例中，如下面进一步详细说明的，控制器109可以执行各种图像和信号处理功能。控制器109还可以生成各种控制信号以控制带电粒子束检查系统的一个或多个组件的操作。The controller 109 can be connected to various parts of the EBI system 100 of Figure 1, such as the source conversion unit 220, the electronic detection device 240, the main projection system 230, or the motorized stage 209. In some embodiments, as described in further detail below, the controller 109 can perform various image and signal processing functions. The controller 109 can also generate various control signals to control the operation of one or more components of the charged particle beam inspection system.
偏转扫描单元232在操作中被配置为将主子束211、212和213偏转为晶片208的表面的一部分中的个体扫描区域上的扫描探测斑221、222和223。响应于主子束211、212和213或探测斑221、222和223在晶片208上的入射，电子从晶片208出现并且生成三个次级电子束261、262和263。次级电子束261、262和263中的每个通常包括次级电子(电子能量≤50eV)和反向散射电子(电子能量在50eV至主子束211、212和213的着陆能量之间)。束分离器233被配置为使次级电子束261、262和263朝向次级投影系统250偏转。次级投影系统250随后将次级电子束261、262和263聚焦到电子检测设备240的检测元件241、242和243上。检测元件241、242和243被布置为检测对应的次级电子束261、262和263，并且生成对应信号，该对应信号被发送到控制器109或信号处理系统(未示出)，例如，以构造晶片208的对应扫描区域的图像。The deflection scanning unit 232 is configured in operation to deflect the main beamlets 211, 212 and 213 into scanning detection spots 221, 222 and 223 on individual scanning areas in a portion of the surface of the wafer 208. In response to the incidence of the main beamlets 211, 212 and 213 or the detection spots 221, 222 and 223 on the wafer 208, electrons emerge from the wafer 208 and generate three secondary electron beams 261, 262 and 263. Each of the secondary electron beams 261, 262 and 263 generally includes secondary electrons (electron energy ≤ 50 eV) and backscattered electrons (electron energy between 50 eV and the landing energy of the main beamlets 211, 212 and 213). The beam splitter 233 is configured to deflect the secondary electron beams 261, 262 and 263 toward the secondary projection system 250. The secondary projection system 250 then focuses the secondary electron beams 261, 262, and 263 onto the detection elements 241, 242, and 243 of the electron detection device 240. The detection elements 241, 242, and 243 are arranged to detect the corresponding secondary electron beams 261, 262, and 263 and generate corresponding signals, which are sent to the controller 109 or a signal processing system (not shown), for example, to construct an image of the corresponding scan area of the wafer 208.
在一些实施例中，检测元件241、242和243分别检测对应次级电子束261、262和263，并且向图像处理系统(例如，控制器109)生成对应强度信号输出(未示出)。在一些实施例中，每个检测元件241、242和243可以包括一个或多个像素。检测元件的强度信号输出可以是由检测元件内的所有像素生成的信号的总和。In some embodiments, the detection elements 241, 242, and 243 detect the corresponding secondary electron beams 261, 262, and 263, respectively, and generate corresponding intensity signal outputs (not shown) to an image processing system (e.g., controller 109). In some embodiments, each detection element 241, 242, and 243 may include one or more pixels. The intensity signal output of a detection element may be the sum of the signals generated by all pixels within the detection element.
在一些实施例中，控制器109可以包括图像处理系统，该图像处理系统包括图像获取器(未示出)和存储装置(未示出)。图像获取器可以包括一个或多个处理器。例如，图像获取器可以包括计算机、服务器、主机、终端、个人计算机、任何种类的移动计算设备等、或其组合。图像获取器可以通过诸如电导体、光纤线缆、便携式存储介质、IR、蓝牙、互联网、无线网络、无线无线电等介质或其组合通信耦合到装置104的电子检测设备240。在一些实施例中，图像获取器可以从电子检测设备240接收信号并且可以构造图像。图像获取器因此可以获取晶片208的图像。图像获取器还可以执行各种后处理功能，诸如生成轮廓、在所获取的图像上叠加指示符等等。图像获取器可以被配置为执行所获取的图像的亮度和对比度等的调节。在一些实施例中，存储装置可以是存储介质，诸如硬盘、闪存驱动器、云存储、随机存取存储器(RAM)、其他类型的计算机可读存储器等。存储装置可以与图像获取器耦合，并且可以用于将所扫描的原始图像数据保存为原始图像并保存后处理图像。In some embodiments, the controller 109 may include an image processing system including an image acquisition device (not shown) and a storage device (not shown). The image acquisition device may include one or more processors. For example, the image acquisition device may include a computer, a server, a host, a terminal, a personal computer, any kind of mobile computing device, etc., or a combination thereof. The image acquisition device may be coupled to the electronic detection device 240 of the device 104 through a medium such as an electrical conductor, an optical fiber cable, a portable storage medium, IR, Bluetooth, the Internet, a wireless network, a wireless radio, or a combination thereof. In some embodiments, the image acquisition device may receive a signal from the electronic detection device 240 and may construct an image. The image acquisition device may thus acquire an image of the wafer 208. The image acquisition device may also perform various post-processing functions, such as generating a contour, superimposing an indicator on the acquired image, and the like. The image acquisition device may be configured to perform adjustments to the brightness and contrast of the acquired image, etc. In some embodiments, the storage device may be a storage medium such as a hard disk, a flash drive, a cloud storage, a random access memory (RAM), other types of computer-readable memory, and the like. The storage device may be coupled to the image acquisition device and may be used to save the scanned raw image data as a raw image and save a post-processed image.
在一些实施例中，图像获取器可以基于从电子检测设备240接收的一个或多个成像信号来获取样本的一个或多个图像。成像信号可以对应于用于进行带电粒子成像的扫描操作。所获取的图像可以是包括多个成像区域的单个图像，或者可以涉及多个图像。单个图像可以存储在存储装置中。单个图像可以是可以被划分为多个区域的原始图像。每个区域可以包括一个成像区域，该一个成像区域包含晶片208的特征。所获取的图像可以包括晶片208的单个成像区域的在时间序列上多次采样的多个图像，或者可以包括晶片208的不同成像区域的多个图像。多个图像可以存储在存储装置中。在一些实施例中，控制器109可以被配置为对晶片208的相同位置的多个图像执行图像处理步骤。In some embodiments, the image acquirer may acquire one or more images of the sample based on one or more imaging signals received from the electronic detection device 240. The imaging signal may correspond to a scanning operation for performing charged particle imaging. The acquired image may be a single image including multiple imaging areas, or may involve multiple images. A single image may be stored in a storage device. A single image may be an original image that may be divided into multiple areas. Each area may include an imaging area that contains features of the wafer 208. The acquired image may include multiple images of a single imaging area of the wafer 208 sampled multiple times in a time series, or may include multiple images of different imaging areas of the wafer 208. Multiple images may be stored in a storage device. In some embodiments, the controller 109 may be configured to perform an image processing step on multiple images of the same position of the wafer 208.
在一些实施例中，控制器109可以包括用于获取检测到的次级电子的分布的测量电路系统(例如，模数转换器)。在检测时间窗口期间收集的电子分布数据与入射在晶片表面上的主子束211、212和213中的每个的对应扫描路径数据相结合可以用于重构被检查的晶片结构的图像。重构的图像可以用于揭示晶片208的内部或外部结构的各种特征，并且从而可以用于揭示可能存在于晶片中的任何缺陷。In some embodiments, the controller 109 may include measurement circuitry (e.g., an analog-to-digital converter) for acquiring the distribution of the detected secondary electrons. The electron distribution data collected during the detection time window combined with the corresponding scan path data of each of the primary beamlets 211, 212, and 213 incident on the wafer surface can be used to reconstruct an image of the inspected wafer structure. The reconstructed image can be used to reveal various features of the internal or external structure of the wafer 208, and thereby can be used to reveal any defects that may be present in the wafer.
在一些实施例中，控制器109可以控制机动载物台209在晶片208检查期间移动晶片208。在一些实施例中，控制器109可以使得机动载物台209能够以恒定速度在一个方向上连续地移动晶片208。在其他实施例中，控制器109可以使得机动载物台209能够根据扫描过程的步骤而随时间改变晶片208的移动速度。In some embodiments, the controller 109 may control the motorized stage 209 to move the wafer 208 during inspection of the wafer 208. In some embodiments, the controller 109 may enable the motorized stage 209 to continuously move the wafer 208 in one direction at a constant speed. In other embodiments, the controller 109 may enable the motorized stage 209 to change the movement speed of the wafer 208 over time according to the steps of the scanning process.
尽管如图2所示的电子束工具104使用三个主电子束，但应当理解，电子束工具可以使用单带电粒子束成像系统(“单束系统”)、或者具有两个或更多个主电子束的多带电粒子束成像系统(“多束系统”)。本公开不限制在电子束工具104中使用的主电子束的数目。Although the electron beam tool 104 shown in Figure 2 uses three main electron beams, it should be understood that the electron beam tool can use a single charged particle beam imaging system ("single beam system"), or a multi-charged particle beam imaging system ("multi-beam system") having two or more main electron beams. The present disclosure does not limit the number of main electron beams used in the electron beam tool 104.
再次参考图1，分析和预测系统199(“系统199”)可以与控制器109直接或间接通信。例如，系统199可以是被配置为无线地、远程地或通过有线连接以及其他通信方法与控制器109、EBI系统100、任何其他装置、系统或数据库进行通信的计算机。在如本公开中讨论的一些实施例中，系统199可以被配置为通过用户接口从用户接收指令，基于用户输入执行过程的模拟和数学建模，预测过程结果，并且生成描绘所预测的过程结果的图像。Referring again to FIG. 1 , the analysis and prediction system 199 (“system 199”) may communicate directly or indirectly with the controller 109. For example, the system 199 may be a computer configured to communicate with the controller 109, the EBI system 100, any other device, system, or database wirelessly, remotely, or through a wired connection, as well as other communication methods. In some embodiments as discussed in the present disclosure, the system 199 may be configured to receive instructions from a user through a user interface, perform simulations and mathematical modeling of a process based on user input, predict process outcomes, and generate images depicting the predicted process outcomes.
在一些实施例中，系统199可以包括一个或多个处理器191。处理器可以是能够操纵或处理信息的通用或特定电子设备。例如，处理器可以包括任何数目的中央处理器(或“CPU”)、图形处理单元(或“GPU”)、光处理器、可编程逻辑控制器、微控制器、微处理器、数字信号处理器、知识产权(IP)核心、可编程逻辑阵列(PLA)、可编程阵列逻辑(PAL)、通用阵列逻辑(GAL)、复杂可编程逻辑器件(CPLD)、现场可编程门阵列(FPGA)、片上系统(SoC)、专用集成电路(ASIC)、以及能够进行数据处理的任何类型的电路的任何组合。处理器也可以是包括分布在经由网络耦合的多个机器或设备之间的一个或多个处理器的虚拟处理器。In some embodiments, the system 199 may include one or more processors 191. A processor may be a general or specific electronic device capable of manipulating or processing information. For example, a processor may include any number of central processing units (or "CPUs"), graphics processing units (or "GPUs"), optical processors, programmable logic controllers, microcontrollers, microprocessors, digital signal processors, intellectual property (IP) cores, programmable logic arrays (PLAs), programmable array logic (PALs), general array logic (GALs), complex programmable logic devices (CPLDs), field programmable gate arrays (FPGAs), systems on chips (SoCs), application specific integrated circuits (ASICs), and any combination of any type of circuit capable of data processing. A processor may also be a virtual processor comprising one or more processors distributed between multiple machines or devices coupled via a network.
在一些实施例中，系统199还可以包括一个或多个存储器192。存储器可以是能够存储处理器可访问的代码和数据(例如，经由总线)的通用或特定电子设备。例如，存储器可以包括任何数目的随机存取存储器(RAM)、只读存储器(ROM)、光盘、磁盘、硬盘驱动器、固态驱动器、闪存驱动器、安全数字(SD)卡、记忆棒、紧凑型闪存(CF)卡或任何类型的存储设备的任何组合。代码可以包括操作系统(OS)和用于特定任务的一个或多个应用程序(或“app”)。存储器也可以是包括分布在经由网络耦合的多个机器或设备之间的一个或多个存储器的虚拟存储器。In some embodiments, the system 199 may also include one or more memories 192. The memory may be a general or specific electronic device capable of storing processor-accessible code and data (e.g., via a bus). For example, the memory may include any number of random access memories (RAM), read-only memories (ROM), optical disks, magnetic disks, hard drives, solid-state drives, flash drives, secure digital (SD) cards, memory sticks, compact flash (CF) cards, or any combination of any type of storage device. The code may include an operating system (OS) and one or more application programs (or "apps") for specific tasks. The memory may also be a virtual memory including one or more memories distributed between multiple machines or devices coupled via a network.
图3是与本公开的一些实施例一致的与晶片分析和缺陷预测相关联的示例晶片分析系统300的框图。在一些实施例中，晶片分析系统300包括训练模块302和分析模块304。训练模块302包括训练图像获取器310、标记数据获取器305和模型训练器320。分析模块304包括图像分析器330，图像分析器330包括分类器模型332(例如，由模型训练器320生成的)和回归模型334。图像分析器330可以处理由图像获取器340获取的图像以生成结果350。3 is a block diagram of an example wafer analysis system 300 associated with wafer analysis and defect prediction consistent with some embodiments of the present disclosure. In some embodiments, the wafer analysis system 300 includes a training module 302 and an analysis module 304. The training module 302 includes a training image acquirer 310, a labeled data acquirer 305, and a model trainer 320. The analysis module 304 includes an image analyzer 330, which includes a classifier model 332 (e.g., generated by the model trainer 320) and a regression model 334. The image analyzer 330 can process the image acquired by the image acquirer 340 to generate a result 350.
在一些实施例中，晶片分析系统300包括一个或多个处理器和存储器。例如，晶片分析系统300可以包括一个或多个计算机、服务器、主机、终端、个人计算机、任何种类的移动计算设备等、或其组合。在一些实施例中，训练模块302和分析模块304在单独的计算设备上实现。在其他实施例中，训练模块302和分析模块304可以在同一计算设备上实现。应当理解，晶片分析系统300可以包括被集成为带电粒子束检查系统(例如，图1的电子束检查系统100)的一部分的一个或多个组件或模块。晶片分析系统300还可以包括与带电粒子束检查系统分离并且通信耦合到带电粒子束检查系统的一个或多个组件或模块。在一些实施例中，晶片分析系统300可以包括一个或多个组件(例如，软件模块)，该组件可以在如本文中讨论的控制器109或系统199中实现。In some embodiments, the wafer analysis system 300 includes one or more processors and memory. For example, the wafer analysis system 300 may include one or more computers, servers, mainframes, terminals, personal computers, any type of mobile computing devices, etc., or a combination thereof. In some embodiments, the training module 302 and the analysis module 304 are implemented on separate computing devices. In other embodiments, the training module 302 and the analysis module 304 may be implemented on the same computing device. It should be understood that the wafer analysis system 300 may include one or more components or modules that are integrated as a part of a charged particle beam inspection system (e.g., the electron beam inspection system 100 of FIG. 1). The wafer analysis system 300 may also include one or more components or modules that are separated from the charged particle beam inspection system and are communicatively coupled to the charged particle beam inspection system. In some embodiments, the wafer analysis system 300 may include one or more components (e.g., software modules) that may be implemented in a controller 109 or system 199 as discussed herein.
在图3所示的一些实施例中，训练模块302包括训练图像获取器310。训练图像获取器310可以被配置为获取训练图像(或训练数据)，诸如图4A所示的晶片上的区域的多个SEM图像。所获取的训练图像可以被馈送到用于训练分类器模型332的模型训练器320。在一些实施例中，训练图像获取器310可以从数据库、控制器109、系统199或电子束工具104获取训练图像。例如，训练图像获取器310可以是如本文中讨论的控制器109的图像获取器。在一些实施例中，相应训练图像可以对应于晶片上的区域，包括任何类型的1D和2D特征，诸如单个接触孔、多个接触孔、一个或多个线、管芯或整个晶片。本公开不限于晶片上的任何特定类型的特征。在一些实施例中，晶片的区域可以基于分析的目的来选择，诸如确定接触孔是否失败、线区域是否有缺陷、或者芯片的一部分是否失败、以及(多个)什么类型的缺陷可能导致(多个)这样的失败。在一些实施例中，训练图像可以包括在不同阶段进行处理的样品的SEM图像，诸如在抗蚀剂显影之后或在蚀刻之后。在一些实施例中，训练图像的训练数据(诸如与灰度值相对应的像素值)可以从训练图像中被提取以用于训练过程。In some embodiments shown in FIG. 3 , the training module 302 includes a training image acquirer 310. The training image acquirer 310 may be configured to acquire training images (or training data), such as a plurality of SEM images of an area on a wafer as shown in FIG. 4A . The acquired training images may be fed to a model trainer 320 for training a classifier model 332. In some embodiments, the training image acquirer 310 may acquire training images from a database, a controller 109, a system 199, or an electron beam tool 104. For example, the training image acquirer 310 may be an image acquirer of a controller 109 as discussed herein. In some embodiments, the corresponding training images may correspond to an area on a wafer, including any type of 1D and 2D features, such as a single contact hole, a plurality of contact holes, one or more lines, a die, or an entire wafer. The present disclosure is not limited to any particular type of feature on a wafer. In some embodiments, an area of a wafer may be selected based on the purpose of the analysis, such as determining whether a contact hole fails, whether a line area is defective, or whether a portion of a chip fails, and what type of defect may cause such failure. In some embodiments, the training images may include SEM images of samples processed at different stages, such as after resist development or after etching. In some embodiments, training data (such as pixel values corresponding to grayscale values) of the training images may be extracted from the training images for use in the training process.
在一些实施例中，训练模块302可以包括标记数据获取器305，标记数据获取器305被配置为获取与由训练图像获取器310获取的训练图像相关联的标记数据。在一些实施例中，每个训练图像可以用与多个类别相关联的相应系数来标记，诸如多个可解释模式(或感兴趣模式)。每个图像可以被标记为单个类别或多个类别的组合。在一些实施例中，可解释模式可以对应于晶片上的特征的(例如缺陷(或失败原因))不同类别。特征或缺陷的一些示例可以包括小临界尺寸(CD)、沿着特定方向的偏移、椭圆度、模糊边缘、印刷接触孔、缺失接触孔和桥接接触孔等。在一些实施例中，标记数据可以由专家基于其先验知识来确定。在一些实施例中，标记数据可以使用自动程序来计算，诸如主成分分析(PCA)或奇异值分解(SVD)方法或本领域公知的任何其他合适的过程。In some embodiments, the training module 302 may include a labeled data acquirer 305, which is configured to acquire labeled data associated with the training images acquired by the training image acquirer 310. In some embodiments, each training image can be labeled with corresponding coefficients associated with multiple categories, such as multiple interpretable patterns (or patterns of interest). Each image can be labeled as a single category or a combination of multiple categories. In some embodiments, the interpretable patterns can correspond to different categories of features (e.g., defects (or failure causes)) on the wafer. Some examples of features or defects may include small critical dimensions (CD), offsets along a specific direction, ellipticity, fuzzy edges, printed contact holes, missing contact holes, and bridged contact holes. In some embodiments, the labeled data can be determined by an expert based on his prior knowledge. In some embodiments, the labeled data can be calculated using an automatic program, such as a principal component analysis (PCA) or a singular value decomposition (SVD) method or any other suitable process known in the art.
在一些实施例中，训练模块302的模型训练器320可以基于训练图像和对应标记数据来训练分类器模型332。在一些实施例中，分类器模型332可以是逻辑回归模型、诸如支持向量机等机器学习模型、或深度神经网络(诸如卷积神经网络)、或适合于预测分类的任何其他模型。在一些实施例中，分类器模型332可以用于预测与一个或多个可解释模式相关联的一个或多个类别中的每个类别是否存在于对应SEM图像中。在一些实施例中，由模型训练器320获取的训练结果包括分类器模型332的优化权重，如下面更详细描述的。In some embodiments, the model trainer 320 of the training module 302 can train a classifier model 332 based on the training images and the corresponding labeled data. In some embodiments, the classifier model 332 can be a logistic regression model, a machine learning model such as a support vector machine, or a deep neural network (such as a convolutional neural network), or any other model suitable for predicting classification. In some embodiments, the classifier model 332 can be used to predict whether each of one or more categories associated with one or more interpretable patterns is present in the corresponding SEM image. In some embodiments, the training results obtained by the model trainer 320 include optimized weights of the classifier model 332, as described in more detail below.
图4A是根据本公开的一些实施例的由训练图像获取器310获取的一组训练图像的示例。例如，如图4A所示，训练图像包括与不同特征或缺陷相关联的接触孔的SEM图像。训练图像可以对应于晶片上的相同或不同位置处的接触孔。4A is an example of a set of training images acquired by the training image acquirer 310 according to some embodiments of the present disclosure. For example, as shown in FIG4A , the training images include SEM images of contact holes associated with different features or defects. The training images may correspond to contact holes at the same or different locations on the wafer.
