Method and system for generating a training set that can be used for inspection of semiconductor samples

Abstract

A system and method for generating a training set that can be used to inspect semiconductor samples is provided. The method includes: obtaining a simulation model that can simulate the effect of a physical process on a fabrication process (FP) image according to a parameter value of a physical process; applying the simulation model to an image to be enhanced for the training set, thereby generating one or more enhanced images corresponding to one or more different values ​​of the parameter of the physical process; and including the generated one or more enhanced images in the training set. The training set can be used for sample inspection, automatic defect review, automatic defect classification, automatic navigation during inspection, automatic segmentation of FP images, automatic metrology based on FP images, and other inspection processes including machine learning using a trained deep neural network.

Description

Technical Field

[0001] The presently disclosed subject matter relates generally to the field of specimen inspection and, more particularly, to methods and systems for automated specimen inspection. Background Art

[0002] Current demands for high density and performance associated with ultra-large scale integration of manufactured devices require sub-micron features, increased transistor and circuit speeds, and improved reliability. Such demands require that device features be formed with high precision and uniformity, which in turn requires careful monitoring of the manufacturing process, including automated inspection of the device while it is still in the form of a semiconductor wafer.

[0003] The term "sample" used in this specification should be broadly interpreted to cover any kind of wafers, masks and other structures used to manufacture semiconductor integrated circuits, combinations and / or parts thereof, magnetic heads, flat panel displays and other semiconductor articles.

[0004] The term "inspection" used in this specification should be broadly interpreted to cover any kind of metrology-related operations and operations related to the detection and / or classification of defects in samples during sample manufacturing. Inspection is provided by using a non-destructive inspection tool during or after the manufacture of the sample to be inspected. By non-limiting example, the inspection process may include runtime scanning (in the form of a single scan or multiple scans), sampling, review, measurement, classification and / or other operations provided for a sample or a portion of a sample using the same or different inspection tools. Similarly, inspection can be provided before manufacturing the sample to be inspected, and may include, for example, generating (multiple) inspection plans and / or other setup operations. It will be noted that, unless otherwise specifically stated, "inspection" or its derivative terms used in this specification are not limited to the resolution or size of the inspection area. By non-limiting example, various non-destructive inspection tools include scanning electron microscopes, atomic force microscopes, optical inspection tools, etc.

[0005] By way of non-limiting example, the inspection at runtime may employ a two-stage process, e.g., inspecting a sample, followed by reviewing the sampled locations for potential defects. During the first stage, the sample surface is inspected at high speed and relatively low resolution. In the first stage, a defect map is generated to display suspect locations on the sample with a high probability of defects. During the second stage, at least a portion of the suspect locations are analyzed in more detail using relatively high resolution. In some cases, both stages may be implemented by the same inspection tool, and in some other cases, the two stages are implemented by different inspection tools.

[0006] Inspection processes can be used at various steps during semiconductor manufacturing to detect and classify defects on samples. The efficiency of inspection can be increased by automation of (multiple) processes, such as Automatic Defect Classification (ADC), Automatic Defect Review (ADR), etc. Summary of the invention

[0007] According to certain aspects of the presently disclosed subject matter, a method for generating a training set that can be used for semiconductor sample inspection is provided. The method includes: obtaining a simulation model capable of simulating the effect of a physical process on a fabrication process (FP) image (e.g., a SEM image but not limited thereto), wherein the simulation may depend on a parameter value of the physical process; applying the simulation model to an image to be enhanced for the training set, thereby generating one or more enhanced images corresponding to one or more different values ​​of the parameter of the physical process; and including the generated one or more enhanced images in the training set.

[0008] The training set can be used for sample inspection using the trained deep neural network, automatic defect review, automatic defect classification, automatic navigation during inspection, automatic segmentation of FP images, automatic metrology based on FP images, and other inspection processes involving machine learning.

[0009] The images to be augmented can be "real world" images, synthetic images, previously augmented images, or other images available in the training set.

[0010] According to further aspects of the presently disclosed subject matter, the simulation model can be a parametric model or an estimated model of various physical effects. By way of non-limiting example, the effects of the physical process can be noise, gray level (GL) calibration between tools, focus error, color variation, charging, vibration, quad detector miscalibration, etc.

[0011] According to further aspects of the presently disclosed subject matter, the generated one or more enhanced images can be included in a training set in association with ground truth data.

[0012] According to further aspects of the presently disclosed subject matter, one or more enhanced images can be generated by applying the simulation model together with other data (eg, CAD data, reference data, etc.).

[0013] According to other aspects of the presently disclosed subject matter, a system useful for inspection of semiconductor samples is provided that is configured to operate according to the above-described method.

[0014] According to other aspects of the presently disclosed subject matter, a non-transitory computer-readable medium is provided, comprising instructions that, when executed by a computer, cause the computer to perform the above-described method.