图4B是根据本公开的一些实施例的使用诸如PCA等自动方法处理的与图4A的该组训练图像相对应的一组标记图像的示例。标记图像对应于不同可解释模式。在一些实施例中，PCA被应用于该组训练图像，以确定与主分量相关联(例如，与可解释模式相对应)的各个系数。在一些实施例中，如图4B所示，首先确定该组训练图像的均值。然后，计算各个训练图像与均值的偏差，以分别确定主分量的系数。所获取的与主分量相关联的系数可以将相应特征或缺陷表征为与对应训练图像相关联的标记数据。Figure 4B is an example of a set of labeled images corresponding to the set of training images of Figure 4A processed using an automatic method such as PCA according to some embodiments of the present disclosure. The labeled images correspond to different interpretable patterns. In some embodiments, PCA is applied to the set of training images to determine the individual coefficients associated with the principal components (e.g., corresponding to the interpretable patterns). In some embodiments, as shown in Figure 4B, the mean of the set of training images is first determined. Then, the deviations of the individual training images from the mean are calculated to determine the coefficients of the principal components, respectively. The coefficients associated with the principal components obtained can characterize the corresponding features or defects as labeled data associated with the corresponding training images.
图5是与本公开的一些实施例一致的表示用于训练分类器模型(例如，分类器模型332)的示例方法500的过程流程图。在一些实施例中，一个或多个步骤由图3中的系统300的一个或多个组件(例如，训练模块302)、图1中的控制器109或系统199来执行。5 is a process flow diagram representing an example method 500 for training a classifier model (e.g., classifier model 332) consistent with some embodiments of the present disclosure. In some embodiments, one or more steps are performed by one or more components (e.g., training module 302) of system 300 in FIG. 3, controller 109 in FIG. 1, or system 199.
如图5所示，在步骤510中，可以获取多个训练图像的图像数据。在一些实施例中，训练图像可以是电子显微镜图像，诸如如图4A所示的SEM图像。训练图像可以通过图3中的训练图像获取器310或者图1中的控制器109或系统199来获取。训练图像可以从图1中的控制器109、系统199、或电子束工具104、或任何其他合适的数据库来获取。在一些实施例中，多个训练图像的图像数据可以包括每个训练图像的像素值。As shown in FIG5 , in step 510, image data of a plurality of training images may be acquired. In some embodiments, the training images may be electron microscope images, such as SEM images as shown in FIG4A . The training images may be acquired by the training image acquirer 310 in FIG3 or the controller 109 or system 199 in FIG1 . The training images may be acquired from the controller 109, the system 199, or the electron beam tool 104 in FIG1 , or any other suitable database. In some embodiments, the image data of the plurality of training images may include pixel values for each training image.
在步骤520中，可以获取与训练图像相关联的标记数据。在一些实施例中，标记数据可以由图3中的标记数据获取器305或者图1中的控制器109或系统199获取。在一些实施例中，标记数据获取器305可以使用诸如PCA/SVD等自动过程来处理在步骤510中获取的训练图像(例如，如图4B所示)。在一些实施例中，训练图像可以由专家基于先验知识进行分析和标记。在一些其他实施例中，训练图像可以通过被实现为软件程序的自动化过程来标记。在一些实施例中，标记数据可以包括与相应可解释模式相关联的系数，这些可解释模式可以表征与每个训练图像相关联的特征或缺陷，诸如小临界尺寸(CD)、沿着特定方向的偏移、椭圆度、模糊边缘、印刷接触孔、缺失接触孔、桥接接触孔等。In step 520, labeled data associated with the training image may be acquired. In some embodiments, the labeled data may be acquired by the labeled data acquirer 305 in FIG. 3 or the controller 109 or system 199 in FIG. 1 . In some embodiments, the labeled data acquirer 305 may process the training image acquired in step 510 (e.g., as shown in FIG. 4B ) using an automated process such as PCA/SVD. In some embodiments, the training image may be analyzed and labeled by an expert based on prior knowledge. In some other embodiments, the training image may be labeled by an automated process implemented as a software program. In some embodiments, the labeled data may include coefficients associated with corresponding interpretable patterns that may characterize features or defects associated with each training image, such as small critical dimensions (CD), offsets along a particular direction, ellipticity, fuzzy edges, printed contact holes, missing contact holes, bridged contact holes, and the like.
在步骤530中，可以基于训练图像和相关联的标记数据来训练(例如，经由图3中的模型训练器320)分类器模型(例如，图3中的分类器模型332)。在一些实施例中，分类器模型可以包括适合于分类的任何模型，诸如逻辑回归模型、支持向量机或深度神经网络。任何合适的训练过程都可以用于优化分类器模型的权重。In step 530, a classifier model (e.g., classifier model 332 in FIG. 3 ) may be trained (e.g., via model trainer 320 in FIG. 3 ) based on the training images and associated labeled data. In some embodiments, the classifier model may include any model suitable for classification, such as a logistic regression model, a support vector machine, or a deep neural network. Any suitable training process may be used to optimize the weights of the classifier model.
再次参考图3，在一些实施例中，晶片分析系统300的分析模块304包括图像获取器340，图像获取器340被配置为获取晶片上的区域的输入显微镜图像(例如，输入SEM图像)，以用于缺陷分析(例如，预测与晶片相关联的失败原因)。在一些实施例中，图像获取器340可以从如图1所示的电子束工具104、控制器109或系统199获取输入图像。例如，图像获取器340可以是如本文中讨论的控制器109的图像获取器。在一些实施例中，输入图像可以对应于晶片上需要缺陷预测的区域，诸如单个接触孔、多个接触孔、一个或多个线、管芯或整个晶片。在一些实施例中，输入图像可以是在半导体处理期间的不同阶段拍摄的晶片的SEM图像，诸如在光刻中显影之后或者在蚀刻之后。例如，分类器模型332可以使用蚀刻后处理的晶片的训练图像和相关联的标记数据来训练，如本文中讨论的，而分类器模型332可以用于预测在光刻中显影后的晶片的输入图像的缺陷类别，以用于早期缺陷检测/预测。例如，与从输入图像的分解中获取的各个可解释模式相关联的系数可以作为输入被馈送到分类器模型332中，以获取输出作为与各个可解释模型相关联的评估结果(例如，图6中的评估结果640指示是否存在某个可解释模式)。此外，用于训练分类器模型332的训练图像(例如，蚀刻后晶片的SEM图像)可以对应于晶片上与用于预测的输入SEM图像中的区域不同的区域。例如，分类器模型332可以使用捕获一个或多个晶片上的不同位置上的接触孔的训练图像来训练。在训练之后，这样的分类器模型332可以用于预测与特定接触孔相关联的缺陷类别，该特定接触孔位于与训练图像中包括的任何接触孔相同或不同的位置处。Referring again to FIG. 3 , in some embodiments, the analysis module 304 of the wafer analysis system 300 includes an image acquirer 340 configured to acquire an input microscope image (e.g., an input SEM image) of an area on a wafer for defect analysis (e.g., predicting a failure cause associated with the wafer). In some embodiments, the image acquirer 340 may acquire the input image from the electron beam tool 104, the controller 109, or the system 199 as shown in FIG. 1 . For example, the image acquirer 340 may be an image acquirer of the controller 109 as discussed herein. In some embodiments, the input image may correspond to an area on the wafer for which defect prediction is desired, such as a single contact hole, a plurality of contact holes, one or more lines, a die, or an entire wafer. In some embodiments, the input image may be an SEM image of a wafer taken at different stages during semiconductor processing, such as after development in lithography or after etching. For example, the classifier model 332 can be trained using training images of wafers processed after etching and associated labeled data, as discussed herein, and the classifier model 332 can be used to predict defect categories of input images of wafers developed in lithography for early defect detection/prediction. For example, coefficients associated with each interpretable pattern obtained from the decomposition of the input image can be fed into the classifier model 332 as input to obtain output as evaluation results associated with each interpretable model (e.g., the evaluation result 640 in FIG. 6 indicates whether a certain interpretable pattern exists). In addition, the training images used to train the classifier model 332 (e.g., SEM images of wafers after etching) can correspond to different areas on the wafer than the areas in the input SEM images used for prediction. For example, the classifier model 332 can be trained using training images that capture contact holes at different locations on one or more wafers. After training, such a classifier model 332 can be used to predict defect categories associated with a specific contact hole that is located at the same or different location as any contact hole included in the training image.
在一些实施例中，分析模块304包括被配置为分析输入图像的图像分析器330。在一些实施例中，分析模块304可以将输入图像分解为与不同缺陷类别相对应的多个可解释模式图像，并且获取与各个可解释模式相关联的系数以用于表征输入图像。输入图像可以使用PCA或本领域公知的任何其他合适的方法来分解。In some embodiments, the analysis module 304 includes an image analyzer 330 configured to analyze an input image. In some embodiments, the analysis module 304 may decompose the input image into a plurality of interpretable pattern images corresponding to different defect categories, and obtain coefficients associated with each interpretable pattern for characterizing the input image. The input image may be decomposed using PCA or any other suitable method known in the art.
在一些实施例中，图像分析器330包括由训练模块302使用图5的方法500生成的分类器模型332。分类器模型332可以包括神经网络模型(例如，如图6所示)或逻辑分类器模型(例如，如图10所示)。在分类器模型332包括神经网络模型的实施例中，与各个的分解后的可解释模式相关联的系数可以由神经网络模型评估以获取评估结果，该评估结果指示与某个可解释模式相对应的相应缺陷类别是否可能存在于输入SEM图像中。In some embodiments, the image analyzer 330 includes a classifier model 332 generated by the training module 302 using the method 500 of FIG5. The classifier model 332 may include a neural network model (e.g., as shown in FIG6) or a logical classifier model (e.g., as shown in FIG10). In embodiments where the classifier model 332 includes a neural network model, coefficients associated with each of the decomposed interpretable patterns may be evaluated by the neural network model to obtain an evaluation result indicating whether a corresponding defect category corresponding to a certain interpretable pattern is likely to exist in the input SEM image.
在一些实施例中，图像分析器330还包括回归模型334，回归模型334被配置为计算相应的分解后的可解释模式对输入SEM图像的贡献。在一些实施例中，回归模型334包括多项式模型，诸如线性模型、二次模型、或具有不同阶数组合的多项式模型。例如，基于神经网络模型和通过将分类器模型332应用于与输入图像相关联的分解后的模式的系数而获取的评估结果，线性模型可以对神经网络模型与评估结果之间的线性关系进行近似。这样的量化的线性关系可以用于确定各个可解释模式对输入SEM图像的贡献。因此，结果350可以标识与对输入图像有显著贡献的一个或多个可解释模式相对应的一个或多个缺陷类别。In some embodiments, the image analyzer 330 further includes a regression model 334 configured to calculate the contribution of the corresponding decomposed interpretable pattern to the input SEM image. In some embodiments, the regression model 334 includes a polynomial model, such as a linear model, a quadratic model, or a polynomial model with different order combinations. For example, based on the neural network model and the evaluation results obtained by applying the classifier model 332 to the coefficients of the decomposed pattern associated with the input image, the linear model can approximate the linear relationship between the neural network model and the evaluation results. Such a quantified linear relationship can be used to determine the contribution of each interpretable pattern to the input SEM image. Therefore, the result 350 can identify one or more defect categories corresponding to one or more interpretable patterns that significantly contribute to the input image.
图6示出了与本公开的一些实施例一致的表示用于预测晶片上的缺陷类别(例如，由可解释模式表示)的示例过程600的流程图。在一些实施例中，过程600的一个或多个步骤由图3中的系统300的一个或多个组件(例如，分析模块304)或者图1中的控制器109或系统199来执行。如本公开中公开的，过程600不仅可以基于SEM图像预测缺陷是否存在，还可以预测缺陷类别(例如，在本文中也可互换地称为“缺陷类型”或“缺陷原因”或“可解释模式”)。因此，过程600有利于缺陷的早期检测和校正，因为可以在早期阶段预测缺陷类别，诸如在光刻中的显影之后。FIG6 shows a flow chart representing an example process 600 for predicting defect classes (e.g., represented by interpretable patterns) on a wafer consistent with some embodiments of the present disclosure. In some embodiments, one or more steps of process 600 are performed by one or more components (e.g., analysis module 304) of system 300 in FIG3 or controller 109 or system 199 in FIG1. As disclosed in the present disclosure, process 600 can predict not only the presence or absence of defects based on SEM images, but also the defect class (e.g., also interchangeably referred to herein as "defect type" or "defect cause" or "interpretable pattern"). Therefore, process 600 facilitates early detection and correction of defects because the defect class can be predicted at an early stage, such as after development in lithography.
在一些实施例中，图像获取器340从控制器109、系统199或电子束工具104获取输入SEM图像610。如图6所示，在一些实施例中，输入SEM图像610反映在光刻中的显影处理之后的接触孔的区域。In some embodiments, the image acquirer 340 acquires the input SEM image 610 from the controller 109, the system 199, or the electron beam tool 104. As shown in FIG6, in some embodiments, the input SEM image 610 reflects the area of the contact hole after the development process in photolithography.
在一些实施例中，图像分析器330将输入SEM图像610分解为多个模式图像620，每个模式图像620对应于特定可解释模式，例如，每个可解释模式与缺陷类别相关。图像分析器330获取分别与可解释模式相关联的系数(例如，C1、C2、C3、C4、C5……)。模式图像620可以通过使用PCA(例如，如图4A-图4B所示)或任何其他合适的方法分解输入SEM图像610来获取。在一些实施例中，多个模式620可以对应于不同缺陷类别，诸如小CD、偏移、椭圆度、模糊边缘、印刷接触孔、缺失接触孔或桥接接触孔。例如，具有小CD的缺陷对应于被改变的接触孔的尺寸或半径。在另一示例中，偏移缺陷对应于接触孔从其预期位置(例如，从设计)向上或向下移动的位置。因此，与原始输入SEM图像相比，特定的分解后的模式可以表现出偏差(例如，在尺寸、位置、形状、对比度、强度或任何其他因素方面)。In some embodiments, the image analyzer 330 decomposes the input SEM image 610 into a plurality of pattern images 620, each pattern image 620 corresponding to a specific interpretable pattern, for example, each interpretable pattern is associated with a defect category. The image analyzer 330 obtains coefficients (e.g., C1, C2, C3, C4, C5, ...) associated with the interpretable patterns, respectively. The pattern images 620 can be obtained by decomposing the input SEM image 610 using PCA (e.g., as shown in FIG. 4A-FIG. 4B) or any other suitable method. In some embodiments, the plurality of patterns 620 can correspond to different defect categories, such as small CD, offset, ellipticity, fuzzy edges, printed contact holes, missing contact holes, or bridged contact holes. For example, a defect with a small CD corresponds to a size or radius of a contact hole that is changed. In another example, an offset defect corresponds to a position where a contact hole is moved up or down from its intended position (e.g., from the design). Therefore, a particular decomposed pattern can exhibit a deviation (e.g., in size, position, shape, contrast, intensity, or any other factor) compared to the original input SEM image.
在一些实施例中，图像分析器330进一步基于使用一个或多个模型的分解后的模式图像620来确定哪个(哪些)可解释模式以及在多大程度上对输入SEM图像610有贡献。例如，图像分析器330可以计算来自各个可解释模式的量化贡献，以便确定各个可解释模式是否对输入SEM图像有显著贡献。因此，可以确定输入SEM图像是否具有对应缺陷。In some embodiments, the image analyzer 330 further determines which (and to what extent) the interpretable pattern(s) contribute to the input SEM image 610 based on the decomposed pattern image 620 using one or more models. For example, the image analyzer 330 can calculate the quantitative contribution from each interpretable pattern to determine whether each interpretable pattern contributes significantly to the input SEM image. Therefore, it can be determined whether the input SEM image has a corresponding defect.
在如图6所示的一些实施例中，训练模块302可以基于训练SEM图像来训练分类器模型630(例如，类似于参考图3-图5讨论的分类器模型332)。图像分析器330然后可以使用分类器模型630(例如，图6所示的预训练神经网络模型)来计算相关联的评估结果640。在一些实施例中，分类器模型620的输入包括分别与可解释模式相关联的系数(例如，C1、C2、C3、C4、C5……)。例如，分类器模型(或分类器模型332)可以由下式表示：In some embodiments as shown in FIG. 6 , the training module 302 can train a classifier model 630 (e.g., similar to the classifier model 332 discussed with reference to FIGS. 3-5 ) based on the training SEM images. The image analyzer 330 can then use the classifier model 630 (e.g., the pre-trained neural network model shown in FIG. 6 ) to calculate the associated evaluation results 640. In some embodiments, the input of the classifier model 620 includes coefficients (e.g., C1, C2, C3, C4, C5, etc.) respectively associated with the interpretable patterns. For example, the classifier model (or classifier model 332) can be represented by the following formula:
y＝f(x1，x2，…，xn) (1)y=f(x1 , x2 ,..., xn ) (1)
其中f表示非线性模型(例如，分类器模型630)的函数，y是函数的输出(例如，评估结果)，xi(i＝1、2、……、n)是函数的输入。在一些实施例中，输入xi(i＝1、2、……、n)可以包括与使用PCA从分解过程获取的相应可解释模式相关联的系数。本公开不限于函数或非线性模型的任何特定形式。Wherein f represents a function of a nonlinear model (e.g., classifier model 630), y is an output of the function (e.g., an evaluation result), andxi (i=1, 2, ..., n) is an input of the function. In some embodiments, the inputxi (i=1, 2, ..., n) may include coefficients associated with corresponding interpretable patterns obtained from a decomposition process using PCA. The present disclosure is not limited to any particular form of function or nonlinear model.
在一些实施例中，对于每个分解后的模式图像，对应评估结果y指示输入SEM图像中是否存在对应缺陷类别。例如，如图6所示，当评估结果y是正值(例如，“预测打印”)时，可以确定输入SEM图像610中不存在与这样的模式相对应的缺陷。另一方面，当评估结果y是负值(例如，“预测缺失”)时，可以确定输入SEM图像610中存在与这样的模式相对应的缺陷。然而，这种讨论仅仅是示例性的。在不脱离本公开的范围的情况下，该模型可以被训练以预测任何其他分类或任何数目的类别。In some embodiments, for each decomposed pattern image, the corresponding evaluation result y indicates whether the corresponding defect category exists in the input SEM image. For example, as shown in FIG6 , when the evaluation result y is a positive value (e.g., “predicted print”), it can be determined that there are no defects corresponding to such a pattern in the input SEM image 610. On the other hand, when the evaluation result y is a negative value (e.g., “predicted missing”), it can be determined that there are defects corresponding to such a pattern in the input SEM image 610. However, this discussion is merely exemplary. Without departing from the scope of the present disclosure, the model can be trained to predict any other classification or any number of categories.
在获取评估结果640之后，可以使用多项式回归模型(诸如线性模型650)来分类器模型630(例如，由等式(1)中的非线性函数f表示)与评估结果640(例如，由等式(1)中的评估结果y表示)之间的线性关系进行近似。可以从近似线性关系中获取参数，诸如与相应可解释模式相关联的系数，以确定相应可解释模型的贡献。例如，线性模型650可以由下式表示：After obtaining the evaluation result 640, a polynomial regression model (such as a linear model 650) can be used to approximate the linear relationship between the classifier model 630 (e.g., represented by the nonlinear function f in equation (1)) and the evaluation result 640 (e.g., represented by the evaluation result y in equation (1)). Parameters, such as coefficients associated with the corresponding interpretable mode, can be obtained from the approximate linear relationship to determine the contribution of the corresponding interpretable model. For example, the linear model 650 can be represented by the following formula:
其中y是输出(与等式(1)中相同)，f0表示常数(constant)，Δxi(i＝1、2、……、n)是与可解释模式的各个系数相关的输入值，wi是与从线性近似获取的相应可解释模式相关联的权重(例如，wi独立于输入值Δxi)，bi是与相应可解释模式相关联的值(例如，取决于输入值Δxi)。在一些实施例中，线性近似可以使用任何合适的线性模式，诸如泰勒展开。因此，可以计算相应可解释模式的量化贡献(例如，wi)，并且可以相应地确定相应模式是否对输入SEM图像有足够的贡献。例如，如果某个权重wi接近0，则可以确定对应模式对输入SEM图像610没有贡献。如果某个权重wi是正的，则可以确定对应模式对非失败的肯定判决有贡献，例如，在输入SEM图像610中不存在对应缺陷类型。如果某个权重wi是负的，则可以确定对应模式对失败有贡献，例如，在输入SEM图像610中存在对应缺陷类型。Where y is the output (same as in equation (1)), f0 represents a constant,Δxi (i=1, 2, ..., n) are input values associated with the respective coefficients of the interpretable pattern,wi is a weight associated with the corresponding interpretable pattern obtained from the linear approximation (e.g.,wi is independent of the input value Δxi) , andbi is a value associated with the corresponding interpretable pattern (e.g., dependent on the input value Δxi) . In some embodiments, the linear approximation may use any suitable linear pattern, such as a Taylor expansion. Thus, the quantitative contribution of the corresponding interpretable pattern (e.g.,wi ) may be calculated, and it may be determined accordingly whether the corresponding pattern has a sufficient contribution to the input SEM image. For example, if a certain weightwi is close to 0, it may be determined that the corresponding pattern has no contribution to the input SEM image 610. If a certain weightwi is positive, it may be determined that the corresponding pattern has contributed to a positive decision of non-failure, for example, the corresponding defect type is not present in the input SEM image 610. If a certain weightwi is negative, then it may be determined that the corresponding pattern contributes to the failure, for example, the corresponding defect type exists in the input SEM image 610.
图7示出了与本公开的一些实施例一致的可视化了将非线性分类器模型(例如，分类器模型630)和线性化模型(例如，线性模型650)应用于被分解成两个可解释模式的多个SEM图像的图700。应当理解，图7的二维(2D)可视化旨在仅是说明性的，而不是限制性的。本领域普通技术人员将理解，类似的概念可以应用于多维空间，以用于预测与和SEM图像相关联的多种类型的缺陷相对应的两个以上的可解释模式。FIG7 shows a graph 700 that visualizes the application of a nonlinear classifier model (e.g., classifier model 630) and a linearized model (e.g., linear model 650) to multiple SEM images decomposed into two interpretable modes consistent with some embodiments of the present disclosure. It should be understood that the two-dimensional (2D) visualization of FIG7 is intended to be illustrative only and not limiting. One of ordinary skill in the art will appreciate that similar concepts can be applied to multi-dimensional spaces for predicting more than two interpretable modes corresponding to multiple types of defects associated with SEM images.
在一些实施例中，获取与多个晶片上的各个区域相对应的多个SEM图像。每个SEM图像可以通过标记为PC1和PC2的两个可解释模式进行分解和表征。图7中的每个点对应于分解成两个可解释模式的SEM图像。例如，每个点的坐标对应于与从PCA获取的两个分解后的可解释模式相关联的系数。如图7所示，实线表示分类器模型630的非线性决策边界。In some embodiments, multiple SEM images corresponding to various regions on multiple wafers are obtained. Each SEM image can be decomposed and characterized by two interpretable modes labeled PC1 and PC2. Each point in FIG. 7 corresponds to a SEM image decomposed into two interpretable modes. For example, the coordinates of each point correspond to the coefficients associated with the two decomposed interpretable modes obtained from PCA. As shown in FIG. 7, the solid line represents the nonlinear decision boundary of the classifier model 630.
在一些实施例中，图700中与输入图像610相对应的点P可以在图700中标识，并且感兴趣区域内的点P的相邻点可以用于图6中讨论的线性化过程。例如，考虑图700中的点P及其相邻点，可以建立线性模型650以线性化非线性分类器模型630，以表示分类器模型630与其对相应可解释模式的评估结果之间的线性化关系。如图7所示，虚线示出了点P及在感兴趣区域(例如，接触孔)内的其相邻点的非线性决策边界的线性化结果(例如，局部线性近似)。In some embodiments, a point P in the graph 700 corresponding to the input image 610 can be identified in the graph 700, and the neighboring points of the point P in the region of interest can be used for the linearization process discussed in FIG6. For example, considering the point P and its neighboring points in the graph 700, a linear model 650 can be established to linearize the nonlinear classifier model 630 to represent the linearized relationship between the classifier model 630 and its evaluation results for the corresponding interpretable pattern. As shown in FIG7, the dotted line shows the linearization result (e.g., local linear approximation) of the nonlinear decision boundary of the point P and its neighboring points in the region of interest (e.g., contact hole).
图8示出了根据本公开的一些实施例的通过对输入SEM图像执行过程600而获取的可视化预测结果的示例。在一些实施例中，图8中的条形图可视化了可解释模式对决策结果的贡献。例如，每个条可以被绘制为向下指向(例如，来自上述等式(2)的负wi值)，以指示存在对应类型的缺陷，或者向上指向(例如，来自上述等式(2)的正wi值)，以指示不存在对应类型的缺陷。此外，每个条的大小由根据上述等式(2)而确定的bi值来确定，以说明对应类型的缺陷对输入SEM图像的贡献的重要性。因此，哪种类别的缺陷以及贡献有多大可以由每个条的方向和大小来确定。FIG8 shows an example of a visualized prediction result obtained by performing process 600 on an input SEM image according to some embodiments of the present disclosure. In some embodiments, the bar chart in FIG8 visualizes the contribution of the interpretable pattern to the decision result. For example, each bar can be drawn to point downward (e.g., a negativewi value from equation (2) above) to indicate the presence of a defect of the corresponding type, or to point upward (e.g., a positivewi value from equation (2) above) to indicate the absence of a defect of the corresponding type. In addition, the size of each bar is determined by thebi value determined according to equation (2) above to illustrate the importance of the contribution of the corresponding type of defect to the input SEM image. Therefore, which type of defect and how much contribution can be determined by the direction and size of each bar.
例如，如图8所示，显影后晶片上的区域的第一输入SEM图像(“CH 1”)可以被确定为第二条具有向负方向延伸并且具有最大大小。第二条(模式2)对应于CD类型的分量。因此，可以预测，蚀刻后(或在半导体处理期间的其他阶段)晶片上的该区域可能具有小CD的缺陷(或主要缺陷)。For example, as shown in FIG8 , the first input SEM image (“CH 1”) of an area on a wafer after development can be determined to have a second stripe extending in a negative direction and having a maximum size. The second stripe (mode 2) corresponds to a component of the CD type. Therefore, it can be predicted that this area on the wafer after etching (or at other stages during semiconductor processing) is likely to have a small CD defect (or major defect).
如图8所示，显影后晶片上的区域的第二输入SEM图像(“CH 2”)可以被确定为第三条具有向负方向延伸并且具有最大大小。第三条(模式3)对应于y偏移类型的分量。因此，可以预测，蚀刻后晶片上的该区域可能具有沿着y方向强烈偏移的缺陷(或主要缺陷)。As shown in Figure 8, the second input SEM image of the area on the wafer after development ("CH 2") can be determined as the third bar having a maximum magnitude extending in the negative direction. The third bar (mode 3) corresponds to a component of the y-shift type. Therefore, it can be predicted that this area on the wafer after etching may have a defect (or a major defect) that is strongly shifted along the y-direction.
在一些实施例中，过程600还可以预测导致晶片上区域失败的多个缺陷类型。例如，第三输入SEM图像、第四输入SEM图像和第五输入SEM图像(“CH 3”、“CH 4”和“CH 5”)可以被确定为分别主要具有以下缺陷：椭圆度和一些y偏移、导致非垂直壁的模糊边缘、以及小CD和强y偏移。应当理解，根据图像分解的算法，可以使用多种模式来指示另一分类，例如缺陷相关分类。因此，本公开可以用于基于分类与多个模式之间的特定相关性通过任何其他合适的标准对图像进行分类。In some embodiments, process 600 may also predict multiple defect types that cause a region on a wafer to fail. For example, the third input SEM image, the fourth input SEM image, and the fifth input SEM image ("CH 3", "CH 4", and "CH 5") may be determined to have primarily the following defects, respectively: ellipticity and some y-shift, fuzzy edges resulting in non-vertical walls, and small CD and strong y-shift. It should be understood that, depending on the algorithm of the image decomposition, multiple patterns may be used to indicate another classification, such as a defect-related classification. Therefore, the present disclosure may be used to classify images by any other suitable criteria based on a specific correlation between a classification and multiple patterns.
图9A-图9E进一步示出了根据本公开的一些实施例的通过对各种输入SEM图像执行过程600而获取的可视化预测结果(例如，如条形图)的示例。例如，如图9A所示，图9A中的每个SEM图像的条形图示出了第二条(模式2)向负方向延伸并且具有最大大小。因此，图9A中的每个晶片在蚀刻之后可能具有小CD的缺陷(或主要缺陷)。9A-9E further illustrate examples of visual prediction results (e.g., as bar graphs) obtained by performing process 600 on various input SEM images according to some embodiments of the present disclosure. For example, as shown in FIG9A , the bar graph of each SEM image in FIG9A shows that the second bar (mode 2) extends in the negative direction and has the largest size. Therefore, each wafer in FIG9A may have a small CD defect (or major defect) after etching.
类似地，从图9B中的条形图中可以看出，图9B的每个晶片在蚀刻后都可能具有垂直偏移的缺陷(或主要缺陷)(对应于模式3中的长的负的第三条)。此外，基于条的大小，可以解释，图9B中的第一晶片和第五晶片可能具有更主要的小CD缺陷，如相应条形图中最长的负的第二条所证明的。Similarly, it can be seen from the bar graph in FIG9B that each wafer of FIG9B may have a vertically offset defect (or dominant defect) after etching (corresponding to the long negative third bar in Mode 3). In addition, based on the size of the bars, it can be interpreted that the first and fifth wafers in FIG9B may have more dominant small CD defects, as evidenced by the longest negative second bar in the corresponding bar graphs.
图9C中的条形图示出，图9C的每个晶片在蚀刻后可能具有至少一个椭圆度的缺陷(例如，主要缺陷或多个缺陷中的缺陷)，如长的负的第四条所证明的。基于相应SEM图像中的条的大小，可以理解，椭圆度缺陷可能是第四晶片中的主要缺陷。此外，第一晶片、第二晶片、第三晶片和第五晶片可能具有更主要的小CD缺陷，如相应条形图中最长的负的第二条所证明的。The bar graph in FIG9C shows that each wafer of FIG9C may have at least one ellipticity defect (e.g., a major defect or a defect among multiple defects) after etching, as evidenced by the long negative fourth bar. Based on the size of the bars in the corresponding SEM images, it can be understood that the ellipticity defect may be the major defect in the fourth wafer. In addition, the first wafer, the second wafer, the third wafer, and the fifth wafer may have more major small CD defects, as evidenced by the longest negative second bar in the corresponding bar graphs.
图9D中的条形图示出，图9D的每个晶片在蚀刻后可能包括边缘模糊的缺陷(例如，主要缺陷或多个缺陷中的缺陷)，如长的负的第七条所证明的。可以理解，模糊边缘可能是第四晶片中的主要缺陷。此外，第一晶片、第二晶片、第三晶片和第五晶片可能具有更主要的小CD缺陷，如相应条形图中最长的负的第二条所证明的。此外，第二晶片可能不具有与垂直偏移相关联的缺陷，如与第三分量相对应的明显正的条所证明的。The bar graphs in FIG. 9D show that each of the wafers in FIG. 9D may include a fuzzy edge defect (e.g., a major defect or a defect among multiple defects) after etching, as evidenced by the long negative seventh bar. It can be appreciated that the fuzzy edge may be a major defect in the fourth wafer. In addition, the first wafer, the second wafer, the third wafer, and the fifth wafer may have more dominant small CD defects, as evidenced by the longest negative second bar in the corresponding bar graphs. In addition, the second wafer may not have defects associated with vertical offset, as evidenced by the significantly positive bar corresponding to the third component.
图9E中的条形图示出，图9E的每个晶片在蚀刻后可能包括水平偏移的缺陷(例如，主要缺陷或多个缺陷中的缺陷)，如长的负的第四条所证明的。此外，每个晶片可能具有更主要的小CD缺陷，如相应条形图中最长的负的第二条所证明的。此外，第一晶片、第二晶片和第四晶片可能不具有与椭圆度相关联的缺陷，如与第四分量相对应的正的条所证明的。The bar graphs in FIG9E show that each wafer of FIG9E may include a horizontally offset defect (e.g., a major defect or a defect among multiple defects) after etching, as evidenced by the long negative fourth bar. In addition, each wafer may have a more dominant small CD defect, as evidenced by the longest negative second bar in the corresponding bar graph. In addition, the first wafer, the second wafer, and the fourth wafer may not have defects associated with ellipticity, as evidenced by the positive bars corresponding to the fourth component.
图10示出了根据本公开的一些实施例的使用逻辑分类器模型而获取的可视化预测结果(例如，如条形图)的示例。在一些实施例中，用于对不同类别的缺陷进行分类的分类器模型可以包括逻辑分类器模型。例如，逻辑分类器模型可以包括逻辑回归模型，逻辑回归模型可以用于通过输出指示某些类别是否可能存在的二进制结果来预测不同缺陷类别。因此，可以直接标识输入SEM图像的重要分量(例如，重要缺陷类别)，如图10所示。在一些实施例中，黑条可以对应于根据上述等式(2)而计算的wi，以指示是否存在某种缺陷类型。白条可以对应于根据上述等式(2)而确定的bi，以指示来自对应类型的缺陷对输入SEM图像的贡献的重要性。例如，基于图10所示的白条的大小，可以确定第二分量、第三分量、第四分量、第七分量、第十分量和第十五分量是对输入SEM图像的主要贡献缺陷。FIG. 10 shows an example of a visual prediction result (e.g., as a bar chart) obtained using a logistic classifier model according to some embodiments of the present disclosure. In some embodiments, a classifier model for classifying different categories of defects may include a logistic classifier model. For example, a logistic classifier model may include a logistic regression model, which may be used to predict different defect categories by outputting a binary result indicating whether certain categories may exist. Therefore, important components (e.g., important defect categories) of the input SEM image may be directly identified, as shown in FIG. 10. In some embodiments, the black bar may correspond to wi calculated according to the above equation (2) to indicate whether a certain defect type exists. The white bar may correspond to bi determined according to the above equation (2) to indicate the importance of the contribution of the defect from the corresponding type to the input SEM image. For example, based on the size of the white bar shown in FIG. 10, it can be determined that the second component, the third component, the fourth component, the seventh component, the tenth component, and the fifteenth component are the main contributing defects to the input SEM image.
图11A是与本公开的一些实施例一致的表示用于基于输入SEM图像来预测失败模式(例如，缺陷类别)的示例方法1100的过程流程图。在一些实施例中，一个或多个步骤由图3中的系统300、图1中的控制器109或系统199、或图1中的系统100的一个或多个组件来执行。FIG. 11A is a process flow diagram representing an example method 1100 for predicting a failure mode (e.g., defect class) based on an input SEM image consistent with some embodiments of the present disclosure. In some embodiments, one or more steps are performed by one or more components of the system 300 in FIG. 3 , the controller 109 or the system 199 in FIG. 1 , or the system 100 in FIG. 1 .
在步骤1110中，通过分解输入电子显微镜图像(例如，图6中的输入图像610)来获取与多个可解释模式相对应的多个模式图像(例如，图6中的模式图像620)。可以获取分别与多个可解释模式相关联的多个系数(例如，图6的C1、C2、C3……)，以用于表征输入电子显微镜图像。例如，输入电子显微镜图像包括SEM图像，诸如输入SEM图像610，该SEM图像是由图像获取器340从控制器109、系统199或电子束工具104获取的。在一些实施例中，输入电子显微镜图像可以反映晶片上与特定特征相对应的区域，诸如接触孔区域。在一些实施例中，输入电子显微镜图像可以对应于在光刻中的已经被显影处理之后的晶片，并且过程1100可以预测蚀刻之后晶片上可能存在的一个或多个缺陷。在一些实施例中，输入电子显微镜图像可以使用PCA或任何其他合适的方法来分解以获取与相应可解释模式相关联的系数。相应可解释模式可以与晶片上对应区域的特征(诸如缺陷类型)相关联。In step 1110, a plurality of pattern images (e.g., pattern image 620 in FIG. 6 ) corresponding to a plurality of interpretable patterns are obtained by decomposing an input electron microscope image (e.g., input image 610 in FIG. 6 ). A plurality of coefficients (e.g., C1, C2, C3, ... in FIG. 6 ) respectively associated with a plurality of interpretable patterns may be obtained for characterizing the input electron microscope image. For example, the input electron microscope image includes an SEM image, such as input SEM image 610, which is acquired by image acquirer 340 from controller 109, system 199, or electron beam tool 104. In some embodiments, the input electron microscope image may reflect an area on a wafer corresponding to a specific feature, such as a contact hole area. In some embodiments, the input electron microscope image may correspond to a wafer after a development process in photolithography, and process 1100 may predict one or more defects that may exist on the wafer after etching. In some embodiments, the input electron microscope image may be decomposed using PCA or any other suitable method to obtain coefficients associated with corresponding interpretable patterns. The corresponding interpretable pattern may be associated with a feature (such as a defect type) of a corresponding area on the wafer.
在步骤1120中，评估与多个可解释模式相关联的系数。在一些实施例中，分类器模型332或630(例如，由过程500训练的如图6所示的卷积神经网络模型)可以被应用以评估多个模式图像的系数。例如，相应评估结果指示在输入电子显微镜图像中存在对应可解释模式的可能性(例如，二进制结果)。在一些实施例中，从PCA获取的多个模式图像的系数被应用于神经网络模型630的输入层中的节点，并且输出是评估结果，诸如图6所示的二进制结果。In step 1120, coefficients associated with multiple interpretable patterns are evaluated. In some embodiments, a classifier model 332 or 630 (e.g., a convolutional neural network model as shown in FIG. 6 trained by process 500) can be applied to evaluate the coefficients of multiple pattern images. For example, the corresponding evaluation results indicate the possibility (e.g., a binary result) of the presence of a corresponding interpretable pattern in the input electron microscope image. In some embodiments, the coefficients of the multiple pattern images obtained from PCA are applied to the nodes in the input layer of the neural network model 630, and the output is an evaluation result, such as the binary result shown in FIG. 6.
在步骤1130中，基于从步骤1120获取的评估结果来确定多个可解释模式对输入电子显微镜图像的贡献。在一些实施例中，回归模型(诸如线性模型650)或任何其他合适的多项式模型(诸如二次模型)可以用于非线性分类器模型与评估结果之间的关系进行近似。例如，如图7所示，可以选择与输入SEM图像相对应的点P，并且可以标识感兴趣区域以执行线性近似。在如图8和图9A-图9E中讨论的一些实施例中，可以使用从上述线性化或等式(2)获取的诸如wi等系数来确定某些模式是否对输入SEM图像有贡献(例如，是否可能存在对应缺陷)，并且大小值bi指示对应模式对输入SEM图像有多重要。In step 1130, the contribution of the plurality of interpretable patterns to the input electron microscope image is determined based on the evaluation results obtained from step 1120. In some embodiments, a regression model (such as the linear model 650) or any other suitable polynomial model (such as a quadratic model) can be used to approximate the relationship between the nonlinear classifier model and the evaluation results. For example, as shown in FIG. 7, a point P corresponding to the input SEM image can be selected, and a region of interest can be identified to perform a linear approximation. In some embodiments such as discussed in FIG. 8 and FIG. 9A-9E, coefficients such as wi obtained from the above linearization or equation (2) can be used to determine whether certain patterns contribute to the input SEM image (e.g., whether corresponding defects may exist), and the magnitude value bi indicates how important the corresponding pattern is to the input SEM image.
在步骤1140中，可以基于来自步骤1130的所确定的贡献来预测晶片上的一个或多个特性，诸如缺陷类型。在一些实施例中，在步骤1130中确定的贡献可以在图表或图形中可视化，诸如图8和图9A-图9E所示的条形图，以向用户提供对半导体制造中失败的各种原因的直接理解。在一些实施例中，从过程1100获取的预测结果可以用于根据标识的晶片上的区域中的一个或多个模式来调节一个或多个工艺参数。In step 1140, one or more characteristics on the wafer, such as defect types, may be predicted based on the determined contributions from step 1130. In some embodiments, the contributions determined in step 1130 may be visualized in a chart or graph, such as the bar graphs shown in FIG8 and FIG9A-9E, to provide a user with an immediate understanding of the various causes of failure in semiconductor manufacturing. In some embodiments, the prediction results obtained from process 1100 may be used to adjust one or more process parameters based on one or more patterns in the identified regions on the wafer.
图11B示出了与本公开的一些实施例一致的可视化了基于多个SEM图像的可解释模式而进行的聚类的图1150和晶片上的聚类结果的映射图1160。在一些实施例中，图1150的聚类由图3中的系统300的一个或多个组件(例如，分析模块304)或者图1中的控制器109或系统199执行。在一些实施例中，图像获取器340从控制器109、系统199或电子束工具104获取多个输入SEM图像。在一些实施例中，每个SEM图像包括晶片上的一个或多个区域，其中区域对应于接触孔(例如，如SEM图像610中所示)或另一类型的特征。在一些实施例中，多个输入SEM图像对应于晶片上的不同区域。在一些实施例中，多个输入SEM图像对应于同一批次中的不同晶片上的区域。FIG. 11B shows a graph 1150 and a map 1160 of clustering results on a wafer that visualizes clustering based on interpretable patterns of multiple SEM images consistent with some embodiments of the present disclosure. In some embodiments, the clustering of FIG. 1150 is performed by one or more components (e.g., analysis module 304) of system 300 in FIG. 3 or controller 109 or system 199 in FIG. 1. In some embodiments, image acquirer 340 acquires multiple input SEM images from controller 109, system 199, or electron beam tool 104. In some embodiments, each SEM image includes one or more regions on a wafer, where the regions correspond to contact holes (e.g., as shown in SEM image 610) or another type of feature. In some embodiments, multiple input SEM images correspond to different regions on a wafer. In some embodiments, multiple input SEM images correspond to regions on different wafers in the same batch.