[0015] One of the advantages of certain embodiments of the presently disclosed subject matter is the enhanced strength of individual detection processes trained with a training set generated as disclosed herein. BRIEF DESCRIPTION OF THE DRAWINGS

[0016] In order to understand the invention and to appreciate how it may be implemented in practice, embodiments will now be described, by way of non-limiting examples only, with reference to the accompanying drawings, in which:

[0017] Figure 1 shows a generalized block diagram of an inspection system according to certain implementations of the presently disclosed subject matter;

[0018] Figure 2 shows a generalized flow chart for using a deep neural network (DNN) to automatically determine inspection-related data based on a manufacturing process (FP) image, according to certain embodiments of the presently disclosed subject matter;

[0019] Figure 3 shows a generalized functional block diagram of a training set generator constructed in accordance with certain embodiments of the presently disclosed subject matter; and

[0020] Figure 4 A generalized flow chart for generating a training set using augmented images according to certain implementations of the presently disclosed subject matter is shown. DETAILED DESCRIPTION

[0021] In the following detailed description, numerous specific details are set forth in order to provide a thorough understanding of the present invention. However, those skilled in the art will appreciate that the presently disclosed subject matter may be practiced without these specific details. In other cases, well-known methods, processes, components, and circuits are not described in detail in order not to unnecessarily obscure the presently disclosed subject matter.

[0022] Unless otherwise specifically stated, it will be apparent from the following discussion that it will be understood that throughout the specification, discussions utilizing terms such as "processing," "computing," "representing," "comparing," "generating," "training," "segmenting," "aligning," or similar terms refer to computer actions and / or processes that manipulate and / or transform data into other data, the data being represented as physical quantities (such as electrical quantities) and / or the data representing physical objects. The term "computer" should be broadly interpreted to cover any kind of hardware-based electronic device having data processing capabilities, including, by way of non-limiting example, a Fabrication Process Examination Information (FPEI) system and corresponding parts thereof.

[0023] The terms "non-transitory memory" and "non-transitory storage medium" as used herein should be broadly interpreted to cover any volatile or non-volatile computer memory suitable for the presently disclosed subject matter.

[0024] The term "defect" as used in this specification should be interpreted broadly to cover any kind of abnormal or undesirable feature formed on or in a sample.

[0025] The term "design data" as used in this specification should be interpreted broadly to cover any data indicative of a hierarchical physical design (layout) of a sample. The design data may be provided by a respective designer and / or may be derived from the physical design (e.g., through complex simulations, simple geometry and Boolean operations, etc.). The design data may be provided in different formats, such as, by way of non-limiting example, GDSII format, OASIS format, etc. The design data may be presented in a vector format, a grayscale intensity image format, or in other ways.

[0026] It will be appreciated that, unless otherwise specifically stated, certain features of the present subject matter described in the context of separate embodiments may also be provided in combination in a single embodiment. Conversely, various features of the present subject matter described in the context of a single embodiment may also be provided individually or in any suitable sub-combination. In the following detailed description, numerous specific details are set forth in order to provide a thorough understanding of the method and apparatus.

[0027] With this in mind, note that Figure 1 , Figure 1 A functional block diagram of an inspection system according to certain implementations of the presently disclosed subject matter is shown. Figure 1The inspection system 100 shown in the figure can be used for inspection of a sample (e.g., a wafer and / or a portion of a wafer) as part of a sample manufacturing process. The inspection system 100 shown includes a computer-based system 103 that can use images obtained during sample manufacturing (hereinafter referred to as manufacturing process (FP) images) to automatically determine metrology-related and / or defect-related information. The system 103 is hereinafter referred to as a manufacturing process inspection information (FPEI) system. The FPEI system 103 can be operably connected to one or more low-resolution inspection tools 101 and / or one or more high-resolution inspection tools 102 and / or other inspection tools. The inspection tool is constructed to capture FP images and / or review (multiple) captured FP images and / or implement or provide measurements related to (multiple) captured images. The FPEI system can be further operably connected to a CAD server 110 and a data repository 109.

[0028] The FPEI system 103 includes a processor and memory circuitry (PMC) 104 operably connected to a hardware-based input interface 105 and a hardware-based output interface 106. The PMC 104 is configured to provide all processing required for operating the FPEI system, as described in detail in the reference Figure 2-4 As described in more detail, and includes a processor (not shown separately) and a memory (not shown separately). The processor of PMC 104 may be constructed to execute several functional modules according to computer-readable instructions implemented on a non-transitory computer-readable memory contained in the PMC. Such functional modules are hereinafter referred to as being contained in the PMC. The functional modules contained in PMC 104 include an operably connected training set generator 111. The functional modules may further include a deep neural network (DNN) 112. DNN 112 includes a DNN module 114, which is constructed to implement data processing using (multiple) deep neural networks for outputting application-related data based on manufacturing (FP) input data. Optionally, DNN 112 may include a front DNN module 113 and / or a rear DNN module 115, the front DNN module 113 being constructed to provide processing before sending the input data to the DNN module, and the rear DNN module 115 being constructed to provide post-processed data generated by the DNN module. Reference will be made to Figure 2-4 The operation of the FPEI system 103, and PMC 104 and the functional blocks therein are described in more detail.