在一些实施例中，每个输入SEM图像可以使用图6的过程600或图11A的过程1100的一个或多个步骤来处理，以标识多个可解释模式并且量化来自各个可解释模式的贡献。可解释模式的示例可以对应于不同特征类别(例如，缺陷或失败的可能原因)，诸如小CD、沿着某个方向的偏移、椭圆度、模糊边缘、印刷接触孔、缺失接触孔、桥接接触孔等。例如，如图1150所示，对于每个输入SEM图像，一个或多个可解释模式和来自对应可解释模式的量化贡献通过条形图中的条可视化，这些条具有不同通道，具有对应方向和大小，如图8中讨论的。在一些实施例中，由多个分量组成的多维向量可以用于表征每个输入SEM图像中的特征，其中每个分量对应于可解释模式，具有对应可解释模式对预测的贡献的相关联的量化。例如，可以使用10维向量来表征图11B中的相应SEM图像中的特征，并且该10维向量包括10个分量，每个分量对应于条形图中的特定通道中的条。In some embodiments, each input SEM image may be processed using one or more steps of process 600 of FIG. 6 or process 1100 of FIG. 11A to identify multiple interpretable patterns and quantify contributions from each interpretable pattern. Examples of interpretable patterns may correspond to different feature categories (e.g., possible causes of defects or failures), such as small CD, offset along a certain direction, ellipticity, fuzzy edges, printed contact holes, missing contact holes, bridged contact holes, etc. For example, as shown in FIG. 1150, for each input SEM image, one or more interpretable patterns and quantified contributions from corresponding interpretable patterns are visualized by bars in a bar graph, the bars having different channels with corresponding directions and magnitudes, as discussed in FIG. 8. In some embodiments, a multi-dimensional vector consisting of multiple components may be used to characterize features in each input SEM image, wherein each component corresponds to an interpretable pattern with an associated quantification of the contribution of the corresponding interpretable pattern to the prediction. For example, a 10-dimensional vector may be used to characterize features in the corresponding SEM image in FIG. 11B, and the 10-dimensional vector includes 10 components, each corresponding to a bar in a specific channel in the bar graph.
在一些实施例中，基于晶片上不同区域的多个SEM图像的向量的某个(某些)分量的相似性，对向量进行聚类。使用图1150所示的二维空间作为示例，每个点对应于晶片上的区域的SEM图像，该SEM图像与二维向量(从过程600或1100获取)相关联。具有相似分量的向量在二维空间中用相似坐标表示，因此分布在彼此靠近的一个集群中，基于对对应SEM图像的分析，表明晶片上的这些区域具有相似缺陷原因。在一些实施例中，聚类也可以被应用于作为一个组一起处理的批次中的多个晶片。在一些实施例中，可以使用任何类型的合适的聚类算法，诸如K均值聚类或均值偏移聚类等。In some embodiments, vectors are clustered based on the similarity of certain components of vectors of multiple SEM images of different regions on the wafer. Using the two-dimensional space shown in Figure 1150 as an example, each point corresponds to an SEM image of a region on the wafer, which is associated with a two-dimensional vector (obtained from process 600 or 1100). Vectors with similar components are represented by similar coordinates in the two-dimensional space and are therefore distributed in a cluster close to each other, indicating that these regions on the wafer have similar causes of defects based on analysis of the corresponding SEM images. In some embodiments, clustering can also be applied to multiple wafers in a batch that are processed together as a group. In some embodiments, any type of suitable clustering algorithm can be used, such as K-means clustering or mean shift clustering.
在聚类之后，可以在晶片上提供聚类结果的映射图1160，以便于分析失败原因。在一些实施例中，聚类图1150中的每个点对应于该区域的SEM图像，并且还与区域信息相关联，诸如晶片上的位置信息(例如，在哪个管芯中，或者晶片上的坐标)、对该区域执行的一个或多个制造过程的时间信息(例如，何时执行了涂覆、曝光、烘焙、显影、蚀刻、抛光等)等。在一些实施例中，如映射图1160所示，图1150中获取的点集群可以映射在晶片上。可以在映射期间获取晶片上的位置分布或处理时间信息，并且可以分析与缺陷分布相关联的光刻参数以了解缺陷的起源。例如，与(集群中的)给定失败模式相对应的接触孔可以映射在晶片上，以可视化每个接触孔与该集群的质心之间的距离。After clustering, a map 1160 of the clustering results can be provided on the wafer to facilitate analysis of the cause of failure. In some embodiments, each point in the cluster map 1150 corresponds to an SEM image of the area, and is also associated with regional information, such as location information on the wafer (e.g., in which die, or coordinates on the wafer), time information of one or more manufacturing processes performed on the area (e.g., when coating, exposure, baking, development, etching, polishing, etc. were performed), etc. In some embodiments, as shown in the map 1160, the cluster of points obtained in the map 1150 can be mapped on the wafer. The position distribution or processing time information on the wafer can be obtained during mapping, and the lithography parameters associated with the defect distribution can be analyzed to understand the origin of the defect. For example, the contact holes corresponding to a given failure mode (in the cluster) can be mapped on the wafer to visualize the distance between each contact hole and the centroid of the cluster.
在一些实施例中，可以基于映射结果来分析失败原因。例如，在映射图1160中，具有小尺寸的接触孔集群(例如，集群1)可以分布在晶片上的同一管芯内。因此，可以调节一个或多个光刻参数，诸如光剂量，以解决在对应管芯上生成的这样的缺陷。在另一示例中，在晶片上的同一区域内出现的具有模糊边缘的多个接触孔可以对应于焦点相关缺陷或失败。因此，可以调节在光刻期间施加到该区域的光焦点。在另一示例中，具有特定y偏移或x偏移的多个接触孔可以在映射图1160上形成图案，表明在蚀刻或抛光工艺期间生成的缺陷。因此，可以调节蚀刻或抛光参数，诸如蚀刻速率、抛光接触压力、晶片旋转速度等，以改进蚀刻或抛光工艺。In some embodiments, the cause of failure can be analyzed based on the mapping results. For example, in the mapping 1160, a cluster of contact holes with a small size (e.g., cluster 1) can be distributed within the same tube core on the wafer. Therefore, one or more lithography parameters, such as light dose, can be adjusted to address such defects generated on the corresponding tube core. In another example, multiple contact holes with fuzzy edges appearing in the same area on the wafer can correspond to focus-related defects or failures. Therefore, the focus of light applied to the area during lithography can be adjusted. In another example, multiple contact holes with a specific y offset or x offset can form a pattern on the mapping 1160, indicating defects generated during the etching or polishing process. Therefore, etching or polishing parameters, such as etching rate, polishing contact pressure, wafer rotation speed, etc., can be adjusted to improve the etching or polishing process.
在一些实施例中，上述聚类和映射过程可以应用于一批次中的晶片上的各种区域的SEM图像，以更好地理解失败原因。例如，如果一批晶片在相似位置(例如，在相同管芯中)包括相似缺陷，则可以调节在处理该批晶片时应用于这些管芯的参数，以解决缺陷。In some embodiments, the clustering and mapping process described above can be applied to SEM images of various areas on wafers in a batch to better understand the causes of failures. For example, if a batch of wafers includes similar defects in similar locations (e.g., in the same die), the parameters applied to those dies when processing the batch of wafers can be adjusted to address the defects.
在一些实施例中，还可以基于对特定参数的用户选择来绘制映射，诸如特定类型的缺陷或缺陷原因(例如，小CD、模糊边缘等)、晶片上的特定区域(例如，特定管芯)、用于处理晶片或管芯的特定时间或特定步骤，以了解在所选择的区域中或在某个时间或步骤发生了什么类型的缺陷(或缺陷原因)，从而改进对应工艺。在一些实施例中，响应于用户选择，晶片上缺陷的分布可以被可视化给用户，使得用户可以更高效和有效地查看和确定失败原因。在一些实施例中，基于晶片级的缺陷分布而分析的失败原因可以用于预测其他未测量甚至未处理的晶片的潜在问题，以便采取预先的校正来改进工艺。例如，特定失败模式的晶片指纹可以映射在晶片上，并且用于监测和诊断未来的晶片测量和处理。In some embodiments, a map can also be drawn based on user selection of specific parameters, such as a specific type of defect or defect cause (e.g., small CD, fuzzy edge, etc.), a specific area on the wafer (e.g., a specific die), a specific time or a specific step for processing the wafer or die, to understand what type of defect (or defect cause) occurred in the selected area or at a certain time or step, so as to improve the corresponding process. In some embodiments, in response to user selection, the distribution of defects on the wafer can be visualized to the user, so that the user can view and determine the cause of failure more efficiently and effectively. In some embodiments, the cause of failure analyzed based on the defect distribution at the wafer level can be used to predict potential problems of other unmeasured or even unprocessed wafers, so as to take advance corrections to improve the process. For example, a wafer fingerprint of a specific failure mode can be mapped on the wafer and used to monitor and diagnose future wafer measurements and processing.
图11C是与本公开的一些实施例一致的表示用于基于晶片上的多个区域(例如，在晶片级)的输入SEM图像来分析失败模式的示例方法1170的过程流程图。在一些实施例中，一个或多个步骤由图3中的系统300、图1中的控制器109或系统199、或图1中的系统100的一个或多个组件来执行。FIG11C is a process flow diagram representing an example method 1170 for analyzing failure modes based on input SEM images of multiple regions on a wafer (e.g., at the wafer level), consistent with some embodiments of the present disclosure. In some embodiments, one or more steps are performed by one or more components of the system 300 in FIG3, the controller 109 or the system 199 in FIG1, or the system 100 in FIG1.
在步骤1172中，获取与晶片上的不同区域相对应的多个输入电子显微镜图像(例如，包括图6的输入图像610)。在一些实施例中，多个输入电子显微镜图像可以对应于同一批次中的不同晶片上的不同区域。In step 1172, a plurality of input electron microscope images corresponding to different regions on a wafer are acquired (eg, including input image 610 of FIG. 6). In some embodiments, the plurality of input electron microscope images may correspond to different regions on different wafers in the same batch.
在步骤1174中，针对每个输入电子显微镜图像，可以确定包括与可解释模式的贡献相对应的分量的多维向量。在一些实施例中，每个电子显微镜图像通过图11A中的过程1100的一个或多个步骤来处理。例如，针对每个电子显微镜图像，可以分析多个可解释模式，并且确定多个可解释模式对输入电子显微镜图像的贡献。多维向量由表示多个可解释模式和相关联贡献的分量组成。In step 1174, for each input electron microscope image, a multidimensional vector including components corresponding to the contributions of the interpretable patterns can be determined. In some embodiments, each electron microscope image is processed by one or more steps of process 1100 in Figure 11A. For example, for each electron microscope image, multiple interpretable patterns can be analyzed and the contributions of the multiple interpretable patterns to the input electron microscope image can be determined. The multidimensional vector is composed of components representing the multiple interpretable patterns and the associated contributions.
在步骤1176中，基于可解释模式对多个输入电子显微镜图像的多维向量进行聚类。例如，具有相似分量和相关联贡献的向量分布在多维空间上的同一集群中(例如，如图11B所示，在二维空间中)。In step 1176, the multidimensional vectors of the plurality of input electron microscope images are clustered based on interpretable patterns. For example, vectors with similar components and associated contributions are distributed in the same cluster in a multidimensional space (e.g., in a two-dimensional space as shown in FIG. 11B ).
在步骤1178中，对聚类结果进行映射，例如，如图11B中的映射图1160所示，并且基于映射结果来分析失败原因。例如，可以将与某个可解释模式相对应的集群内的向量映射到晶片上，以可视化晶片上对应区域的分布。在一些实施例中，用户还可以选择参数，诸如选择晶片上的管芯以检查管芯上发生了什么类型的缺陷(或缺陷原因)、或者选择某个缺陷(或失败原因)并且查看晶片上该类型的缺陷的分布。In step 1178, the clustering results are mapped, for example, as shown in the mapping diagram 1160 in FIG. 11B, and the failure causes are analyzed based on the mapping results. For example, vectors within a cluster corresponding to a certain interpretable pattern can be mapped to a wafer to visualize the distribution of corresponding areas on the wafer. In some embodiments, the user can also select parameters, such as selecting a die on a wafer to check what type of defect (or defect cause) occurred on the die, or selecting a defect (or failure cause) and viewing the distribution of that type of defect on the wafer.
如本文中所述，通过执行过程500、600、1100或1150，系统300不仅可以提供不同类型的缺陷对晶片上的区域的失败的贡献的量化评估，还可以向用户提供对个体区域的这样的量化分析的解释以及晶片级的缺陷分布。因此，用户可以更直接地了解半导体制造中的失败的各种原因。此外，预测缺陷的早期检测和校正变得可能。例如，训练图像和输入图像可以对应于已经在半导体处理的不同阶段处理的晶片上的不同区域或层。这继而可以有助于半导体工艺的系统改进。此外，与现有技术相比，训练和使用分类器模型可以不那么复杂，也不那么耗时，因为在本公开中，与分解后的模式相关联的权重(例如，系数)被用作输入，而不是使用像素值作为输入。因此，输入层中涉及的节点更少，神经网络也不那么复杂，从而为半导体处理提供更高效和更有效的基于机器学习的缺陷预测。As described herein, by performing processes 500, 600, 1100 or 1150, system 300 can not only provide a quantitative assessment of the contribution of different types of defects to the failure of an area on a wafer, but also provide the user with an explanation of such a quantitative analysis of individual areas and defect distribution at the wafer level. Therefore, the user can more directly understand the various causes of failure in semiconductor manufacturing. In addition, early detection and correction of predicted defects become possible. For example, the training image and the input image can correspond to different areas or layers on a wafer that has been processed at different stages of semiconductor processing. This in turn can contribute to the system improvement of semiconductor processes. In addition, compared with the prior art, training and using a classifier model can be less complex and less time-consuming because in the present disclosure, the weights (e.g., coefficients) associated with the decomposed pattern are used as inputs instead of using pixel values as inputs. Therefore, there are fewer nodes involved in the input layer and the neural network is less complex, thereby providing more efficient and more effective machine learning-based defect prediction for semiconductor processing.
此外，本公开中公开的过程也可以用于根本原因分析。例如，可以基于与半导体处理期间的不同处理步骤、阶段或参数相对应的可解释模式来训练分类器模型。然后，可以获取来自不同处理步骤、阶段或参数的贡献的排名，以用于特征重要性检测。在一些实施例中，代替使用通用的机器学习模型，还可以使用用于特征重要性检测的模型特定方法。例如，当使用针对随机森林模型的决策树时，可以根据该特征在树中的位置来确定特征重要性(例如，该特征在该树中的位置越高，该特征可以越重要)。In addition, the processes disclosed in the present disclosure can also be used for root cause analysis. For example, a classifier model can be trained based on interpretable patterns corresponding to different processing steps, stages, or parameters during semiconductor processing. Then, a ranking of contributions from different processing steps, stages, or parameters can be obtained for feature importance detection. In some embodiments, instead of using a general machine learning model, a model-specific method for feature importance detection can also be used. For example, when using a decision tree for a random forest model, feature importance can be determined based on the position of the feature in the tree (for example, the higher the position of the feature in the tree, the more important the feature can be).
根本原因分析对于标识IC芯片上各种缺陷的显著原因非常重要，因此可以基于标识的原因来优化半导体制造过程。目前，根本原因分析主要由经验丰富的工程师手动执行。例如，经验丰富的人员可以检查显微镜图像上的缺陷，分析有缺陷的化学成分，或对IC芯片上的电气失败进行测试，以了解缺陷原因。然而，这种手动过程可能耗时、容易出错，并且仅限于少量样本中的少量缺陷类型。因此，需要一种用于根本原因分析的自动过程，该过程能够满足具有减小的特征尺寸、增加的特征密度以及更详细和准确的缺陷分析的大规模IC芯片处理的需要。Root cause analysis is important for identifying significant causes of various defects on IC chips so that the semiconductor manufacturing process can be optimized based on the identified causes. Currently, root cause analysis is mainly performed manually by experienced engineers. For example, experienced personnel can examine defects on microscope images, analyze defective chemical compositions, or test electrical failures on IC chips to understand the cause of defects. However, this manual process can be time consuming, error-prone, and limited to a small number of defect types in a small number of samples. Therefore, there is a need for an automated process for root cause analysis that can meet the needs of large-scale IC chip processing with reduced feature sizes, increased feature density, and more detailed and accurate defect analysis.
为了解决这个问题，本公开提供了一种适用于大规模半导体制造过程中的自动根本原因分析的方法和系统。例如，可以基于图像特征数据和工艺特征数据来训练预测模型，诸如随机森林模型。预测模型可以用于基于用于制造用于预测的输入样本的工艺特征数据(诸如工艺参数)来预测缺陷的形成。此外，预测模型可以提供对特征进行排名的特征重要性信息，以指示缺陷的根本原因的排序。To solve this problem, the present disclosure provides a method and system for automatic root cause analysis in a large-scale semiconductor manufacturing process. For example, a prediction model, such as a random forest model, can be trained based on image feature data and process feature data. The prediction model can be used to predict the formation of defects based on process feature data (such as process parameters) used to manufacture input samples for prediction. In addition, the prediction model can provide feature importance information for ranking features to indicate the ranking of the root causes of defects.
图12是与本公开的一些实施例一致的被配置为基于从预测模型获取的特征排名结果来执行根本原因分析的示例系统1200的框图。在一些实施例中，系统1200包括训练模块1202和预测模块1204。训练模块1202包括训练数据获取器1210、标记数据获取器1205和模型训练器1220。预测模块1204包括分析器1230，分析器1230包括模型1232(例如，由预训练的训练器1220生成的)。分析器1230可以分析由数据获取器1240获取的输入数据，以生成特征排名结果12650。分析器1230还可以基于输入数据生成缺陷预测结果1250。12 is a block diagram of an example system 1200 configured to perform root cause analysis based on feature ranking results obtained from a prediction model consistent with some embodiments of the present disclosure. In some embodiments, the system 1200 includes a training module 1202 and a prediction module 1204. The training module 1202 includes a training data acquirer 1210, a labeled data acquirer 1205, and a model trainer 1220. The prediction module 1204 includes an analyzer 1230, which includes a model 1232 (e.g., generated by a pre-trained trainer 1220). The analyzer 1230 can analyze the input data acquired by the data acquirer 1240 to generate a feature ranking result 12650. The analyzer 1230 can also generate a defect prediction result 1250 based on the input data.
在一些实施例中，系统1200包括一个或多个处理器和存储器。例如，系统1200可以包括一个或多个计算机、服务器、主机、终端、个人计算机、任何种类的移动计算设备等、或其组合。在一些实施例中，训练模块1202和预测模块1204在单独的计算设备上实现。在其他实施例中，训练模块1202和预测模块1204可以在同一计算设备上实现。应当理解，系统1200可以包括被集成为带电粒子束检查系统(例如，图1的电子束检查系统100)的一部分的一个或多个组件或模块。系统1200还可以包括与带电粒子束检查系统分离并且通信耦合到带电粒子束检查系统的一个或多个组件或模块。在一些实施例中，系统1200可以包括一个或多个组件(例如，软件模块)，该组件可以在如本文中讨论的控制器109或系统199中实现。In some embodiments, system 1200 includes one or more processors and memory. For example, system 1200 may include one or more computers, servers, mainframes, terminals, personal computers, any kind of mobile computing devices, etc., or a combination thereof. In some embodiments, training module 1202 and prediction module 1204 are implemented on separate computing devices. In other embodiments, training module 1202 and prediction module 1204 may be implemented on the same computing device. It should be understood that system 1200 may include one or more components or modules that are integrated as a part of a charged particle beam inspection system (e.g., the electron beam inspection system 100 of FIG. 1). System 1200 may also include one or more components or modules that are separated from the charged particle beam inspection system and are communicatively coupled to the charged particle beam inspection system. In some embodiments, system 1200 may include one or more components (e.g., software modules) that may be implemented in controller 109 or system 199 as discussed herein.
在图12所示的一些实施例中，训练模块1202包括训练数据获取器1210。训练数据获取器1210可以被配置为获取训练数据。训练数据可以包括从IC芯片的多个SEM图像中提取的图像数据1207、以及与IC芯片的制造相关联的工艺数据1208。所获取的训练数据可以被馈送到用于训练模型1232(例如，预测模型)的模型训练器1220。在一些实施例中，训练数据获取器1210可以从数据库、控制器109、系统199或电子束工具104获取训练数据。例如，训练数据获取器1210可以包括如本文中讨论的用于获取IC芯片的多个SEM图像的控制器109的图像获取器。In some embodiments shown in FIG. 12 , the training module 1202 includes a training data acquirer 1210. The training data acquirer 1210 may be configured to acquire training data. The training data may include image data 1207 extracted from a plurality of SEM images of an IC chip, and process data 1208 associated with the manufacture of the IC chip. The acquired training data may be fed to a model trainer 1220 for training a model 1232 (e.g., a predictive model). In some embodiments, the training data acquirer 1210 may acquire training data from a database, a controller 109, a system 199, or an electron beam tool 104. For example, the training data acquirer 1210 may include an image acquirer of a controller 109 for acquiring a plurality of SEM images of an IC chip as discussed herein.
在一些实施例中，训练数据的图像数据1207可以包括与IC芯片上的各种特征或缺陷相关联的像素值、位置信息等。在一些实施例中，相应SEM图像可以捕获与特征(例如，一个或多个接触孔、一个或多个线等)、管芯或整个晶片相对应的区域。在一些实施例中，图像数据1207可以从在不同阶段处理的样品的SEM图像中提取，诸如在光刻中的显影之后、在蚀刻之后、在金属层沉积之后、在化学机械抛光(CMP)之后等。In some embodiments, the image data 1207 of the training data may include pixel values, location information, etc. associated with various features or defects on the IC chip. In some embodiments, the corresponding SEM image may capture an area corresponding to a feature (e.g., one or more contact holes, one or more lines, etc.), a die, or an entire wafer. In some embodiments, the image data 1207 may be extracted from SEM images of samples processed at different stages, such as after development in photolithography, after etching, after metal layer deposition, after chemical mechanical polishing (CMP), etc.
在一些实施例中，工艺训练数据获取器1210可以进一步获取与用于制造IC芯片的不同工艺相关联的工艺数据1208，工艺数据1208用于对多个SEM图像做出贡献，以用于训练。在一些实施例中，工艺数据1208包括但不限于从不同半导体工艺或检查阶段收集的制造数据、用于设计IC芯片上的微电路的设计数据、材料信息(例如，成分)、以及其他类型的可能的缺陷原因。In some embodiments, the process training data acquirer 1210 may further acquire process data 1208 associated with different processes used to manufacture IC chips, the process data 1208 being used to contribute to the plurality of SEM images for training. In some embodiments, the process data 1208 includes, but is not limited to, manufacturing data collected from different semiconductor processes or inspection stages, design data used to design microcircuits on IC chips, material information (e.g., composition), and other types of possible defect causes.
在一些实施例中，制造数据包括与光刻工艺、蚀刻工艺、检查条件和制造过程中涉及的其他工艺相关的参数(例如，也称为特征)。例如，光刻参数包括但不限于光束焦点、剂量和透镜像差值、晶片平整和双重图案化工艺中的重叠校正等。蚀刻参数包括但不限于蚀刻温度、蚀刻化学物质(例如，气体)浓度水平和蚀刻持续时间等。检查参数包括但不限于使用亮场或暗场显微镜图像的光学检查条件、放大率、扫描区域等，其可以指示检查期间的缺陷检查。In some embodiments, the manufacturing data includes parameters (e.g., also referred to as features) related to the lithography process, the etching process, the inspection conditions, and other processes involved in the manufacturing process. For example, the lithography parameters include, but are not limited to, beam focus, dose and lens aberration values, wafer flattening, and overlap correction in double patterning processes, etc. The etching parameters include, but are not limited to, etching temperature, etching chemical (e.g., gas) concentration level, and etching duration, etc. The inspection parameters include, but are not limited to, optical inspection conditions using bright field or dark field microscope images, magnification, scanning area, etc., which can indicate defect detection during inspection.
在一些实施例中，设计数据对应于将在晶片上的多个分层上形成的设计架构。设计数据可以以图像文件来呈现，并且可以包括不同层上的各种图案的特征信息(例如，形状、尺寸等)。例如，设计数据可以与和要在晶片上制造的各种结构、器件和系统相关联的信息相关，包括但不限于衬底、掺杂区、多晶硅栅极(poly-gate)层、电阻层、介电层、金属层、晶体管、处理器、存储器、金属连接、触点、过孔、片上系统(SoC)、片上网络(NoC)、或任何其他合适的结构。设计数据还可以包括存储器块、逻辑块、互连等的IC布局设计。例如，设计数据可以包括参数或特性，包括但不限于图案密度、图案/特征在微芯片掩模版/场上的位置，其可以与IC芯片上的缺陷相关联。在一些实施例中，设计数据可以是图形数据库系统(GDS)格式、图形数据库系统II(GDS II)格式、开放艺术作品系统交换标准(OASIS)格式、加州理工学院中间格式(CIF)等。In some embodiments, the design data corresponds to a design architecture to be formed on multiple layers on a wafer. The design data may be presented as an image file and may include feature information (e.g., shape, size, etc.) of various patterns on different layers. For example, the design data may be related to information associated with various structures, devices, and systems to be manufactured on a wafer, including but not limited to substrates, doped regions, polysilicon gate layers, resistive layers, dielectric layers, metal layers, transistors, processors, memories, metal connections, contacts, vias, systems on chip (SoC), networks on chip (NoC), or any other suitable structure. The design data may also include IC layout designs of memory blocks, logic blocks, interconnections, etc. For example, the design data may include parameters or characteristics, including but not limited to pattern density, the location of patterns/features on microchip masks/fields, which may be associated with defects on IC chips. In some embodiments, the design data may be in a Graphic Database System (GDS) format, a Graphic Database System II (GDS II) format, an Open Artwork System Exchange Standard (OASIS) format, a California Institute of Technology intermediate format (CIF), etc.
在一些实施例中，工艺数据1208还可以包括与其他类型的可能的缺陷原因相关的数据，诸如管芯上的划痕或残留物、以及材料成分等，这些数据可以指示哪种工艺是根本原因。In some embodiments, process data 1208 may also include data related to other types of possible defect causes, such as scratches or residues on the die, and material composition, which may indicate which process is the root cause.
在一些实施例中，训练模块1202可以包括标记数据获取器1205，标记数据获取器1205被配置为获取与SEM图像相关联的标记数据1206，从该SEM图像获取用于训练的图像数据1207。在一些实施例中，SEM图像由分类标记来标记，诸如缺陷或非缺陷。附加于或替代二进制标记，SEM图像还可以通过与不同缺陷类型相对应的多个类别来标记。例如，缺陷类别包括但不限于桥接、颈缩、缺失、合并、小CD、边缘模糊、椭圆度等。分类标记可以用于构建用于预测分类的分类型预测模型。在一些实施例中，SEM图像可以由专家基于其先验知识进行检查和标记。在一些实施例中，SEM图像可以使用自动程序来自动分析和标记。例如，主成分分析(PCA)或奇异值分解(SVD)方法。在一些实施例中，SEM图像通过回归标记来标记，例如基于连续图案大小。回归标记可以用于建立回归类型的预测模型。例如，计算机视觉算法可以用于将感兴趣区域(或感兴趣图案)分类为缺陷或非缺陷，或者测量连续图案大小。例如，感兴趣图案可以被分类为图案中断或图案桥接。在另一示例中，感兴趣图案也可以通过分布在不同范围内的大小来测量，诸如5-10nm范围等。In some embodiments, the training module 1202 may include a labeled data acquirer 1205, which is configured to acquire labeled data 1206 associated with the SEM image, and acquire image data 1207 for training from the SEM image. In some embodiments, the SEM image is labeled by a classification label, such as a defect or non-defect. In addition to or in place of a binary label, the SEM image can also be labeled by a plurality of categories corresponding to different defect types. For example, defect categories include, but are not limited to, bridging, necking, missing, merging, small CD, edge blur, ellipticity, etc. The classification label can be used to build a classification prediction model for predicting classification. In some embodiments, the SEM image can be inspected and labeled by an expert based on his prior knowledge. In some embodiments, the SEM image can be automatically analyzed and labeled using an automatic program. For example, principal component analysis (PCA) or singular value decomposition (SVD) methods. In some embodiments, the SEM image is labeled by a regression label, for example based on the size of a continuous pattern. The regression label can be used to establish a regression type prediction model. For example, a computer vision algorithm can be used to classify an area of interest (or a pattern of interest) as a defect or non-defect, or to measure the size of a continuous pattern. For example, the pattern of interest may be classified as a pattern interruption or a pattern bridge. In another example, the pattern of interest may also be measured by sizes distributed in different ranges, such as a 5-10 nm range.
在一些实施例中，训练模块1202的模型训练器1220可以基于训练数据和对应标记数据来训练预测模型1232。在一些实施例中，预测模型1232基于特征排名算法。例如，对可以与缺陷形成相关的各种特征或参数进行排名，从而可以选择与缺陷原因更相关的因素，并且可以去除噪声或无关变量，以进行更有效和高效的根本原因分析。在一些实施例中，预测模型1232基于模型特定算法或模型不可知算法。In some embodiments, the model trainer 1220 of the training module 1202 can train the prediction model 1232 based on the training data and the corresponding labeled data. In some embodiments, the prediction model 1232 is based on a feature ranking algorithm. For example, various features or parameters that may be related to defect formation are ranked so that factors that are more relevant to the cause of the defect can be selected, and noise or irrelevant variables can be removed to perform more effective and efficient root cause analysis. In some embodiments, the prediction model 1232 is based on a model-specific algorithm or a model-agnostic algorithm.
在一些实施例中，模型1232包括建立在多个随机决策树上的随机森林模型。随机森林模型可以对在训练数据的不同子集上训练的多个深度决策树进行平均，目的是减小方差。在一些实施例中，(由模型训练器1220)基于来自训练数据的多个随机选择的特征或参数来训练每个随机决策树，其中训练数据的相应特征可以被放置在树的节点处。在每个节点处，确定所选择的对应特征是否能够充分地对减小方差的总体目标做出贡献(例如，该特征是重要特征还是要去除的令人讨厌的特征)。因此，在训练期间，可以在相应决策树的每个节点处执行一系列评估，并且对从多个树获取的结果进行平均以确定特征重要性。可以计算杂质值来对特征进行排名。In some embodiments, model 1232 includes a random forest model built on multiple random decision trees. The random forest model can average multiple deep decision trees trained on different subsets of training data in order to reduce variance. In some embodiments, each random decision tree is trained (by model trainer 1220) based on multiple randomly selected features or parameters from training data, where the corresponding features of the training data can be placed at the nodes of the tree. At each node, it is determined whether the selected corresponding feature can fully contribute to the overall goal of reducing variance (e.g., whether the feature is an important feature or an annoying feature to be removed). Therefore, during training, a series of evaluations can be performed at each node of the corresponding decision tree, and the results obtained from multiple trees are averaged to determine feature importance. Impurity values can be calculated to rank features.
在一些实施例中，在收集包括图像数据1207、工艺数据1208和标记数据1206的训练数据之后，模型训练器1220确定用于拆分数据的最佳特征。然后，可以将所收集的数据拆分为包含最佳特征的值的子集。在一些实施例中，可以使用不同度量来量化地测量拆分质量。在一些实施例中，对于回归标记(例如，基于连续模式大小，并且用于构建回归型预测模型)，杂质值可以基于如下定义的均方误差来计算：In some embodiments, after collecting training data including image data 1207, process data 1208, and label data 1206, the model trainer 1220 determines the best features for splitting the data. The collected data can then be split into subsets containing the values of the best features. In some embodiments, different metrics can be used to quantitatively measure the quality of the split. In some embodiments, for regression labels (e.g., based on continuous pattern size and used to build a regression-type prediction model), the impurity value can be calculated based on the mean square error defined as follows:
或杂质值可以基于如下定义的绝对误差来计算：Or the impurity value can be calculated based on the absolute error defined as follows:
其中yi是实例的标记，N是实例的数目，μ是由下式确定的均值：whereyi is the label of the instance, N is the number of instances, and μ is the mean determined by:
在一些实施例中，对于用于构建分类型预测模型的分类标记，可以使用Gini杂质来计算杂质值：In some embodiments, for classification labels used to construct classification prediction models, Gini impurity can be used to calculate impurity values:
或可以使用熵来计算杂质值：Or you can use entropy to calculate the impurity value:
其中pi是节点处的标记i的频率，c是类的数目。wherepi is the frequency of label i at the node and c is the number of classes.
在一些实施例中，可以使用贪婪算法来选择特征的选择和用于放置特征的拆分点，以使杂质值最小化。例如，可以迭代地尝试出不同拆分点，并且选择决策树上提供最低杂质值的拆分点。此外，在相应决策树中的每个拆分处，拆分标准的改进是归因于拆分变量的重要性度量，并且对于每个变量单独地在随机森林模型中的所有决策树上累积。In some embodiments, a greedy algorithm can be used to select the selection of features and the split points for placing features to minimize the impurity value. For example, different split points can be iteratively tried out and the split point that provides the lowest impurity value on the decision tree is selected. In addition, at each split in the corresponding decision tree, the improvement of the splitting criterion is attributed to the importance measure of the splitting variable and accumulated across all decision trees in the random forest model for each variable separately.
在一些实施例中，在确定最佳拆分决策树的最佳拆分点处的最佳特征之后，模型训练器1220进一步使用基于最佳特征拆分的数据的子集来递归地生成新的树节点，直到可以获取优化的最佳精度和最小化的拆分次数。因此，模型训练器1220可以构建去相关的决策树的集合以获取平均结果，从而减小估计的预测函数的方差。在一些实施例中，在训练期间，随机森林中的每个决策树可以从数据点的随机样本中学习，并且一些样本可以在单个决策树中使用多次。在一些实施例中，仅从训练数据收集的所有特征的子集被考虑用于在相应决策树中拆分每个节点。In some embodiments, after determining the best features at the best split point of the best split decision tree, the model trainer 1220 further uses a subset of the data split based on the best features to recursively generate new tree nodes until the optimized best accuracy and the minimized number of splits can be obtained. Therefore, the model trainer 1220 can construct a collection of decorrelated decision trees to obtain average results, thereby reducing the variance of the estimated prediction function. In some embodiments, during training, each decision tree in the random forest can learn from a random sample of data points, and some samples can be used multiple times in a single decision tree. In some embodiments, only a subset of all features collected from the training data is considered for splitting each node in the corresponding decision tree.
在一些其他实施例中，预测模型1232基于用于缺陷预测的多项式回归模型。序列选择算法(诸如序列前向选择(SFS)或序列后向选择(SBS)算法)可以用于确定特征重要性。还可以计算每个特征的贡献的百分比。例如，SFS算法从一组空数据开始，每一步添加一个特征，该特征针对对应步骤的目标函数(诸如分类精度)提供最高值。重复该过程，直到添加了所需数目的特征。另一方面，SBS算法从一组完整的变量开始，一次删除一个特征，该特征的删除使目标性能下降最低。在一些实施例中，SFS和SBS也可以组合。在一些实施例中，预测模型1232可以包括任何其他合适的机器学习模型，诸如线性回归模型、逻辑回归模型、XGBoost模型等。In some other embodiments, the prediction model 1232 is based on a polynomial regression model for defect prediction. A sequence selection algorithm (such as a sequence forward selection (SFS) or a sequence backward selection (SBS) algorithm) can be used to determine feature importance. The percentage of contribution of each feature can also be calculated. For example, the SFS algorithm starts with a set of empty data and adds a feature at each step that provides the highest value for the objective function (such as classification accuracy) of the corresponding step. Repeat the process until the required number of features are added. On the other hand, the SBS algorithm starts with a complete set of variables and deletes one feature at a time, and the deletion of the feature minimizes the target performance. In some embodiments, SFS and SBS can also be combined. In some embodiments, the prediction model 1232 may include any other suitable machine learning model, such as a linear regression model, a logistic regression model, an XGBoost model, and the like.
在一些实施例中，预测模块1204包括数据获取器1240，数据获取器1240被配置为获取用于预测与晶片相关联的失败的根本原因的输入数据1242。在一些实施例中，输入数据1242包括从来自如图1所示的电子束工具104、控制器109或系统199的输入图像(诸如SEM图像)中提取的图像数据。在一些实施例中，输入图像可以对应于晶片上需要缺陷预测的感兴趣区域，诸如一个或多个接触孔、一个或多个线、管芯或整个晶片的区域。在一些实施例中，输入图像可以是在半导体处理期间的不同阶段拍摄的晶片的SEM图像，诸如在光刻中显影之后或者在蚀刻之后。图像数据可以包括与晶片上感兴趣区域中的特征或缺陷相关联的像素值、位置信息等。In some embodiments, the prediction module 1204 includes a data acquirer 1240, which is configured to acquire input data 1242 for predicting the root cause of failure associated with the wafer. In some embodiments, the input data 1242 includes image data extracted from an input image (such as a SEM image) from the electron beam tool 104, the controller 109, or the system 199 as shown in Figure 1. In some embodiments, the input image may correspond to an area of interest on the wafer where defect prediction is required, such as one or more contact holes, one or more lines, a die, or an area of the entire wafer. In some embodiments, the input image may be an SEM image of a wafer taken at different stages during semiconductor processing, such as after development in lithography or after etching. The image data may include pixel values, position information, etc. associated with features or defects in the area of interest on the wafer.
在一些实施例中，输入数据1242还包括与晶片的制造、检查或其他处理步骤相关的工艺数据。例如，工艺数据包括制造数据，诸如光刻参数、蚀刻参数、检查条件或其他制造过程。工艺数据还可以包括设计数据或可以与缺陷原因相关的其他因素。In some embodiments, the input data 1242 also includes process data related to the manufacture, inspection, or other processing steps of the wafer. For example, the process data includes manufacturing data, such as lithography parameters, etching parameters, inspection conditions, or other manufacturing processes. The process data may also include design data or other factors that may be related to the cause of the defect.
在一些实施例中，预测模块1204包括分析器1230，分析器1230被配置为使用模型1232来分析输入数据1242，模型1232包括如上所述的由模型训练器1220生成的多个决策树。在一些实施例中，通过对回归型模型的输入数据1242上的所有个体决策树的预测进行平均，或者在分类型模型的情况下通过从决策树的结果中获取多数票，可以进行对输入数据1241的预测。In some embodiments, the prediction module 1204 includes an analyzer 1230 configured to analyze the input data 1242 using a model 1232 including a plurality of decision trees generated by the model trainer 1220 as described above. In some embodiments, the prediction of the input data 1241 may be made by averaging the predictions of all individual decision trees on the input data 1242 for a regression type model, or by taking a majority vote from the results of the decision trees in the case of a classification type model.
在一些实施例中，在使用已训练模型1232拟合输入数据1242之后，分析器1230可以生成缺陷预测结果1250。例如，缺陷预测结果1250可以包括基于输入数据1242是否存在一个或多个缺陷、可以存在什么类型的缺陷、或者缺陷在晶片上的位置。在一些实施例中，缺陷预测结果1250可以用于使晶片上与缺陷的预测位置相对应的区域被测量或评估。例如，根据预测的缺陷类型或位置，控制器109和系统199的一个或多个组件可以控制电子束工具104扫描对应区域。在一些实施例中，分析器1230还可以生成特征排名结果1260。例如，可以标识与预测的重要特征(例如，处理步骤或参数)相对应的决策树的节点的子集，这些特征可以比其他特征对缺陷的形成有更显著的贡献。在一些实施例中，可以进一步输出可视化的排名结果以促进根本原因分析。In some embodiments, after fitting the input data 1242 using the trained model 1232, the analyzer 1230 can generate a defect prediction result 1250. For example, the defect prediction result 1250 may include whether one or more defects exist based on the input data 1242, what type of defects may exist, or the location of the defects on the wafer. In some embodiments, the defect prediction result 1250 can be used to cause an area on the wafer corresponding to the predicted location of the defect to be measured or evaluated. For example, based on the predicted defect type or location, the controller 109 and one or more components of the system 199 can control the electron beam tool 104 to scan the corresponding area. In some embodiments, the analyzer 1230 can also generate a feature ranking result 1260. For example, a subset of nodes of a decision tree corresponding to predicted important features (e.g., processing steps or parameters) can be identified, which can contribute more significantly to the formation of defects than other features. In some embodiments, a visualized ranking result can be further output to facilitate root cause analysis.
图13示出了根据本公开的一些实施例的根据特征排名结果1260而进行的特征重要性的可视化的示例。例如，图13中的条形图可视化了基于由预测模块1204分析的排名结果1260相应特征重要性，以便向用户提供对在半导体处理期间可能导致缺陷的相关工艺或参数的直接理解。FIG13 shows an example of visualization of feature importance according to some embodiments of the present disclosure based on the feature ranking results 1260. For example, the bar chart in FIG13 visualizes the corresponding feature importance based on the ranking results 1260 analyzed by the prediction module 1204, so as to provide the user with a direct understanding of the relevant processes or parameters that may cause defects during semiconductor processing.
图14是与本公开的一些实施例一致的表示用于执行自动根本原因分析的示例方法1400的过程流程图。在一些实施例中，一个或多个步骤由图12中的系统1200、图1中的控制器109或系统199、或图1中的系统100的一个或多个组件来执行。FIG14 is a process flow diagram representing an example method 1400 for performing automatic root cause analysis consistent with some embodiments of the present disclosure. In some embodiments, one or more steps are performed by one or more components of system 1200 in FIG12, controller 109 or system 199 in FIG1, or system 100 in FIG1.
在步骤1410中，可以(例如，通过数据获取器1240)获取与晶片的输入电子显微镜图像(例如，SEM图像)相关联的输入数据(例如，图12的输入数据1242)。在一些实施例中，输入数据包括从输入图像中提取的图像数据，诸如与晶片上感兴趣区域中的特征或缺陷相关联的像素值或位置信息等。在一些实施例中，输入数据还包括与加工晶片的制造、检查或其他步骤相关联的多个工艺特征。例如，工艺特征包括制造数据，诸如光刻参数、蚀刻参数、检查条件或其他制造过程。特征数据还可以包括设计数据或可以与缺陷原因相关的其他因素。In step 1410, input data (e.g., input data 1242 of FIG. 12 ) associated with an input electron microscope image (e.g., SEM image) of a wafer may be acquired (e.g., by data acquirer 1240). In some embodiments, the input data includes image data extracted from the input image, such as pixel values or position information associated with features or defects in a region of interest on the wafer. In some embodiments, the input data also includes a plurality of process features associated with manufacturing, inspection, or other steps of processing the wafer. For example, the process features include manufacturing data, such as lithography parameters, etching parameters, inspection conditions, or other manufacturing processes. The feature data may also include design data or other factors that may be associated with the cause of the defect.