[0029] The DNN module 114 may include a supervised or unsupervised DNN 112, which includes multiple layers organized according to a corresponding DNN architecture. By way of non-limiting example, the multiple layers of the DNN may be organized according to a convolutional neural network (CNN) architecture, a recurrent neural network architecture, a recursive neural network architecture, a generative adversarial network (GAN) architecture, etc. Optionally, at least some of the layers may be organized in multiple DNN subnets. Each layer of the DNN module 114 may include multiple basic computational elements (CEs), which are generally referred to in the art as dimensions, neurons, or nodes.

[0030] Typically, the computational elements of a given layer can be connected to the CE of the previous layer and / or the next layer. Each connection between the CE of the previous layer and the CE of the next layer is associated with a weighted value. A given CE can receive input from the CE of the previous layer via each connection, and each given connection is associated with a weighted value of the input that can be applied to the given connection. The weighted value can determine the relative strength of the connection, thereby determining the relative influence of each input on the output of a given CE. A given CE can be constructed to calculate an activation value (e.g., a weighted sum of inputs), and further derive an output by applying an activation function to the calculated activation. The activation function can be, for example, an identity function, a deterministic function (e.g., a linear function, a sigmoid function, a threshold function or the like), a random function or other suitable function. The output from a given CE can be transmitted to the CE of a subsequent layer via their respective connections. Similarly, as described above, each connection at the output of a CE can be associated with a weighted value, which can be applied to the output of the CE before the output of the CE is received as an input to the CE of a subsequent layer. In addition to the weighted value, there can also be a threshold (including a limit function) associated with the connection and the CE.

[0031] The weight values ​​and / or thresholds of the deep neural network can be initially selected before training, and can be further iteratively adjusted or modified during training to achieve an optimal set of weight values ​​and / or thresholds in the trained DNN. After each iteration, the difference between the actual output produced by the DNN module and the target output associated with the corresponding data training set can be determined. This difference can be called an error value. When the cost function indicating the error value is less than a predetermined value, or when the performance between iterations achieves limited changes, it can be determined that the training has been completed. Optionally, at least a portion of the DNN subnet (if any) can be trained separately before training the entire DNN.

[0032] The DNN input dataset used to adjust the weights / thresholds of the deep neural network is referred to as the training set hereinafter.

[0033] The input of DNN 112 may be pre-processed by a pre-DNN module 113 before being input to DNN module 114, and / or the output of DNN module 114 may be post-processed by a post-DNN module 115 before being output from DNN 112. In this case, the training of DNN 112 also includes determining parameters of the pre-DNN module and / or the post-DNN module. Optionally, the DNN module may be trained so as to minimize the cost function of the entire DNN, while the parameters of the pre-DNN module and / or the post-DNN module may be pre-defined and adjusted during training. A set of training-based parameters that characterize DNN 112 at training time may further include parameters related to pre-DNN and post-DNN processing.

[0034] Note that the teachings of the presently disclosed subject matter are not constrained by the DNN 112 architecture.

[0035] As an alternative or in addition to the DNN 112, the PMC may include an Automatic Defect Review Module (ADR) and / or an Automatic Defect Classification Module (ADC) and / or other machine learning-based inspection modules that can be used for semiconductor sample inspection after training. Optionally, one or more inspection modules may include the DNN 112. Optionally, the DNN 112 may be shared between the inspection modules, or, alternatively, each of the one or more inspection modules may include its own DNN 112.

[0036] If you will refer to Figure 2-4 As described in further detail, the FPEI system is constructed to receive FP input data via an input interface 105. The FP input data may include data generated by an inspection tool (and / or its derivative data and / or metadata associated therewith) and / or data stored in one or more data repositories 109 and / or a CAD server 110 and / or another related data repository. Note that the FP input data may include images (e.g., captured images, images derived from captured images, simulated images, synthetic images, etc.) and associated digital data (e.g., metadata, handcrafted attributes, etc.). It should also be noted that the image data may include data related to a layer of interest and / or one or more other layers of the sample. Optionally, for training purposes, the FP input data may include all available FAB data or a portion of the available FAB data selected according to certain criteria.

[0037] The FPEI system is also configured to process at least a portion of the received FP input data and send the results (or a portion thereof) to a storage system 107, (multiple) inspection tools, a computer-based graphical user interface (GUI) 108 for presenting the results, and / or an external system (e.g., a yield management system (YMS) of the FAB) via an output interface 106. The GUI 108 may also be configured to enable user-specified input related to operating the FPEI system 103.

[0038] By way of non-limiting example, the sample may be inspected by one or more low-resolution inspection tools 101 (e.g., an optical inspection system, a low-resolution SEM, etc.). Resulting data providing low-resolution image information of the sample (hereinafter referred to as low-resolution image data 121) may be transmitted directly or via one or more intermediate systems to the FPEI system 103. Alternatively or additionally, the sample may be inspected by a high-resolution inspection tool 102 (e.g., a subset of potential defect locations selected for review may be reviewed by a scanning electron microscope (SEM) or an atomic force microscope (AFM). Resulting data providing high-resolution image information of the sample (hereinafter referred to as high-resolution image data 122) may be transmitted directly or via one or more intermediate systems to the FPEI system 103.

[0039] Note that image data may be received and processed along with associated metadata (eg, pixel size, text description of defect type, parameters of the image capture process, etc.).

[0040] When processing FP input data (e.g., low-resolution image data and / or high-resolution image data, optionally together with other data as (e.g.) design data, synthesis data, etc.), the FPEI system can send the results (e.g., instruction-related data 123 and / or 124) to any (multiple) inspection tools, store the results (e.g., defect attributes, defect classification, etc.) in the storage system 107, present the results via the GUI 108 and / or send them to an external system (e.g., YMS).

[0041] Those skilled in the art will readily appreciate that the teachings of the presently disclosed subject matter are not limited to Figure 1 The constraints of the system shown in ; equivalent and / or modified functionality may be combined or divided in another manner and may be implemented in any suitable combination of software and firmware and / or hardware.

[0042] Without limiting the scope of the present disclosure in any way, it should also be noted that the inspection tool can be implemented as various types of inspection machines, such as optical imaging machines, electron beam inspection machines, etc. In some cases, the same inspection tool can provide low-resolution image data and high-resolution image data. In some cases, at least one inspection tool can have metrology capabilities.

[0043] Will note, Figure 1 The inspection system shown in can be implemented in a distributed computing environment, where Figure 1 The aforementioned functional modules shown in can be distributed on several local and / or remote devices and can be linked through a communication network. It should also be noted that in other embodiments, at least a portion of the inspection tool 101 and / or 102, the data repository 109, the storage system 107 and / or the GUI 108 can be outside the inspection system 100 and perform data communication operations with the FPEI system 103 via the input interface 105 and the output interface 106. The FPEI system 103 can be implemented as an independent (multiple) computer to be used in combination with the inspection tool. Alternatively, the corresponding functions of the FPEI system can be at least partially integrated with one or more inspection tools.

[0044] For illustrative purposes only, the following description is provided for inspecting semiconductor samples using a trained DNN. Those skilled in the art will readily appreciate that the teachings of the presently disclosed subject matter are equally applicable to generating training sets for other machine learning-based inspection modules, such as ADR, ADC, auto-navigation modules, auto-segmentation modules, and the like.

[0045] refer to Figure 2 , shows a generalized flow chart for automatically determining inspection-related data based on a manufacturing process (FP) image using a DNN 112. As described in U.S. Patent Application No. 2017 / 0177997 assigned to the assignee of the present application, and the entire text of the U.S. Patent Application is incorporated herein by reference, the process includes a setup step, the setup step includes training a deep neural network (DNN) 112, wherein the DNN is trained for a given inspection-related application, and when trained, the DNN characterizes a specific set of applications based on the trained parameters. By way of non-limiting example, the inspection-related application can be one of the following:

[0046] defect classification using attributes generated by the DNN (defining classes may include modifying and / or updating pre-existing class definitions and / or identifying new classes);

[0047] Segmentation of manufacturing process images, including segmenting FP images into segments (e.g., material type, edges, pixel markers, regions of interest, etc.);

[0048] Defect detection (e.g., using the FP image to identify one or more candidate defects (if any), and marking the candidate defects, determining the true value of the candidate defects, obtaining shape information of the defects, etc.);

[0049] Registration between two or more images, including obtaining geometric distortion parameters between the images (which can be global or local, simple as a shift or more complex transformations);

[0050] Cross-modality regression (e.g., reconstructing an image from one or more images from different examination modalities, e.g., SEM or optical images from CAD, height maps from SEM images, high-resolution images from low-resolution images);

[0051] Regression-based image attribute estimation;

[0052] Combinations of the above.

[0053] During the setup step, a DNN 112 trained for a given application is obtained (201). Figure 3-4 In further detail, PMC 104 (e.g., training set generator 111) can be configured to perform generation of a specific application training set that can be further used for training of DNN 112. Generating the training set can include augmenting the training samples (and, if applicable, ground truth data), and including the augmented training images and augmented ground truth data in the training samples. Alternatively or additionally, generating the training set can include generating synthetic images (and, if necessary, obtaining ground truth data), and including the synthetic images in the training samples.