在步骤1420中，通过将多个预训练的决策树模型应用于多个工艺特征，来标识多个过程特性中的一组工艺特征。在一些实施例中，模型训练器1220可以使用训练数据来训练多个预训练的决策树模型，训练数据包括由训练数据获取器1210获取的图像数据1207和工艺数据1208以及由标记数据获取器1205获取的标记数据1206，如图12中讨论的。在一些实施例中，多个预训练的决策树模型是随机森林模型、XGBoost模型或决策树分类模型的一部分。在一些实施例中，可以在训练数据的不同子集(例如，随机选择的工艺特征)上训练多个决策树模型，并且可以通过对多个决策树模型求平均来构造随机森林模型(例如，图12的模型1232)，目的是减小估计的预测函数的方差。在一些实施例中，当训练相应决策树时，可以使用不同度量来测量针对决策树的节点随机选择特征的拆分质量。例如，可以针对决策树的每个节点计算杂质值(如Gini杂质)，以选择一个特征作为相应节点的拆分点，从而导致Gini杂质的最大减少。在一些实施例中，针对多个预训练的决策树模型的特征子集是随机选择的，预训练决策树模型是去相关的，并且该组工艺特征是基于对将多个预训练的决策树模型应用于多个工艺特征而得到的结果进行平均来标识的。In step 1420, a set of process features in a plurality of process characteristics is identified by applying a plurality of pre-trained decision tree models to a plurality of process features. In some embodiments, the model trainer 1220 may train a plurality of pre-trained decision tree models using training data including image data 1207 and process data 1208 acquired by the training data acquirer 1210 and labeled data 1206 acquired by the labeled data acquirer 1205, as discussed in FIG. 12 . In some embodiments, the plurality of pre-trained decision tree models are part of a random forest model, an XGBoost model, or a decision tree classification model. In some embodiments, a plurality of decision tree models may be trained on different subsets of the training data (e.g., randomly selected process features), and a random forest model (e.g., model 1232 of FIG. 12 ) may be constructed by averaging a plurality of decision tree models in order to reduce the variance of the estimated prediction function. In some embodiments, when training the respective decision trees, different metrics may be used to measure the quality of the splitting of the randomly selected features for the nodes of the decision tree. For example, an impurity value (such as Gini impurity) can be calculated for each node of the decision tree to select a feature as a split point for the corresponding node, resulting in a maximum reduction in Gini impurity. In some embodiments, feature subsets for multiple pre-trained decision tree models are randomly selected, the pre-trained decision tree models are decorrelated, and the set of process features is identified based on averaging the results obtained by applying the multiple pre-trained decision tree models to the multiple process features.
在步骤1430中，可以输出该组工艺特征的排名结果(例如，图13中的条形图)。在一些实施例中，分析器1230可以输出特征排名结果1260，特征排名结果1260使用多个预训练的决策树模型(例如，针对随机森林模型)对与基于输入数据1242而预测的重要特征(例如，包括处理步骤或参数)相对应的一组工艺特征进行排名。In step 1430, a ranking result of the set of process features may be output (e.g., a bar chart in FIG. 13 ). In some embodiments, the analyzer 1230 may output a feature ranking result 1260 that ranks a set of process features corresponding to important features (e.g., including processing steps or parameters) predicted based on the input data 1242 using a plurality of pre-trained decision tree models (e.g., for a random forest model).
在一些实施例中，分析器1230还可以输出与一个或多个缺陷的类型或位置相关联的缺陷预测结果1250，该缺陷被预测为形成在与输入数据1242相关联的晶片上。在一些实施例中，与输入数据1242相对应的晶片区域的评估由图1中的系统100、控制器109或系统199的一个或多个组件来执行。例如，可以生成指令以使得本文中讨论的带电粒子束检查系统(例如，系统100或电子束工具104)根据缺陷预测结果1250来执行晶片区域的检查。在一些实施例中，检查可以在与缺陷的预测位置相对应的晶片的一个或多个区域上执行。在一些实施例中，检查还可以根据预测的缺陷类型使用一个或多个检查参数来执行。在一些实施例中，检查图像和对应检查结果可以用于进一步评估或改进模型1232。In some embodiments, the analyzer 1230 may also output a defect prediction result 1250 associated with the type or location of one or more defects that are predicted to be formed on a wafer associated with the input data 1242. In some embodiments, the evaluation of the wafer area corresponding to the input data 1242 is performed by one or more components of the system 100, the controller 109, or the system 199 in FIG. 1. For example, instructions may be generated so that the charged particle beam inspection system discussed herein (e.g., the system 100 or the electron beam tool 104) performs an inspection of the wafer area according to the defect prediction result 1250. In some embodiments, the inspection may be performed on one or more areas of the wafer corresponding to the predicted location of the defect. In some embodiments, the inspection may also be performed using one or more inspection parameters based on the predicted defect type. In some embodiments, the inspection image and the corresponding inspection result may be used to further evaluate or improve the model 1232.
可以提供一种非暂态计算机可读介质，该介质存储用于处理器(例如，控制器109、系统199、系统300或系统1200的处理器)执行在过程500、600、1100和1400中讨论的各种步骤等的指令。常见形式的非暂态介质包括例如软盘、柔性盘、硬盘、固态驱动器、磁带或任何其他磁性数据存储介质、光盘只读存储器(CD-ROM)、任何其他光学数据存储介质、具有孔图案的任何物理介质、随机存取存储器(RAM)、可编程只读存储器(PROM)、以及可擦除可编程只读存储器(EPROM)、FLASH-EPROM或任何其他闪存、非易失性随机存取存储器(NVRAM)、高速缓存、寄存器、任何其他存储器芯片或卡盒、以及其网络版本。A non-transitory computer readable medium may be provided that stores instructions for a processor (e.g., a processor of controller 109, system 199, system 300, or system 1200) to perform the various steps discussed in processes 500, 600, 1100, and 1400, etc. Common forms of non-transitory media include, for example, floppy disks, flexible disks, hard disks, solid-state drives, magnetic tapes, or any other magnetic data storage medium, compact disk read-only memory (CD-ROM), any other optical data storage medium, any physical medium with a pattern of holes, random access memory (RAM), programmable read-only memory (PROM), and erasable programmable read-only memory (EPROM), FLASH-EPROM or any other flash memory, non-volatile random access memory (NVRAM), caches, registers, any other memory chips or cartridges, and networked versions thereof.
可以使用以下条款进一步描述实施例：The embodiments may be further described using the following terms:
1.一种分析第一晶片上的第一区域的输入电子显微镜图像的方法，所述方法包括：1. A method for analyzing an input electron microscope image of a first area on a first wafer, the method comprising:
从所述输入电子显微镜图像，获取多个模式图像，所述多个模式图像与多个可解释模式相对应；From the input electron microscope image, acquiring a plurality of pattern images, the plurality of pattern images corresponding to a plurality of interpretable patterns;
评估所述多个模式图像；evaluating the plurality of pattern images;
基于评估结果，确定所述多个可解释模式对所述输入电子显微镜图像的贡献；以及Based on the evaluation results, determining contributions of the plurality of interpretable patterns to the input electron microscope image; and
基于所确定的所述贡献，预测所述第一晶片上的所述第一区域中的一个或多个特性。Based on the determined contributions, one or more characteristics in the first region on the first wafer are predicted.
2.根据条款1所述的方法，其中所述多个可解释模式中的相应可解释模式与所述第一晶片上的所述第一区域的特性相关联。2. The method of clause 1, wherein a respective interpretable pattern of the plurality of interpretable patterns is associated with a characteristic of the first region on the first wafer.
3.根据条款1至2中任一项所述的方法，其中获取所述多个模式图像包括：3. The method according to any one of clauses 1 to 2, wherein acquiring the plurality of pattern images comprises:
将所述输入电子显微镜图像分解为所述多个模式图像。The input electron microscope image is decomposed into the plurality of pattern images.
4.根据条款1至3中任一项所述的方法，其中获取所述多个模式图像包括：4. The method according to any one of clauses 1 to 3, wherein acquiring the plurality of pattern images comprises:
获取与所述输入电子显微镜图像相对应的分别与所述多个可解释模式相关联的系数。Obtain coefficients respectively associated with the plurality of interpretable modes corresponding to the input electron microscope image.
5.根据条款1至4中任一项所述的方法，其中所述一个或多个特性分别对应于一个或多个缺陷类别。5. The method according to any one of clauses 1 to 4, wherein the one or more characteristics correspond to one or more defect categories, respectively.
6.根据条款1至5中任一项所述的方法，其中所述一个或多个缺陷类别包括：小临界尺寸(CD)、沿着特定方向的偏移、椭圆度、模糊边缘、印刷接触孔、缺失接触孔、或桥接接触孔。6. A method according to any one of clauses 1 to 5, wherein the one or more defect categories include: small critical dimension (CD), deviation along a specific direction, ellipticity, fuzzy edges, printed contact holes, missing contact holes, or bridged contact holes.
7.根据条款1至6中任一项所述的方法，其中评估所述多个模式图像包括：7. A method according to any one of clauses 1 to 6, wherein evaluating the plurality of pattern images comprises:
将分类器模型应用于分别与所述多个可解释模式相关联的系数，以获取包括所述评估结果的输出。A classifier model is applied to the coefficients respectively associated with the plurality of interpretable patterns to obtain an output comprising the evaluation result.
8.根据条款1至7中任一项所述的方法，其中所述分类器模型是逻辑回归、支持向量机或神经网络模型。8. A method according to any one of clauses 1 to 7, wherein the classifier model is a logistic regression, a support vector machine or a neural network model.
9.根据条款1至8中任一项所述的方法，其中评估所述多个模式图像包括：9. A method according to any one of clauses 1 to 8, wherein evaluating the plurality of pattern images comprises:
获取所述评估结果，所述评估结果中的每个评估结果指示对应可解释模式存在的可能性。The evaluation results are obtained, each of which indicates a possibility that a corresponding interpretable pattern exists.
10.根据条款1至9中任一项所述的方法，其中确定所述多个可解释模式对所述输入电子显微镜图像的所述贡献包括：10. A method according to any one of clauses 1 to 9, wherein determining the contribution of the plurality of interpretable patterns to the input electron microscopy image comprises:
使用多项式回归模型来对分类器模型进行近似。The classifier model is approximated using a polynomial regression model.
11.根据条款1至10中任一项所述的方法，其中所述多项式回归模型包括线性模型。11. A method according to any one of clauses 1 to 10, wherein the polynomial regression model comprises a linear model.
12.根据条款1至11中任一项所述的方法，其中确定所述多个可解释模式对所述输入电子显微镜图像的所述贡献包括：12. A method according to any one of clauses 1 to 11, wherein determining the contribution of the plurality of interpretable patterns to the input electron microscopy image comprises:
根据使用线性模型的线性近似，分别确定与所述多个可解释模式相关联的权重。Weights associated with the plurality of interpretable patterns are respectively determined based on linear approximation using a linear model.
13.根据条款1至12中任一项所述的方法，还包括：13. The method according to any one of clauses 1 to 12, further comprising:
生成表示所述多个可解释模式对所述输入电子显微镜图像的所述贡献的可视化。A visualization is generated representing the contribution of the plurality of interpretable patterns to the input electron microscopy image.
14.根据条款1至13中任一项所述的方法，还包括：14. The method according to any one of clauses 1 to 13, further comprising:
根据所述晶片上的所述区域中的所述一个或多个特性，调节一个或多个工艺参数。One or more process parameters are adjusted based on the one or more characteristics in the region on the wafer.
15.根据条款1至14中任一项所述的方法，还包括：15. The method according to any one of clauses 1 to 14, further comprising:
基于所述多个可解释模式的所确定的所述贡献，确定缺陷原因。Based on the determined contributions of the plurality of interpretable patterns, a defect cause is determined.
16.根据条款1至15中任一项所述的方法，还包括：16. The method according to any one of clauses 1 to 15, further comprising:
基于以下项来训练所述分类器模型：(1)多个晶片的训练电子显微镜图像、以及(2)所述训练电子显微镜图像的标记数据，所述标记数据对应于与所述训练电子显微镜图像中的每个训练电子显微镜图像相关联的多个可解释模式的系数。The classifier model is trained based on: (1) training electron microscope images of a plurality of wafers, and (2) labeled data for the training electron microscope images, the labeled data corresponding to coefficients of a plurality of interpretable patterns associated with each of the training electron microscope images.
17.根据条款1至16中任一项所述的方法，其中所述输入电子显微镜图像是在第一阶段处已经在所述第一阶段之前被处理的所述第一晶片上的所述第一区域的扫描电子显微镜(SEM)图像，并且其中所述训练电子显微镜图像是在所述第一阶段之后的第二阶段处被处理的所述多个晶片的SEM图像。17. A method according to any one of clauses 1 to 16, wherein the input electron microscope image is a scanning electron microscope (SEM) image of the first area on the first wafer that has been processed before the first stage at a first stage, and wherein the training electron microscope image is an SEM image of the multiple wafers processed at a second stage after the first stage.
18.根据条款1至17中任一项所述的方法，其中所述训练电子显微镜图像中的至少一个训练电子显微镜图像对应于所述多个晶片中的第二晶片上的第二区域，所述第二区域不同于所述第一晶片上的所述第一区域。18. A method according to any one of clauses 1 to 17, wherein at least one of the training electron microscope images corresponds to a second region on a second wafer of the plurality of wafers, the second region being different from the first region on the first wafer.
19.根据条款1至18中任一项所述的方法，还包括：19. The method according to any one of clauses 1 to 18, further comprising:
获取所述第一晶片上的多个区域的多个输入电子显微镜图像；acquiring a plurality of input electron microscope images of a plurality of regions on the first wafer;
针对相应输入电子显微镜图像，确定多维向量，所述多维向量表征所述多个可解释模式、以及所述多个可解释模式对所述相应输入电子显微镜图像的相关联贡献；以及For a corresponding input electron microscopy image, determining a multidimensional vector, the multidimensional vector characterizing the plurality of interpretable patterns and their associated contributions to the corresponding input electron microscopy image; and
对与所述第一晶片的所述多个输入电子显微镜图像相对应的多个多维向量进行聚类。A plurality of multi-dimensional vectors corresponding to the plurality of input electron microscope images of the first wafer are clustered.
20.根据条款19所述的方法，还包括：20. The method according to clause 19, further comprising:
基于所述聚类的结果，确定与多个集群相关联的一个或多个缺陷。Based on the results of the clustering, one or more defects associated with the plurality of clusters are determined.
21.根据条款19至20中任一项所述的方法，还包括：21. The method according to any one of clauses 19 to 20, further comprising:
基于所述聚类的结果来确定失败原因。A failure cause is determined based on the clustering results.
22.根据条款21所述的方法，其中基于所述聚类的所述结果来确定失败原因还包括：22. The method of clause 21, wherein determining a failure cause based on the result of the clustering further comprises:
映射所述第一晶片上与向量的集群相对应的一组区域的位置；以及mapping the locations of a set of regions on the first wafer corresponding to clusters of vectors; and
基于所述第一晶片上的所述一组区域的所述位置、以及与所述集群相关联的所述缺陷，来确定失败原因。A cause of failure is determined based on the locations of the set of regions on the first wafer and the defects associated with the cluster.
23.根据条款19至22中任一项所述的方法，还包括：23. The method according to any one of clauses 19 to 22, further comprising:
接收对所述第一晶片的区域的用户选择；以及receiving a user selection of a region of the first wafer; and
生成在所述第一晶片上的所述区域中确定的缺陷的可视化。A visualization of defects identified in the region on the first wafer is generated.
24.根据条款19至23中任一项所述的方法，还包括：24. The method according to any one of clauses 19 to 23, further comprising:
接收对缺陷类型的用户选择；以及receiving a user selection of a defect type; and
生成所述第一晶片上被确定为具有所述缺陷类型的区域的分布的可视化。A visualization of the distribution of areas on the first wafer determined to have the defect type is generated.
25.根据条款19至24中任一项所述的方法，还包括：25. The method according to any one of clauses 19 to 24, further comprising:
获取一组中包括所述第一晶片在内的多个晶片上的多个区域的多个输入电子显微镜图像；acquiring a plurality of input electron microscope images of a plurality of regions on a plurality of wafers in a group including the first wafer;
针对相应输入电子显微镜图像，确定多维向量，所述多维向量表征所述多个可解释模式、以及所述多个可解释模式对所述相应输入电子显微镜图像的相关联贡献；For a corresponding input electron microscope image, determining a multidimensional vector, the multidimensional vector characterizing the plurality of interpretable patterns and the associated contributions of the plurality of interpretable patterns to the corresponding input electron microscope image;
对与所述组中的所述多个晶片的所述多个输入电子显微镜图像相对应的多个多维向量进行聚类；以及clustering a plurality of multidimensional vectors corresponding to the plurality of input electron microscope images of the plurality of wafers in the group; and
基于所述聚类的结果来确定失败原因。A failure cause is determined based on the clustering results.
26.根据条款25所述的方法，还包括：26. The method according to clause 25, further comprising:
基于所述聚类的所述结果，来确定与多个集群相关联的一个或多个缺陷。Based on the results of the clustering, one or more defects associated with a plurality of clusters are determined.
27.根据条款25至26中任一项所述的方法，还包括：27. The method according to any one of clauses 25 to 26, further comprising:
预测所述组中的第二晶片上的一个或多个缺陷。One or more defects are predicted on a second wafer in the set.
28.一种用于分析第一晶片上的第一区域的输入电子显微镜图像的装置，包括：28. An apparatus for analyzing an input electron microscope image of a first area on a first wafer, comprising:
存储器，存储指令集；以及a memory storing an instruction set; and
至少一个处理器，被配置为执行所述指令集，以使得所述装置执行：at least one processor configured to execute the set of instructions so that the apparatus performs:
从所述输入电子显微镜图像，获取多个模式图像，所述多个模式图像与多个可解释模式相对应；Acquire a plurality of pattern images from the input electron microscope image, the plurality of pattern images corresponding to a plurality of interpretable patterns;
评估所述多个模式图像；evaluating the plurality of pattern images;
基于评估结果，确定所述多个可解释模式对所述输入电子显微镜图像的贡献；以及Based on the evaluation results, determining contributions of the plurality of interpretable patterns to the input electron microscope image; and
基于所确定的所述贡献，预测所述第一晶片上的所述第一区域中的一个或多个特性。Based on the determined contributions, one or more characteristics in the first region on the first wafer are predicted.
29.根据条款28所述的装置，其中所述多个可解释模式中的相应可解释模式与所述第一晶片上的所述第一区域的特性相关联。29. The apparatus of clause 28, wherein a respective interpretable pattern of the plurality of interpretable patterns is associated with a characteristic of the first region on the first wafer.
30.根据条款28至29中任一项所述的装置，其中获取所述多个模式图像包括：30. The apparatus of any of clauses 28 to 29, wherein acquiring the plurality of pattern images comprises:
将所述输入电子显微镜图像分解为所述多个模式图像。The input electron microscope image is decomposed into the plurality of pattern images.
31.根据条款28至30中任一项所述的装置，其中获取所述多个模式图像包括：31. An apparatus according to any of clauses 28 to 30, wherein acquiring the plurality of pattern images comprises:
获取与所述输入电子显微镜图像相对应的分别与所述多个可解释模式相关联的系数。Obtain coefficients respectively associated with the plurality of interpretable modes corresponding to the input electron microscope image.
32.根据条款28至31中任一项所述的装置，其中所述一个或多个特性分别对应于一个或多个缺陷类别。32. Apparatus according to any of clauses 28 to 31, wherein the one or more characteristics correspond to one or more defect classes, respectively.
33.根据条款28至32中任一项所述的装置，其中所述一个或多个缺陷类别：包括小临界尺寸(CD)、沿着特定方向的偏移、椭圆度、模糊边缘、印刷接触孔、缺失接触孔、或桥接接触孔。33. A device according to any of clauses 28 to 32, wherein the one or more defect categories include small critical dimension (CD), deviation along a specific direction, ellipticity, fuzzy edges, printed contact holes, missing contact holes, or bridged contact holes.
34.根据条款28至33中任一项所述的装置，其中评估所述多个模式图像包括：34. An apparatus according to any of clauses 28 to 33, wherein evaluating the plurality of pattern images comprises:
将应用分类器模型应用于分别与所述多个可解释模式相关联的系数，以获取包括所述评估结果的输出。Applying a classifier model to the coefficients respectively associated with the plurality of interpretable patterns to obtain an output including the evaluation result.
35.根据条款28至34中任一项所述的装置，其中所述分类器模型是逻辑回归、支持向量机或神经网络模型。35. An apparatus according to any of clauses 28 to 34, wherein the classifier model is a logistic regression, support vector machine or neural network model.
36.根据条款28至35中任一项所述的装置，其中评估所述多个模式图像包括：36. An apparatus according to any of clauses 28 to 35, wherein evaluating the plurality of pattern images comprises:
获取所述评估结果，所述评估结果中的每个评估结果指示对应可解释模式存在的可能性。The evaluation results are obtained, each of which indicates the possibility of the existence of a corresponding interpretable pattern.
37.根据条款28至36中任一项所述的装置，其中确定所述多个可解释模式对所述输入电子显微镜图像的所述贡献包括：37. An apparatus according to any of clauses 28 to 36, wherein determining the contribution of the plurality of interpretable patterns to the input electron microscopy image comprises:
使用多项式回归模型来对分类器模型进行近似。The classifier model is approximated using a polynomial regression model.
38.根据条款28至37中任一项所述的装置，其中所述多项式回归模型包括线性模型。38. An apparatus according to any of clauses 28 to 37, wherein the polynomial regression model comprises a linear model.
39.根据条款28至38中任一项所述的装置，其中确定所述多个可解释模式对所述输入电子显微镜图像的所述贡献包括：39. An apparatus according to any of clauses 28 to 38, wherein determining the contribution of the plurality of interpretable patterns to the input electron microscopy image comprises:
根据使用所述线性模型的线性近似，分别确定与所述多个可解释模式相关联的权重。Weights associated with the plurality of interpretable patterns are respectively determined based on linear approximation using the linear model.