[0054] Note that the DNN training process can be cyclical and can be repeated several times until the DNN is fully trained. The process can start with an initially generated training set, while the user provides feedback on the results achieved by the DNN based on the initial training set. The feedback provided can include, for example:

[0055] Manual reclassification of one or more pixels, regions and / or defects;

[0056] Prioritization of categories;

[0057] changes in sensitivity, updates of ground truth segmentation, and / or manual definition of regions of interest (ROI) for segmentation applications;

[0058] Redefine masks / bounding boxes for defect detection applications;

[0059] Reselect failed cases for registration application and / or manually register failed cases;

[0060] Reselect features of interest for regression-based applications, etc.

[0061] The PMC may adjust the next training cycle based on the received feedback. The adjustment may include at least one of the following: updating the training set, updating the cost function, updating the front DNN and / or the back / DNN algorithm, etc. Optionally, some training cycles may not be provided to the entire DNN 112, but to one or more higher layers of the front DNN module 113, the back DNN module 115, or the DNN module 114.

[0062] During operation, the PMC of the FPEI system processes FP samples including FP images using the obtained trained DNN 112 (202). The PMC thereby obtains inspection-related data of a specific application characterizing at least one image in the processed FP samples (203). When processing the FP samples, the PMC may also use predefined parameters and / or parameters received from other sources, as well as training-based parameters that characterize the DNN 112 during training.

[0063] The FP images in the FP sample may be from different inspection modalities (e.g., from different inspection tools; from different channels of the same inspection tool, such as bright field and dark field images; from the same inspection tool using different operating parameters; may be derived from design data, etc.). Note that the inspection modalities may differ from one another due to the source of the respective images (e.g., images captured by a scanning electron microscope (SEM), images captured by an optical inspection system, images derived from captured images, images generated based on CAD, etc.), and / or due to the derivation techniques applied to the captured images (e.g., images derived by segmentation, defect contour extraction, height map calculation, etc.), and / or due to inspection parameters (e.g., perspective and / or resolution provided by a certain inspection tool, etc.). By way of non-limiting example, the FP sample may include a set of images associated with defects, which are classified and obtained by the same or different inspection modalities, reference mold images, CAD-based images, data derived from the obtained images (e.g., height maps, defect masks, grades, segmentations, etc.). The FP sample may also include metadata related to the image and / or defects therein (e.g., imaging conditions, pixel size, engineering properties (e.g., defect size, orientation, background fragmentation, etc.), etc.). By way of non-limiting example, the metadata may be generated by the PMC according to predefined instructions stored in the PMC and / or received from various inspection tools.

[0064] Optionally, the training samples may correspond to the FP samples. Each training sample may include at least the same number of images obtained by the same inspection modality and have the same relationship with the images in the corresponding FP sample. Note that in some embodiments, the training samples may lack some images corresponding to the images in the FP samples. These defects can be compensated by different techniques, some of which are known in the art (for example, missing images can be compensated by images generated based on the average of corresponding images from another training sample, etc.). Optionally, the training samples may further include additional images obtained by additional inspection modalities that are typically not available at runtime, and / or by enhancing images that can be used for training, and / or by generating one or more synthetic images.

[0065] For example, the FP image can be selected from an image of a sample (e.g., a wafer or portion thereof) captured during a manufacturing process, derivative images of captured images obtained through various pre-processing stages (e.g., an image of a portion of a wafer or photomask captured by a SEM or optical inspection system, an SEM image roughly centered on a defect to be classified by an ADC, an SEM image of a larger area where defects are to be located by an ADR, registered images of different inspection modalities corresponding to the same mask position, segmented images, height map images, etc.), and computer-generated images based on design data.

[0066] By way of non-limiting example, inspection-related data for a particular application may represent a per-pixel map of values, the meaning of which depends on the application (e.g., a binary map for defect detection; a discrete map representing a family type or generic class of harmful family prediction; a discrete map for defect type classification; a continuous value for cross-modal or die-to-model (D2M) regression, etc.) A map for each pixel may further be obtained, as well as a probability map for each pixel indicating the probability of the value obtained for the pixel.

[0067] Alternatively or additionally, the inspection-related data obtained for a specific application may represent one or more values ​​that summarize the entire image content (not every pixel), such as (for example) defect attributes, defect bounding box candidates and associated defect probabilities for automatic defect review applications, defect classes and class probabilities for automatic defect classification applications, etc.

[0068] Alternatively or additionally, the obtained application-specific defect-related data may not be directly related to the defect, but may be used for defect analysis (e.g., boundaries between layers of the wafer obtained by segmentation of the FP image may be used to define the layers of the defect, defect environment data, e.g., characteristics of the background pattern, etc.). Alternatively or additionally, the inspection-related data may be used for metrology purposes.

[0069] Reference Figure 3, a training set generator 111 is shown. According to certain embodiments of the presently presented subject matter, the training set generator 111 includes an enhanced image generator 301, a synthetic image generator 302, and an output training set module 303, which is operably coupled to the enhanced image generator 301 and the synthetic image generator 302 and is configured to generate a training set including "real world" training samples, synthetic training samples, and / or enhanced training samples. The generated training set can be stored in a memory of the PMC 104.

[0070] The training set generator may obtain data for “real world” training samples (and optionally ground truth data) from data repository 109 , CAD server 110 , or any other suitable data repository via input interface 105 . Alternatively or additionally, ground truth data may be obtained via GUI 108 .

[0071] Depending on the application, a "real world" training sample can be a single image or a set of images of a sample obtained by the same or different inspection modalities. "Real world" training samples can also include digital data associated with the image, such as metadata, hand-crafted attributes (such as defect size, orientation, background fragments, etc.), and others. The value of ground truth data can include images and / or labels associated with training samples for a specific application. Ground truth data can be synthetically generated (e.g., CAD-based images), actually generated (e.g., captured images), generated through machine learning annotations (e.g., labels based on feature extraction and analysis), generated through manual annotations, or a combination of the above, etc.

[0072] Note that, depending on the application, the same image can be used for training samples or ground truth data. By way of non-limiting example, a CAD-based image can be used as a training sample for a segmentation application and can be used as ground truth data for a regression-based application. As described in U.S. Patent Application No. 2017 / 0177997, the ground truth data may vary depending on the application.

[0073] The enhanced image generator 301 is constructed to enhance one or more images from the "real world" training samples, thereby deriving an enhanced image. Image enhancement from the "real world" training samples can be provided by various image processing techniques, including adding noise, blurring, geometric transformation (e.g., rotation, stretching, simulating different angles, cropping, scaling, etc.), tone mapping, changing the vector information of one or more pixels of the image (e.g., adding and / or modifying the perspective or acquisition channel, etc.), etc.

[0074] Alternatively or additionally, images from "real world" training samples may be augmented with synthetic data (e.g., defect related data, simulated connectors or other objects, implants from other images, etc.). By way of non-limiting example, available images (and / or parameters) of known defect types may be used to implant new defects in images, amplify the defect intensity of pre-existing defects in images, remove defects from images, mask defects in images (making them more difficult to detect), etc.

[0075] For example, generating an image with synthetic defects may include:

[0076] - Train CNN DNN (e.g. based on Alex-Net, etc.);

[0077] - obtaining a first image comprising selected pixels;

[0078] - obtaining a second image containing the selected background;

[0079] - Use the trained CNN network to implement a "style transfer" of the selected defect to the selected background. This implementation can be done, for example, using https: / / arxiv.org / pdf / 1508.06576.pdf Provided by the technology disclosed in.

[0080] Alternatively or additionally, however, captured images from "real world" training samples may be enhanced using segmentation, defect contour extraction and / or height map calculation, and / or may be obtained by processing together with corresponding CAD-based images.

[0081] The enhancement techniques may be applied to the image(s) of the "real world" training sample in an application-independent manner. Alternatively, the enhanced training sample may be derived in an application-specific manner, where at least a portion of the corresponding enhanced image is obtained by a technique that is preferred over the specific application. Table 1 shows non-limiting examples of preferred enhancement techniques for specific applications as detailed in U.S. Patent Application No. 2017 / 0177997.

[0082] Table 1. Examples of specific applications where enhancement techniques are preferred

[0083]

[0084] Optionally, the enhancement technique may also be applied to the synthetic image generated by the synthetic image generator 302 , and the training set generated by the output training set module 303 may further include the synthetic image generated by the synthetic image generator 302 and further enhanced by the enhanced image generator 301 .

[0085] Therefore, according to certain embodiments of the presently disclosed subject matter, the number of training samples of the generated training set can be significantly greater than the number of "real-world" training samples. For example, a "real-world" training sample set may include 100 to 50,000 training samples, while a training set generated according to certain embodiments of the presently disclosed subject matter may include at least 100,000 training samples. It is worth noting that capturing images is a slow and potentially expensive process. Generating a sufficient number of training samples in the training set by enhancing captured images and / or generating synthetic images can improve the efficiency of DNN training.

[0086] Furthermore, techniques for enhanced and synthetic image generation enable the generated training sets to be tailored to the needs of a particular application. By way of non-limiting example, a lack of FAB data associated with a particular defect class (e.g., a minority class) can be compensated for by presenting synthetic and / or enhanced images of the corresponding defect. By way of further non-limiting example, a lack of appropriate FAB data associated with a particular layer can be compensated for by synthetic images associated with that layer. Similarly, a lack of data from a particular tool required for tool variation training can be compensated for by synthetic images associated with that tool.

[0087] refer to Figure 4 , shows a generalized flowchart for generating enhanced images for the training set.

[0088] For purposes of illustration only, the following description provides a training set that can be used to train a DNN, which can also be used for inspection of semiconductor samples. Those skilled in the art will readily appreciate that the teachings of the presently disclosed subject matter are equally applicable to training sets that can be used for automatic defect review, automatic defect classification, automatic navigation during inspection, automatic segmentation of FP images, metrology based on FP images, and other inspection processes including machine learning. According to certain embodiments of the presently disclosed subject matter, image enhancement for the training set can achieve robustness of the trained DNN to different process variations.