40.根据条款28至39中任一项所述的装置，其中所述至少一个处理器被配置为执行所述指令集以使得所述装置进一步执行：40. An apparatus according to any of clauses 28 to 39, wherein the at least one processor is configured to execute the set of instructions so that the apparatus further performs:
生成表示所述多个可解释模式对所述输入电子显微镜图像的所述贡献的可视化。A visualization is generated representing the contribution of the plurality of interpretable patterns to the input electron microscopy image.
41.根据条款28至40中任一项所述的装置，其中所述至少一个处理器被配置为执行所述指令集以使得所述装置进一步执行：41. An apparatus according to any of clauses 28 to 40, wherein the at least one processor is configured to execute the set of instructions so that the apparatus further performs:
根据所述晶片上的所述区域中的所述一个或多个特性，调节一个或多个工艺参数。One or more process parameters are adjusted based on the one or more characteristics in the region on the wafer.
42.根据条款28至41中任一项所述的装置，其中所述至少一个处理器被配置为执行所述指令集以使得所述装置进一步执行：42. An apparatus according to any of clauses 28 to 41, wherein the at least one processor is configured to execute the set of instructions so that the apparatus further performs:
基于所述多个可解释模式的所确定的所述贡献，确定缺陷原因。Based on the determined contributions of the plurality of interpretable patterns, a defect cause is determined.
43.根据条款28至42中任一项所述的装置，其中所述至少一个处理器被配置为执行所述指令集以使得所述装置进一步执行：43. An apparatus according to any of clauses 28 to 42, wherein the at least one processor is configured to execute the set of instructions so that the apparatus further performs:
基于以下项来训练所述分类器模型：(1)多个晶片的训练电子显微镜图像、以及(2)所述训练电子显微镜图像的标记数据，所述标记数据对应于与所述训练电子显微镜图像中的每个训练电子显微镜图像相关联的多个可解释模式的系数。The classifier model is trained based on: (1) training electron microscope images of a plurality of wafers, and (2) labeled data for the training electron microscope images, the labeled data corresponding to coefficients of a plurality of interpretable patterns associated with each of the training electron microscope images.
44.根据条款28至43中任一项所述的装置，其中所述输入电子显微镜图像是在第一阶段处已经在所述第一阶段之前被处理的所述第一晶片上的所述第一区域的扫描电子显微镜(SEM)图像，并且其中所述训练电子显微镜图像是在所述第一阶段之后的第二阶段被处理的所述多个晶片的SEM图像。44. An apparatus according to any one of clauses 28 to 43, wherein the input electron microscope image is a scanning electron microscope (SEM) image of the first area on the first wafer that has been processed at a first stage before the first stage, and wherein the training electron microscope image is an SEM image of the multiple wafers that are processed in a second stage after the first stage.
45.根据条款28至44中任一项所述的装置，其中所述训练电子显微镜图像中的至少一个训练电子显微镜图像对应于所述多个晶片中的第二晶片上的第二区域，所述第二区域不同于所述第一晶片上的所述第一区域。45. An apparatus according to any one of clauses 28 to 44, wherein at least one of the training electron microscope images corresponds to a second region on a second wafer of the plurality of wafers, the second region being different from the first region on the first wafer.
46.根据条款28至45中任一项所述的装置，其中所述至少一个处理器被配置为执行所述指令集以使得所述装置进一步执行：46. An apparatus according to any of clauses 28 to 45, wherein the at least one processor is configured to execute the set of instructions so that the apparatus further performs:
获取所述第一晶片上的多个区域的多个输入电子显微镜图像；acquiring a plurality of input electron microscope images of a plurality of regions on the first wafer;
针对相应输入电子显微镜图像，确定多维向量，所述多维向量表征所述多个可解释模式、以及所述多个可解释模式对所述相应输入电子显微镜图像的相关联贡献；以及For a corresponding input electron microscopy image, determining a multidimensional vector, the multidimensional vector characterizing the plurality of interpretable patterns and their associated contributions to the corresponding input electron microscopy image; and
对与所述第一晶片的所述多个输入电子显微镜图像相对应的多个多维向量进行聚类。A plurality of multi-dimensional vectors corresponding to the plurality of input electron microscope images of the first wafer are clustered.
47.根据条款46所述的装置，其中所述至少一个处理器被配置为执行所述指令集以使得所述装置进一步执行：47. An apparatus according to clause 46, wherein the at least one processor is configured to execute the set of instructions so that the apparatus further performs:
基于所述聚类的结果，确定与多个集群相关联的一个或多个缺陷。Based on the results of the clustering, one or more defects associated with the plurality of clusters are determined.
48.根据条款46至47中任一项所述的装置，其中所述至少一个处理器被配置为执行所述指令集以使得所述装置进一步执行：48. An apparatus according to any of clauses 46 to 47, wherein the at least one processor is configured to execute the set of instructions so that the apparatus further performs:
基于所述聚类的结果来确定失败原因。A failure cause is determined based on the clustering results.
49.根据条款48所述的装置，其中基于所述聚类的所述结果来确定失败原因还包括：49. The apparatus of clause 48, wherein determining a cause of failure based on the result of the clustering further comprises:
映射所述第一晶片上与向量的集群相对应的一组区域的位置；以及mapping the locations of a set of regions on the first wafer corresponding to clusters of vectors; and
基于所述第一晶片上的所述一组区域的所述位置、以及与所述集群相关联的所述缺陷，来确定失败原因。A cause of failure is determined based on the locations of the set of regions on the first wafer and the defects associated with the cluster.
50.根据条款46至49中任一项所述的装置，其中所述至少一个处理器被配置为执行所述指令集以使得所述装置进一步执行：50. An apparatus according to any of clauses 46 to 49, wherein the at least one processor is configured to execute the set of instructions so that the apparatus further performs:
接收对所述第一晶片的区域的用户选择；以及receiving a user selection of a region of the first wafer; and
生成在所述第一晶片上的所述区域中确定的缺陷的可视化。A visualization of defects identified in the region on the first wafer is generated.
51.根据条款46至50中任一项所述的装置，其中所述至少一个处理器被配置为执行所述指令集以使得所述装置进一步执行：51. An apparatus according to any of clauses 46 to 50, wherein the at least one processor is configured to execute the set of instructions so that the apparatus further performs:
接收对缺陷类型的用户选择；以及receiving a user selection of a defect type; and
生成所述第一晶片上被确定为具有所述缺陷类型的区域的分布的可视化。A visualization of the distribution of areas on the first wafer determined to have the defect type is generated.
52.根据条款46至51中任一项所述的装置，其中所述至少一个处理器被配置为执行所述指令集以使得所述装置进一步执行：52. An apparatus according to any of clauses 46 to 51, wherein the at least one processor is configured to execute the set of instructions so that the apparatus further performs:
获取一组中包括所述第一晶片在内的多个晶片上的多个区域的多个输入电子显微镜图像；acquiring a plurality of input electron microscope images of a plurality of regions on a plurality of wafers in a group including the first wafer;
针对相应输入电子显微镜图像，确定多维向量，所述多维向量表征所述多个可解释模式、以及所述多个可解释模式对所述相应输入电子显微镜图像的相关联贡献；For a corresponding input electron microscope image, determining a multidimensional vector, the multidimensional vector characterizing the plurality of interpretable patterns and the associated contributions of the plurality of interpretable patterns to the corresponding input electron microscope image;
对与所述组中的所述多个晶片的所述多个输入电子显微镜图像相对应的多个多维向量进行聚类；以及clustering a plurality of multidimensional vectors corresponding to the plurality of input electron microscope images of the plurality of wafers in the group; and
基于所述聚类的结果来确定失败原因。A failure cause is determined based on the clustering results.
53.根据条款52所述的装置，其中所述至少一个处理器被配置为执行所述指令集以使得所述装置进一步执行：53. An apparatus according to clause 52, wherein the at least one processor is configured to execute the set of instructions so that the apparatus further performs:
基于所述聚类的所述结果，来确定与多个集群相关联的一个或多个缺陷。Based on the results of the clustering, one or more defects associated with a plurality of clusters are determined.
54.根据条款52至53中任一项所述的装置，其中所述至少一个处理器被配置为执行所述指令集以使得所述装置进一步执行：54. An apparatus according to any of clauses 52 to 53, wherein the at least one processor is configured to execute the set of instructions so that the apparatus further performs:
预测所述组中的第二晶片上的一个或多个缺陷。One or more defects are predicted on a second wafer in the set.
55.一种存储指令集的非暂态计算机可读介质，所述指令集由计算设备的至少一个处理器可执行以使得所述计算设备执行分析第一晶片上的第一区域的输入电子显微镜图像的方法，所述方法包括：55. A non-transitory computer readable medium storing a set of instructions executable by at least one processor of a computing device to cause the computing device to perform a method of analyzing an input electron microscope image of a first area on a first wafer, the method comprising:
从所述输入电子显微镜图像，获取多个模式图像，所述多个模式图像与多个可解释模式相对应；Acquire a plurality of pattern images from the input electron microscope image, the plurality of pattern images corresponding to a plurality of interpretable patterns;
评估所述多个模式图像；evaluating the plurality of pattern images;
基于评估结果，确定所述多个可解释模式对所述输入电子显微镜图像的贡献；以及Based on the evaluation results, determining contributions of the plurality of interpretable patterns to the input electron microscope image; and
基于所确定的所述贡献，预测所述第一晶片上的所述第一区域中的一个或多个特性。Based on the determined contributions, one or more characteristics in the first region on the first wafer are predicted.
56.根据条款55所述的非暂态计算机可读介质，其中所述多个可解释模式中的相应可解释模式与所述第一晶片上的所述第一区域的特性相关联。56. The non-transitory computer-readable medium of clause 55, wherein a respective one of the plurality of interpretable patterns is associated with a characteristic of the first region on the first wafer.
57.根据条款55至56中任一项所述的非暂态计算机可读介质，其中获取所述多个模式图像包括：57. The non-transitory computer-readable medium of any one of clauses 55 to 56, wherein acquiring the plurality of pattern images comprises:
将所述输入电子显微镜图像分解为所述多个模式图像。The input electron microscope image is decomposed into the plurality of pattern images.
58.根据条款55至57中任一项所述的非暂态计算机可读介质，其中获取所述多个模式图像包括：58. The non-transitory computer-readable medium of any one of clauses 55 to 57, wherein acquiring the plurality of pattern images comprises:
获取与所述输入电子显微镜图像相对应的分别与所述多个可解释模式相关联的系数。Obtain coefficients respectively associated with the plurality of interpretable modes corresponding to the input electron microscope image.
59.根据条款55至58中任一项所述的非暂态计算机可读介质，其中所述一个或多个特性分别对应于一个或多个缺陷类别。59. The non-transitory computer-readable medium of any of clauses 55 to 58, wherein the one or more characteristics correspond to one or more defect categories, respectively.
60.根据条款55至59中任一项所述的非暂态计算机可读介质，其中所述一个或多个缺陷类别包括：小临界尺寸(CD)、沿着特定方向的偏移、椭圆度、模糊边缘、印刷接触孔、缺失接触孔、或桥接接触孔。60. A non-transitory computer readable medium according to any one of clauses 55 to 59, wherein the one or more defect categories include: small critical dimension (CD), deviation along a specific direction, ellipticity, fuzzy edges, printed contact holes, missing contact holes, or bridged contact holes.
61.根据条款55至60中任一项所述的非暂态计算机可读介质，其中评估所述多个模式图像包括：61. The non-transitory computer-readable medium of any one of clauses 55 to 60, wherein evaluating the plurality of pattern images comprises:
将分类器模型应用于分别与所述多个可解释模式相关联的系数，以获取包括所述评估结果的输出。A classifier model is applied to the coefficients respectively associated with the plurality of interpretable patterns to obtain an output comprising the evaluation result.
62.根据条款55至61中任一项所述的非暂态计算机可读介质，其中所述分类器模型是逻辑回归、支持向量机或神经网络模型。62. The non-transitory computer-readable medium of any one of clauses 55 to 61, wherein the classifier model is a logistic regression, a support vector machine, or a neural network model.
63.根据条款55至62中任一项所述的非暂态计算机可读介质，其中评估所述多个模式图像包括：63. The non-transitory computer-readable medium of any one of clauses 55 to 62, wherein evaluating the plurality of pattern images comprises:
获取所述评估结果，所述评估结果中的每个评估结果指示对应可解释模式存在的可能性。The evaluation results are obtained, each of which indicates the possibility of the existence of a corresponding interpretable pattern.
64.根据条款55至63中任一项所述的非暂态计算机可读介质，其中确定所述多个可解释模式对所述输入电子显微镜图像的所述贡献包括：64. The non-transitory computer-readable medium of any one of clauses 55 to 63, wherein determining the contribution of the plurality of interpretable patterns to the input electron microscope image comprises:
使用多项式回归模型来对分类器模型进行近似。The classifier model is approximated using a polynomial regression model.
65.根据条款55至64中任一项所述的非暂态计算机可读介质，其中所述多项式回归模型包括线性模型。65. The non-transitory computer-readable medium of any one of clauses 55 to 64, wherein the polynomial regression model comprises a linear model.
66.根据条款55至65中任一项所述的非暂态计算机可读介质，其中确定所述多个可解释模式对所述输入电子显微镜图像的所述贡献包括：66. The non-transitory computer-readable medium of any one of clauses 55 to 65, wherein determining the contribution of the plurality of interpretable patterns to the input electron microscope image comprises:
根据使用线性模型的线性近似，分别确定与所述多个可解释模式相关联的权重。Weights associated with the plurality of interpretable patterns are respectively determined based on linear approximation using a linear model.
67.根据条款55至66中任一项所述的非暂态计算机可读介质，其中所述指令集由所述计算设备的至少一个处理器可执行以使得所述计算设备进一步执行：67. The non-transitory computer-readable medium of any one of clauses 55 to 66, wherein the set of instructions is executable by at least one processor of the computing device to cause the computing device to further perform:
生成表示所述多个可解释模式对所述输入电子显微镜图像的所述贡献的可视化。A visualization is generated representing the contribution of the plurality of interpretable patterns to the input electron microscopy image.
68.根据条款55至67中任一项所述的非暂态计算机可读介质，其中所述指令集由所述计算设备的至少一个处理器可执行以使得所述计算设备进一步执行：68. The non-transitory computer-readable medium of any one of clauses 55 to 67, wherein the set of instructions is executable by at least one processor of the computing device to cause the computing device to further perform:
根据所述晶片上的所述区域中的所述一个或多个特性，调节一个或多个工艺参数。One or more process parameters are adjusted based on the one or more characteristics in the region on the wafer.
69.根据条款55至68中任一项所述的非暂态计算机可读介质，其中所述指令集由所述计算设备的至少一个处理器可执行以使得所述计算设备进一步执行：69. The non-transitory computer-readable medium of any one of clauses 55 to 68, wherein the set of instructions is executable by at least one processor of the computing device to cause the computing device to further perform:
基于所述多个可解释模式的所确定的所述贡献，确定缺陷原因。Based on the determined contributions of the plurality of interpretable patterns, a defect cause is determined.
70.根据条款55至69中任一项所述的非暂态计算机可读介质，其中所述指令集由所述计算设备的至少一个处理器可执行以使得所述计算设备进一步执行，所述指令集还包括：70. The non-transitory computer-readable medium of any one of clauses 55 to 69, wherein the set of instructions is executable by at least one processor of the computing device to cause the computing device to further perform:
基于以下项来训练所述分类器模型：(1)多个晶片的训练电子显微镜图像、以及(2)所述训练电子显微镜图像的标记数据，所述标记数据对应于与所述训练电子显微镜图像中的每个训练电子显微镜图像相关联的多个可解释模式的系数。The classifier model is trained based on: (1) training electron microscope images of a plurality of wafers, and (2) labeled data for the training electron microscope images, the labeled data corresponding to coefficients of a plurality of interpretable patterns associated with each of the training electron microscope images.
71.根据条款55至70中任一项所述的非暂态计算机可读介质，其中所述输入电子显微镜图像是在第一阶段处已经在所述第一阶段之前被处理的所述第一晶片上的所述第一区域的扫描电子显微镜(SEM)图像，并且其中所述训练电子显微镜图像是在所述第一阶段之后的第二阶段处被处理的所述多个晶片的SEM图像。71. A non-transitory computer-readable medium according to any one of clauses 55 to 70, wherein the input electron microscope image is a scanning electron microscope (SEM) image of the first area on the first wafer that has been processed before the first stage at a first stage, and wherein the training electron microscope image is an SEM image of the multiple wafers processed at a second stage after the first stage.
72.根据条款55至71中任一项所述的非暂态计算机可读介质，其中所述训练电子显微镜图像中的至少一个训练电子显微镜图像对应于所述多个晶片中的第二晶片上的第二区域，所述第二区域不同于所述第一晶片上的所述第一区域。72. A non-transitory computer-readable medium according to any one of clauses 55 to 71, wherein at least one of the training electron microscope images corresponds to a second region on a second wafer among the plurality of wafers, the second region being different from the first region on the first wafer.
73.根据条款55至72中任一项所述的非暂态计算机可读介质，其中所述指令集由所述计算设备的至少一个处理器可执行以使得所述计算设备进一步执行：73. The non-transitory computer-readable medium of any one of clauses 55 to 72, wherein the set of instructions is executable by at least one processor of the computing device to cause the computing device to further perform:
获取所述第一晶片上的多个区域的多个输入电子显微镜图像；acquiring a plurality of input electron microscope images of a plurality of regions on the first wafer;
针对相应输入电子显微镜图像，确定多维向量，所述多维向量表征所述多个可解释模式、以及所述多个可解释模式对所述相应输入电子显微镜图像的相关联贡献；以及For a corresponding input electron microscopy image, determining a multidimensional vector, the multidimensional vector characterizing the plurality of interpretable patterns and their associated contributions to the corresponding input electron microscopy image; and
对与所述第一晶片的所述多个输入电子显微镜图像相对应的多个多维向量进行聚类。A plurality of multi-dimensional vectors corresponding to the plurality of input electron microscope images of the first wafer are clustered.
74.根据条款73所述的非暂态计算机可读介质，其中所述指令集由所述计算设备的至少一个处理器可执行以使得所述计算设备进一步执行：74. The non-transitory computer-readable medium of clause 73, wherein the set of instructions is executable by at least one processor of the computing device to cause the computing device to further perform:
基于所述聚类的结果，确定与多个集群相关联的一个或多个缺陷。Based on the results of the clustering, one or more defects associated with the plurality of clusters are determined.
75.根据条款73至74中任一项所述的非暂态计算机可读介质，其中所述指令集由所述计算设备的至少一个处理器可执行以使得所述计算设备进一步执行：75. The non-transitory computer-readable medium of any of clauses 73 to 74, wherein the set of instructions is executable by at least one processor of the computing device to cause the computing device to further perform:
基于所述聚类的结果来确定失败原因。A failure cause is determined based on the clustering results.
76.根据条款75所述的非暂态计算机可读介质，其中基于所述聚类的所述结果来确定失败原因还包括：76. The non-transitory computer-readable medium of clause 75, wherein determining a failure cause based on the result of the clustering further comprises:
映射所述第一晶片上与向量集群相对应的一组区域的位置；以及mapping locations of a set of regions on the first wafer corresponding to vector clusters; and
基于所述第一晶片上的所述一组区域的所述位置、以及与所述集群相关联的所述缺陷，来确定失败原因。A cause of failure is determined based on the locations of the set of regions on the first wafer and the defects associated with the cluster.
77.根据条款73至76中任一项所述的非暂态计算机可读介质，其中所述指令集由所述计算设备的至少一个处理器可执行以使得所述计算设备进一步执行：77. The non-transitory computer-readable medium of any of clauses 73 to 76, wherein the set of instructions is executable by at least one processor of the computing device to cause the computing device to further perform:
接收对所述第一晶片的区域的用户选择；以及receiving a user selection of a region of the first wafer; and
生成在所述第一晶片上的所述区域中确定的缺陷的可视化。A visualization of defects identified in the region on the first wafer is generated.
78.根据条款73至77中任一项所述的非暂态计算机可读介质，其中所述指令集由所述计算设备的至少一个处理器可执行以使得所述计算设备进一步执行：78. The non-transitory computer-readable medium of any of clauses 73 to 77, wherein the set of instructions is executable by at least one processor of the computing device to cause the computing device to further perform:
接收对缺陷类型的用户选择；以及receiving a user selection of a defect type; and
生成所述第一晶片上被确定为具有所述缺陷类型的区域的分布的可视化。A visualization of the distribution of areas on the first wafer determined to have the defect type is generated.
79.根据条款73至78中任一项所述的非暂态计算机可读介质，其中所述指令集由所述计算设备的至少一个处理器可执行以使得所述计算设备进一步执行：79. The non-transitory computer-readable medium of any of clauses 73 to 78, wherein the set of instructions is executable by at least one processor of the computing device to cause the computing device to further perform:
获取一组中包括所述第一晶片在内的多个晶片上的多个区域的多个输入电子显微镜图像；acquiring a plurality of input electron microscope images of a plurality of regions on a plurality of wafers in a group including the first wafer;
针对相应输入电子显微镜图像，确定多维向量，所述多维向量表征所述多个可解释模式、以及所述多个可解释模式对所述相应输入电子显微镜图像的相关联贡献；For a corresponding input electron microscope image, determining a multidimensional vector, the multidimensional vector characterizing the plurality of interpretable patterns and the associated contributions of the plurality of interpretable patterns to the corresponding input electron microscope image;
对与所述组中的所述多个晶片的所述多个输入电子显微镜图像相对应的多个多维向量进行聚类；以及clustering a plurality of multidimensional vectors corresponding to the plurality of input electron microscope images of the plurality of wafers in the group; and
基于所述聚类的结果来确定失败原因。A failure cause is determined based on the clustering results.