[0089] For illustrative purposes only, the following description is provided for enhancing "real world" images from a training set. Those skilled in the art will readily appreciate that the teachings of the presently disclosed subject matter are equally applicable to synthetic images generated for a training set and / or previously enhanced images in a training set.

[0090] The process includes obtaining, by the FPEI system, a mathematical model (hereinafter also referred to as a simulation model) that is capable of simulating the effect (usually an undesirable effect, although not necessarily) of a certain physical process on the FP image depending on the value of the physical process parameter(s) (402). A training set generator applies the simulation model to the image to be enhanced for the training set (401), thereby generating one or more enhanced images corresponding to one or more different parameter values ​​of the certain physical process (403). The generated enhanced images can be added to the training set (404). The simulation model can be received by the FPEI system 103 from other blocks of the inspection system (e.g., from the storage system 107) and / or from (multiple) external systems. Optionally, at least a portion of the simulation model can be generated by the FPEI system 103.

[0091] Note that the simulated model may be a parametric model or an estimated model of the corresponding physical influence. It should also be noted that an enhanced image may be obtained by applying two or more models simulating different physical influences.

[0092] Optionally, the enhanced images generated in the training set may be weighted, and these weights are taken into account when applying the training set. As a non-limiting example, weightings may be provided between the original image and the enhanced image. Alternatively or additionally, weightings may be provided between the enhanced images (e.g., depending on the likelihood of corresponding values ​​of the physical process parameter).

[0093] Optionally, the generated augmented image can be added to the training set in association with the ground truth data. The original image and the augmented image can be associated with the same ground truth data (e.g., defect labels). Alternatively, the augmented image can be associated with augmented ground truth data (e.g., augmented class distribution in the training set).

[0094] Optionally, an enhanced image may be obtained by applying the simulation model together with other data (eg, CAD data, reference images, etc.).

[0095] The following are non-limiting examples of applying the above-described methods to reduce the impact of different interference effects in accordance with certain implementations of the presently disclosed subject matter.

[0096] noise

[0097] "Real-world" images contain noise, where each gray-level (GL) intensity contains noise intensities of different scales. Applying a noise simulation model with desired noise parameter values ​​to the original image can result in an enhanced image with different desired levels of added noise. Therefore, a training set including such enhanced images will achieve robustness of the trained DNN to the addition of noise in the FT samples.

[0098] GL-Tool Calibration:

[0099] The grayscale of images captured by different tools can differ by a factor, even when captured at the same resolution at the same location on the same wafer. Therefore, training a DNN on images captured by tools other than the tool that captured the FP images reduces the generalization ability of the DNN. According to certain embodiments of the presently disclosed subject matter, each given tool can be associated with a GL simulation model customized for the given tool. Applying such a model with expected GL intensity values ​​to images captured by a given tool will result in enhanced images of the training set. A training set containing such enhanced images will achieve robustness to GL tool variations and reduce the necessity to collect training data from multiple tools.

[0100] Focus Error:

[0101] Certain defect classes (e.g., large pattern damage, large particles, etc.) are more sensitive to focus errors than other classes. According to certain embodiments of the presently disclosed subject matter, a training set generator can apply a defocus simulation model to defect images of defocus sensitive classes. The resulting enhanced defects with desired defocus values ​​can be added to the training set in association with the class indication label of the corresponding defect.

[0102] Color Variation (CV):

[0103] Color variations can occur within a single image (e.g., due to layer thickness variations) or between a defect and a reference image. CV may result from different tool calibrations and can be characterized by local variations in GL in the image. Applying a CV model to the original image can result in enhanced images with different desired levels of additional variation. Therefore, a training set including such enhanced images will achieve increased robustness to color variations in FT samples.

[0104] Charge:

[0105] The charging effect causes artificial differences between defect images and reference images and can significantly affect the inspection process. Applying a simulation model of the charging effect to the original image can produce an enhanced image, allowing the trained DNN to ignore the differences due to the charging effect.

[0106] vibration:

[0107] Vibration can cause a widening of the image edges. Applying a model that simulates the blurred edges caused by vibration can produce enhanced images that are more robust to the edge smirking effect.

[0108] Quaternary detector miscalibration

[0109] The calibration of the quad detector may be suboptimal, resulting in image overlap within different viewing angles. Applying a simulated model (optionally together with CAD data to enhance edge detection) enables obtaining enhanced images with different desired levels of additional overlap. Thus, the trained DNN can become more robust with respect to tool calibration.

[0110] One of the advantages of certain embodiments of the enhancement process, including the above-described illustrations, is that it reduces the impact of interference on the trained DNN, thereby improving the robustness and effectiveness of the DNN trained for ADC, ADR, matching, metrology, and other inspection tasks.

[0111] It will be understood that the invention is not limited in its application to the details set forth in the description contained herein or illustrated in the drawings.

[0112] It will also be understood that the system according to the present invention can be implemented at least in part on a suitably programmed computer. Likewise, the present invention contemplates a computer program that can be read by a computer to perform the method of the present invention. The present invention also contemplates a non-transitory computer-readable memory that tangibly embodies a program of instructions executable by a computer for performing the method of the present invention.

[0113] The present invention is capable of other embodiments and can be practiced or implemented in various ways. Therefore, it will be understood that the words and terms used herein are for descriptive purposes and should not be considered limiting. Therefore, those skilled in the art will appreciate that the concepts upon which the present disclosure is based can be easily used as the basis for designing other structures, methods and systems for achieving the several purposes of the presently disclosed subject matter.

[0114] It will be easily understood by those skilled in the art that various modifications and variations may be applied to the embodiments of the present invention as described above without departing from the protective scope of the present invention as determined in the accompanying claims.

Claims

1. A method for generating a training set that can be used to inspect semiconductor samples, the method being executed by a computer and comprising: Obtaining a simulation model capable of simulating the effect of a physical process on a real-world fabrication process (FP) image, wherein the simulation depends on parameter values ​​of the physical process; applying the simulation model to the real-world FP images to be enhanced for the training set, thereby generating one or more enhanced images corresponding to one or more different values ​​of a parameter of the physical process; and including the generated one or more enhanced images into the training set, The training set including the one or more enhanced images is used to train a deep neural network (DNN) to process one or more real-world FP images to obtain inspection-related data of a specific application characterizing at least one of the one or more real-world FP images.

2. The method of claim 1, wherein the training set can be used for at least one inspection process selected from the group consisting of: sample inspection using the trained DNN, automatic defect review, automatic defect classification, automatic navigation during inspection, automatic segmentation of FP images, and automatic metrology based on FP images.

3. The method of claim 1, wherein the effect of the physical process on the FP image is an undesirable effect. The method of claim 1 , wherein the FP image is a SEM image.

5. The method of claim 1, wherein the simulation model is a parametric model or an estimation model of the corresponding simulated influences of the physical process.

6. The method of claim 1, wherein the enhanced images generated in the training set are weighted when included in the training set.

7. The method of claim 1, wherein the simulated impact of the physical process is selected from the group consisting of: noise, gray level (GL) calibration between tools, focus error, color variation, charging, vibration, and quad detector miscalibration.

8. The method of claim 1, wherein the one or more enhanced images generated are included in the training set in association with ground truth data.

9. The method of claim 1, wherein the one or more enhanced images are generated by applying the simulation model together with other data.

10. A system useful for inspection of a semiconductor sample, the system comprising a processing and memory circuitry (PMC) operably connected to an input interface, wherein: The input interface is configured to receive real-world manufacturing process (FP) images to be augmented for a training set; and The PMC is constructed to: obtaining a simulation model capable of simulating the effect of a physical process on a real-world FP image, wherein the simulation depends on parameter values ​​of the physical process; applying the simulation model to the real-world FP images to be enhanced for the training set, thereby generating one or more enhanced images corresponding to one or more different values ​​of a parameter of the physical process; including the generated one or more enhanced images into the training set; and A deep neural network (DNN) is trained using the training set including the one or more enhanced images to process one or more real-world FP images to obtain inspection-related data of a specific application characterizing at least one of the one or more real-world FP images.

11. The system of claim 10, wherein the training set can be used for at least one inspection process selected from the group consisting of: inspection of samples using the trained DNN, automatic defect review, automatic defect classification, automatic navigation during inspection, automatic segmentation of FP images, and automatic metrology based on FP images.

12. The system of claim 10, wherein the simulation model is a parametric model or an estimated model of the corresponding simulated impacts of the physical process.

13. The system of claim 10, wherein the PMC is constructed to provide weights to the generated enhanced images before including them in the training set.

14. The system of claim 10, wherein the simulated effects of the physical process are selected from the group consisting of: noise, gray level (GL) calibration between tools, focus error, color variation, charging, vibration, and quad detector miscalibration.

15. The system of claim 10, wherein the generated one or more augmented images are included in the training set in association with ground truth data.

16. The system of claim 10, wherein the PMC is configured to generate the one or more enhanced images by applying the simulation model together with other data.

17. A non-transitory computer readable medium comprising instructions which, when executed by a computer, cause the instructions to perform a method for training a set of semiconductor samples for inspection, the method comprising: Obtaining a simulation model capable of simulating the effect of a physical process on a real-world manufacturing process (FP) image, wherein the simulation depends on parameter values ​​of the physical process; applying the simulation model to the real-world FP images to be enhanced for the training set, thereby generating one or more enhanced images corresponding to one or more different values ​​of a parameter of the physical process; and including the generated one or more enhanced images into the training set, The training set including the one or more enhanced images is used to train a deep neural network (DNN) to process one or more real-world FP images to obtain inspection-related data of a specific application characterizing at least one of the one or more real-world FP images.

Images