80.根据条款79所述的非暂态计算机可读介质，其中所述指令集由所述计算设备的至少一个处理器可执行以使得所述计算设备进一步执行：80. The non-transitory computer-readable medium of clause 79, wherein the set of instructions is executable by at least one processor of the computing device to cause the computing device to further perform:
基于所述聚类的所述结果，来确定与多个集群相关联的一个或多个缺陷。Based on the results of the clustering, one or more defects associated with a plurality of clusters are determined.
81.根据条款79至80中任一项所述的非暂态计算机可读介质，其中所述指令集由所述计算设备的至少一个处理器可执行以使得所述计算设备进一步执行：81. The non-transitory computer-readable medium of any of clauses 79 to 80, wherein the set of instructions is executable by at least one processor of the computing device to cause the computing device to further perform:
预测所述组中的第二晶片上的一个或多个缺陷。One or more defects are predicted on a second wafer in the set.
82.一种训练用于对电子显微镜图像进行分类的分类器模型的方法，所述方法包括：82. A method for training a classifier model for classifying electron microscope images, the method comprising:
获取多个晶片的训练电子显微镜图像；acquiring training electron microscope images of a plurality of wafers;
获取所述训练电子显微镜图像的标记数据，所述标记数据指示与所述训练电子显微镜图像中的每个训练电子显微镜图像相关联的多个可解释模式；以及obtaining label data for the training electron microscopy images, the label data indicating a plurality of interpretable patterns associated with each of the training electron microscopy images; and
基于所述训练电子显微镜图像和所述标记数据来训练所述分类器模型。The classifier model is trained based on the training electron microscope images and the labeled data.
83.根据条款82所述的方法，其中所述多个可解释模式分别对应于多个缺陷类别。83. A method according to clause 82, wherein the multiple interpretable patterns correspond to multiple defect categories respectively.
84.根据条款82至83中任一项所述的方法，其中所述分类器模型是逻辑回归、支持向量机或神经网络模型。84. A method according to any one of clauses 82 to 83, wherein the classifier model is a logistic regression, a support vector machine or a neural network model.
85.根据条款82至84中任一项所述的方法，其中所述训练电子显微镜图像是所述多个晶片的扫描电子显微镜(SEM)图像。85. A method according to any one of clauses 82 to 84, wherein the training electron microscope image is a scanning electron microscope (SEM) image of the plurality of wafers.
86.根据条款82至85中任一项所述的方法，其中所述标记数据是通过包括主成分分析(PCA)或奇异值分解(SVD)在内的自动程序获取的。86. A method according to any one of clauses 82 to 85, wherein the labelled data is obtained by an automated procedure including principal component analysis (PCA) or singular value decomposition (SVD).
87.一种用于训练用于对电子显微镜图像进行分类的分类器模型的装置，所述装置包括：87. An apparatus for training a classifier model for classifying electron microscope images, the apparatus comprising:
存储器，存储指令集；以及a memory storing an instruction set; and
至少一个处理器，被配置为执行所述指令集以使得所述装置执行：at least one processor configured to execute the set of instructions so that the apparatus performs:
获取多个晶片的训练电子显微镜图像；acquiring training electron microscope images of a plurality of wafers;
获取所述训练电子显微镜图像的标记数据，所述标记数据指示与所述训练电子显微镜图像中的每个训练电子显微镜图像相关联的多个可解释模式；以及obtaining label data for the training electron microscopy images, the label data indicating a plurality of interpretable patterns associated with each of the training electron microscopy images; and
基于所述训练电子显微镜图像和所述标记数据来训练所述分类器模型。The classifier model is trained based on the training electron microscope images and the labeled data.
88.根据条款87所述的装置，其中所述多个可解释模式分别对应于多个缺陷类别。88. An apparatus according to clause 87, wherein the plurality of interpretable patterns correspond to a plurality of defect categories, respectively.
89.根据条款87至88中任一项所述的装置，其中所述分类器模型是逻辑回归、支持向量机或神经网络模型。89. An apparatus according to any one of clauses 87 to 88, wherein the classifier model is a logistic regression, support vector machine or neural network model.
90.根据条款87至89中任一项所述的装置，其中所述训练电子显微镜图像是所述多个晶片的扫描电子显微镜(SEM)图像。90. An apparatus according to any of clauses 87 to 89, wherein the training electron microscope image is a scanning electron microscope (SEM) image of the plurality of wafers.
91.根据条款87至90中任一项所述的装置，其中所述标记数据是通过包括主成分分析(PCA)或奇异值分解(SVD)在内的自动程序获取的。91. An apparatus according to any one of clauses 87 to 90, wherein the labelled data is obtained by an automated procedure including principal component analysis (PCA) or singular value decomposition (SVD).
92.一种存储指令集的非暂态计算机可读介质，所述指令集由计算设备的至少一个处理器可执行以使得所述计算设备执行训练用于对电子显微镜图像进行分类的分类器模型的方法，所述方法包括：92. A non-transitory computer readable medium storing a set of instructions executable by at least one processor of a computing device to cause the computing device to perform a method of training a classifier model for classifying electron microscopy images, the method comprising:
获取多个晶片的训练电子显微镜图像；acquiring training electron microscope images of a plurality of wafers;
获取所述训练电子显微镜图像的标记数据，所述标记数据指示与所述训练电子显微镜图像中的每个训练电子显微镜图像相关联的多个可解释模式；以及obtaining label data for the training electron microscopy images, the label data indicating a plurality of interpretable patterns associated with each of the training electron microscopy images; and
基于所述训练电子显微镜图像和所述标记数据来训练所述分类器模型。The classifier model is trained based on the training electron microscope images and the labeled data.
93.根据条款92所述的非暂态计算机可读介质，其中所述多个可解释模式分别对应于多个缺陷类别。93. The non-transitory computer-readable medium of clause 92, wherein the plurality of interpretable patterns correspond to a plurality of defect categories, respectively.
94.根据条款92至93中任一项所述的非暂态计算机可读介质，其中所述分类器模型是逻辑回归、支持向量机或神经网络模型。94. The non-transitory computer-readable medium of any one of clauses 92 to 93, wherein the classifier model is a logistic regression, a support vector machine, or a neural network model.
95.根据条款92至94中任一项所述的非暂态计算机可读介质，其中所述训练电子显微镜图像是所述多个晶片的扫描电子显微镜(SEM)图像。95. The non-transitory computer-readable medium of any one of clauses 92 to 94, wherein the training electron microscope image is a scanning electron microscope (SEM) image of the plurality of wafers.
96.根据条款92至95中任一项所述的非暂态计算机可读介质，其中所述标记数据是通过包括主成分分析(PCA)或奇异值分解(SVD)在内的自动程序获取的。96. A non-transitory computer-readable medium according to any one of clauses 92 to 95, wherein the labeling data is obtained by an automatic procedure including principal component analysis (PCA) or singular value decomposition (SVD).
97.一种用于基于晶片的输入电子显微镜图像进行自动根本原因分析的方法，所述方法包括：97. A method for automatic root cause analysis based on an input electron microscope image of a wafer, the method comprising:
获取与所述输入电子显微镜图像相关联的输入数据，所述输入数据包括所述晶片的多个工艺特征；acquiring input data associated with the input electron microscope image, the input data comprising a plurality of process features of the wafer;
通过将多个预训练的决策树模型应用于所述多个工艺特征，来从所述多个工艺特征中标识一组工艺特征；以及identifying a set of process features from the plurality of process features by applying a plurality of pre-trained decision tree models to the plurality of process features; and
输出所述一组工艺特征的排名结果。The ranking result of the set of process features is outputted.
98.根据条款97所述的方法，其中所述多个工艺特征包括与对所述晶片的处理相关联的光刻参数、蚀刻参数或检查参数。98. The method of clause 97, wherein the plurality of process characteristics comprises lithography parameters, etching parameters, or inspection parameters associated with processing of the wafer.
99.根据条款97至98中任一项所述的方法，还包括：99. The method according to any one of clauses 97 to 98, further comprising:
基于以下项来训练所述多个决策树模型：多个晶片的多个电子显微镜图像的图像数据、与所述多个晶片的处理相关联的工艺数据、以及指示与所述电子显微镜图像中的每个电子显微镜图像相关联的缺陷信息的标记数据。The plurality of decision tree models are trained based on image data of a plurality of electron microscope images of a plurality of wafers, process data associated with processing of the plurality of wafers, and labeling data indicative of defect information associated with each of the electron microscope images.
100.根据条款97至99中任一项所述的方法，其中所述多个预训练的决策树模型是随机森林模型、XGBoost模型或决策树分类模型的一部分。100. A method according to any one of clauses 97 to 99, wherein the plurality of pre-trained decision tree models are part of a random forest model, an XGBoost model, or a decision tree classification model.
101.根据条款97至100中任一项所述的方法，其中所述多个预训练的决策树模型是去相关的，并且所述一组工艺特征是基于对将所述多个预训练的决策树模型应用于所述多个工艺特征而得到的结果进行平均来标识的。101. A method according to any one of clauses 97 to 100, wherein the multiple pre-trained decision tree models are decorrelated and the set of process features are identified based on averaging the results obtained by applying the multiple pre-trained decision tree models to the multiple process features.
102.根据条款97至101中任一项所述的方法，还包括：输出与将在所述晶片上形成的一个或多个缺陷的类型或位置相关联的缺陷预测结果。102. The method of any one of clauses 97 to 101, further comprising: outputting a defect prediction result associated with a type or location of one or more defects to be formed on the wafer.
103.根据条款97至102中任一项所述的方法，还包括：使得检查系统检查所述晶片上与所输出的缺陷预测结果相对应的区域。103. The method of any one of clauses 97 to 102, further comprising causing an inspection system to inspect an area on the wafer corresponding to the output defect prediction result.
104.一种用于基于晶片的输入电子显微镜图像进行自动根本原因分析的装置，包括：104. An apparatus for performing automatic root cause analysis based on an input electron microscope image of a wafer, comprising:
存储器，存储指令集；以及a memory storing an instruction set; and
至少一个处理器，被配置为执行所述指令集以使得所述装置执行：at least one processor configured to execute the set of instructions so that the apparatus performs:
获取与所述输入电子显微镜图像相关联的输入数据，所述输入数据包括所述晶片的多个工艺特征；acquiring input data associated with the input electron microscope image, the input data comprising a plurality of process features of the wafer;
通过将多个预训练的决策树模型应用于所述多个工艺特征，来从所述多个工艺特征中标识一组工艺特征；以及identifying a set of process features from the plurality of process features by applying a plurality of pre-trained decision tree models to the plurality of process features; and
输出所述一组工艺特征的排名结果。The ranking result of the set of process features is outputted.
105.根据条款104所述的装置，其中所述多个工艺特征包括与对所述晶片的处理相关联的光刻参数、蚀刻参数或检查参数。105. The apparatus of clause 104, wherein the plurality of process characteristics comprises lithography parameters, etching parameters, or inspection parameters associated with processing of the wafer.
106.根据权利要求104至105中任一项所述的装置，其中所述至少一个处理器被配置为执行所述指令集以使得所述装置进一步执行：106. The apparatus of any one of claims 104 to 105, wherein the at least one processor is configured to execute the set of instructions so that the apparatus further performs:
基于以下项来训练所述多个决策树模型：多个晶片的多个电子显微镜图像的图像数据、与所述多个晶片的处理相关联的工艺数据、以及指示与所述电子显微镜图像中的每个电子显微镜图像相关联的缺陷信息的标记数据。The plurality of decision tree models are trained based on image data of a plurality of electron microscope images of a plurality of wafers, process data associated with processing of the plurality of wafers, and labeling data indicative of defect information associated with each of the electron microscope images.
107.根据条款104至106中任一项所述的装置，其中所述多个预训练的决策树模型是随机森林模型、XGBoost模型或决策树分类模型的一部分。107. An apparatus according to any one of clauses 104 to 106, wherein the plurality of pre-trained decision tree models are part of a random forest model, an XGBoost model, or a decision tree classification model.
108.根据条款104至107中任一项所述的装置，其中所述多个预训练的决策树模型是去相关的，并且所述一组工艺特征是基于对将所述多个预训练的决策树模型应用于所述多个工艺特征而得到的结果进行平均来标识的。108. An apparatus according to any one of clauses 104 to 107, wherein the plurality of pre-trained decision tree models are decorrelated and the set of process features are identified based on averaging the results obtained by applying the plurality of pre-trained decision tree models to the plurality of process features.
109.根据条款104至108中任一项所述的装置，其中所述至少一个处理器被配置为执行所述指令集以使得所述装置进一步执行：109. An apparatus as described in any of clauses 104 to 108, wherein the at least one processor is configured to execute the set of instructions so that the apparatus further performs:
输出与将在所述晶片上形成的一个或多个缺陷的类型或位置相关联的缺陷预测结果。Defect prediction results associated with the type or location of one or more defects to be formed on the wafer are output.
110.根据条款104至109中任一项所述的装置，其中所述至少一个处理器被配置为执行所述指令集以使得所述装置进一步执行：110. An apparatus as described in any of clauses 104 to 109, wherein the at least one processor is configured to execute the set of instructions so that the apparatus further performs:
使得检查系统检查所述晶片上与所输出的缺陷预测结果相对应的区域。The inspection system is enabled to inspect an area on the wafer corresponding to the output defect prediction result.
111.一种存储指令集的非暂态计算机可读介质，所述指令集由计算设备的至少一个处理器可执行以使得所述计算设备执行用于基于晶片的输入电子显微镜图像进行自动根本原因分析的方法，所述方法包括：111. A non-transitory computer readable medium storing a set of instructions executable by at least one processor of a computing device to cause the computing device to perform a method for automatic root cause analysis based on an input electron microscope image of a wafer, the method comprising:
获取与所述输入电子显微镜图像相关联的输入数据，所述输入数据包括所述晶片的多个工艺特征；acquiring input data associated with the input electron microscope image, the input data comprising a plurality of process features of the wafer;
通过将多个预训练的决策树模型应用于所述多个工艺特征，来从所述多个工艺特征中标识一组工艺特征；以及identifying a set of process features from the plurality of process features by applying a plurality of pre-trained decision tree models to the plurality of process features; and
输出所述一组工艺特征的排名结果。The ranking result of the set of process features is outputted.
112.根据条款111所述的非暂态计算机可读介质，其中所述多个工艺特征包括与对所述晶片的处理相关联的光刻参数、蚀刻参数或检查参数。112. The non-transitory computer-readable medium of clause 111, wherein the plurality of process characteristics comprises lithography parameters, etch parameters, or inspection parameters associated with processing of the wafer.
113.根据条款111至112中任一项所述的非暂态计算机可读介质，其中由所述计算设备的至少一个处理器可执行的所述指令集使得所述计算设备进一步执行：113. The non-transitory computer-readable medium of any of clauses 111 to 112, wherein the set of instructions executable by at least one processor of the computing device causes the computing device to further perform:
基于以下项来训练所述多个决策树模型：多个晶片的多个电子显微镜图像的图像数据、与所述多个晶片的处理相关联的工艺数据、以及指示与所述电子显微镜图像中的每个电子显微镜图像相关联的缺陷信息的标记数据。The plurality of decision tree models are trained based on image data of a plurality of electron microscope images of a plurality of wafers, process data associated with processing of the plurality of wafers, and labeling data indicative of defect information associated with each of the electron microscope images.
114.根据条款111至113中任一项所述的非暂态计算机可读介质，其中所述多个预训练的决策树模型是随机森林模型、XGBoost模型或决策树分类模型的一部分。114. A non-transitory computer-readable medium as described in any of clauses 111 to 113, wherein the plurality of pre-trained decision tree models are part of a random forest model, an XGBoost model, or a decision tree classification model.
115.根据条款111至114中任一项所述的非暂态计算机可读介质，其中所述多个预训练的决策树模型是去相关的，并且所述一组工艺特征是基于对将所述多个预训练的决策树模型应用于所述多个工艺特征而得到的结果进行平均来标识的。115. A non-transitory computer-readable medium as described in any one of clauses 111 to 114, wherein the multiple pre-trained decision tree models are decorrelated and the set of process features are identified based on averaging the results obtained by applying the multiple pre-trained decision tree models to the multiple process features.
116.根据条款111至115中任一项所述的非暂态计算机可读介质，其中由所述计算设备的至少一个处理器可执行的所述指令集使得所述计算设备进一步执行：116. The non-transitory computer-readable medium of any of clauses 111 to 115, wherein the set of instructions executable by at least one processor of the computing device causes the computing device to further perform:
输出与将在所述晶片上形成的一个或多个缺陷的类型或位置相关联的缺陷预测结果。Defect prediction results associated with the type or location of one or more defects to be formed on the wafer are output.
117.根据条款111至116中任一项所述的非暂态计算机可读介质，其中由所述计算设备的至少一个处理器可执行的所述指令集使得所述计算设备进一步执行：117. The non-transitory computer-readable medium of any of clauses 111 to 116, wherein the set of instructions executable by at least one processor of the computing device causes the computing device to further perform:
使得检查系统检查所述晶片上与所输出的缺陷预测结果相对应的区域。The inspection system is enabled to inspect an area on the wafer corresponding to the output defect prediction result.
应当理解，本公开的实施例不限于上文所述和附图中所示的确切结构，并且在不偏离其范围的情况下可以进行各种修改和改变。已经结合各种实施例描述了本公开，通过考虑本文中公开的本发明的说明书和实践，本发明的其他实施例对于本领域技术人员将是很清楚的。本说明书和示例仅被认为是示例性的，本发明的真正范围和精神由以下权利要求指示。It should be understood that the embodiments of the present disclosure are not limited to the exact structures described above and shown in the accompanying drawings, and various modifications and changes may be made without departing from the scope thereof. The present disclosure has been described in conjunction with various embodiments, and other embodiments of the present invention will be clear to those skilled in the art by considering the specification and practice of the invention disclosed herein. This specification and examples are to be considered exemplary only, and the true scope and spirit of the present invention are indicated by the following claims.
以上描述旨在说明，而非限制。因此，对于本领域技术人员来说很清楚的是，在不脱离下面所陈述的权利要求的范围的情况下，可以如所描述的那样进行修改。The above description is intended to be illustrative, not limiting. It will therefore be apparent to one skilled in the art that modifications may be made as described without departing from the scope of the claims set forth below.
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| US63/128,764 | 2020-12-21 | ||
| US202163159389P | 2021-03-10 | 2021-03-10 | |
| US63/159,389 | 2021-03-10 | ||
| PCT/EP2021/084913WO2022135948A1 (en) | 2020-12-21 | 2021-12-09 | Data-driven prediction and identification of failure modes based on wafer-level analysis and root cause analysis for semiconductor processing |
| Publication Number | Publication Date |
|---|---|
| CN116648722Atrue CN116648722A (en) | 2023-08-25 |
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202180086259.0APendingCN116648722A (en) | 2020-12-21 | 2021-12-09 | Data driven failure mode prediction and identification based on wafer level analysis and root cause analysis of semiconductor processing |
| Country | Link |
|---|---|
| CN (1) | CN116648722A (en) |
| Publication | Publication Date | Title |
|---|---|---|
| TWI769371B (en) | Inspection methods and systems for semiconductor devices, and non-transitory computer readable medium of the same | |
| KR102826986B1 (en) | Automatic optimization of an examination recipe | |
| TWI708301B (en) | Feature selection and automated process window monitoring through outlier detection | |
| TWI706376B (en) | Systems, methods and non-transitory computer-readable storage media for defect detection | |
| US7570800B2 (en) | Methods and systems for binning defects detected on a specimen | |
| US20240062356A1 (en) | Data-driven prediction and identification of failure modes based on wafer-level analysis and root cause analysis for semiconductor processing | |
| TWI872117B (en) | System and method for generating predictive images for wafer inspection using machine learning | |
| JP2022507543A (en) | Use of deep learning defect detection and classification methods for pixel-level image quantification | |
| TWI683103B (en) | Determining one or more characteristics of a pattern of interest on a specimen | |
| US8260034B2 (en) | Multi-modal data analysis for defect identification | |
| US20240212131A1 (en) | Improved charged particle image inspection | |
| CN117522779B (en) | Method and system for inspecting a sample | |
| CN116364569A (en) | Defect detection using computationally efficient segmentation methods | |
| US20230139085A1 (en) | Processing reference data for wafer inspection | |
| TWI853692B (en) | Methods and systems for grouping a plurality of patterns and related non-transitory computer readable medium | |
| TWI895570B (en) | Apparatus for analyzing an input electron microscope image of an area on a wafer | |
| CN116648722A (en) | Data driven failure mode prediction and identification based on wafer level analysis and root cause analysis of semiconductor processing | |
| US20250166166A1 (en) | Systems and methods for defect location binning in charged-particle systems | |
| WO2025011912A1 (en) | Systems and methods for defect inspection in charged-particle systems | |
| TW202509985A (en) | Method for efficient dynamic sampling plan generation and accurate probe die loss projection | |
| CN117957496A (en) | System and method for inspection in charged particle systems by classification and identification of fault mechanisms |
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination |