Deep learning image denoising for semiconductor-based applications

Abstract

This invention provides a method and system for determining information about a sample. The system includes a computer subsystem and one or more components executed by the computer subsystem. The one or more components include a deep learning model configured to remove noise from an image of a sample generated by an imaging subsystem. The computer subsystem is configured to determine information about the sample from the noise-removed image.

Description

[Technical Field]

[0001] This invention generally relates to a method and system for determining information about a sample. Some embodiments relate to a deep learning model configured for removing noise from an image of the sample used to determine information about the sample. [Previous Technology]

[0002] The following descriptions and examples are not acknowledged as prior art by virtue of their inclusion in this paragraph.

[0003] Manufacturing semiconductor devices (such as logic and memory devices) typically involves processing a substrate (such as a semiconductor wafer) using numerous semiconductor processes to form various features and multiple layers of such semiconductor devices. For example, lithography is a semiconductor process that involves transferring a pattern from a photomask to a photoresist disposed on a semiconductor wafer. Additional examples of semiconductor processes include (but are not limited to) chemical mechanical polishing (CMP), etching, deposition, and ion implantation. Multiple semiconductor devices can be fabricated in an arrangement on a single semiconductor wafer and then separated into individual semiconductor devices.

[0004] Inspection procedures are used at various steps during a semiconductor manufacturing process to detect defects on samples in order to drive higher yields and thus higher profits in the process. Inspection has always been an important part of manufacturing semiconductor devices. However, as the size of semiconductor devices shrinks, inspection becomes even more important for the successful manufacture of acceptable semiconductor devices, because even small defects can cause device failure.

[0005] There are many different ways to process images, outputs, etc., generated by tools configured as described above for their judgment information. Typically, in the case of inspection, a test image is compared with a reference image so that any differences between the test image and the reference image can be identified and these differences can be used to detect defects on the sample. Regardless of how the judgment information is determined, mitigating the impact of image noise on the judged information can be challenging, and a great deal of research and development has been carried out to overcome the obstacles posed by image noise.

[0006] Some currently used defect detection algorithms are designed to reduce noise on the tool. For example, some currently used algorithms use a linear combination of reference frames from multiple dies on a wafer to compute a reference frame, resulting in a reference image with less noise. Another approach is to use a differential filter, which is typically a matrix expected to improve the signal-to-noise ratio (SNR) of a defect when convolved with a different image. In cases where a scaled-down mask is printed on a wafer and the wafer is inspected to detect defects on the scaled-down mask, dies or scaled-down masks from one or two columns can be stacked to reduce random noise and improve the SNR for detecting recurring defects. Another approach to reduce the impact of noise is to use deep learning to learn a filter that is applied after detection to extract new properties for perturbation point event filtering (NEF).

[0007] The noise reduction strategies described above have several drawbacks. For example, the inventors are unaware of any current methods for noise reduction used to remove noise from test images. In one example, the method of linearly combining similar noise implementations only performs this operation on a reference image. Furthermore, even after the reference generation described above, designed for noise reduction, residual noise may still exist in the differential image generated using the reference. In many cases, further noise reduction may be required to achieve the desired accuracy of the determined information. In another example, the differential image filter is a standard kernel and is not adapted to different noise distributions. Defining a custom differential image filter requires specialized expertise. Moreover, stacked dies or magnified masks generated for comparison with a golden database reference address specific use cases (such as magnified mask inspection performed using a wafer printed with a magnified mask) but cannot be applied to the detection of non-repeating defects.

[0008] Therefore, it would be advantageous to develop a system and method for determining information about a sample that does not have one or more of the aforementioned disadvantages. [Summary of the Invention]

[0009] The following description of the various embodiments should not be construed in any way as limiting the subject matter of the accompanying technical solutions.

[0010] One embodiment relates to a system configured to determine information about a sample. The system includes a computer subsystem and one or more components executed by the computer subsystem. The one or more components include a deep learning model configured to remove noise from an image of a sample generated by an imaging subsystem. The computer subsystem is configured to determine information about the sample from the noise-removed image. The system can be further configured as described herein.

[0011] Another embodiment relates to a computer-implemented method for determining information about a sample. The method includes removing noise from an image of a sample generated by an imaging subsystem by inputting the image into a deep learning model configured to perform noise removal. The deep learning model is included in one or more components executed by a computer subsystem. The method also includes determining information about the sample from the noise-removed image. The input and the determination are performed by the computer subsystem.

[0012] The steps of this method may be performed as further described herein. This method may include any other steps of any other method described herein. This method may be performed by any of the systems described herein.

[0013] Another embodiment relates to a non-transitory computer-readable medium storing program instructions that can be executed on a computer system to perform a computer-implemented method for determining information of a sample. The computer-implemented method includes the steps of the method described above. The computer-readable medium can be further configured as described herein. The steps of the computer-implemented method can be performed as further described herein. Additionally, the computer-implemented method to which the program instructions can be executed may include any other steps of any other method(s) described herein.

Implementation Method

[0015] A “disruptive point” (which may sometimes be used interchangeably with a “disruptive point defect” or a “disruptive point event”) is generally defined, as used herein, as a defect of no concern to the user and / or an event detected on a sample that is not a true actual defect on that sample. Disruptive points that are not actually defects may be detected as events due to non-defect noise sources on a sample (e.g., grains in a metal line on the sample, signals from the underlying material or material on the sample, line edge roughness (LER), relatively small critical size (CD) variations in patterned features, thickness variations, etc.) and / or due to the edge properties of the detection system itself or its configuration for detection.

[0016] As used herein, the term "Defect of Interest (DOI)" is defined as a defect detected on a sample that is an actual defect on that sample. Therefore, a DOI is of interest to a user because the user is generally concerned with how many and what kinds of actual defects are present on the tested sample. In some contexts, the term "DOI" is used to refer to a subset of all actual defects on a sample, containing only the actual defects of interest to a user. For example, multiple types of DOIs may exist on any given sample, and one or more of them may be of greater interest to a user than one or more other types. However, in the context of the embodiments described herein, the term "DOI" is used to refer to any and all actual defects on a sample.

[0017] Referring now to the drawings, it should be noted that the drawings are not drawn to scale. Specifically, the scale of some components in the drawings is greatly exaggerated to highlight the characteristics of such components. It should also be noted that the drawings are not drawn to the same scale. The same component symbols have been used to indicate components that may be similarly configured and shown in more than one drawing. Unless otherwise mentioned herein, any component described and shown may include any suitable commercially available component.

[0018] Generally, the embodiments described herein are systems and methods for determining information about a sample. More specifically, the embodiments described herein are configured for image noise removal for applications such as optical defect detection.

[0019] One particularly useful application of the embodiments described herein is the use of a convolutional neural network (CNN) to remove noise from broadband plasma (BBP) patch images. BBP tools are among the most sensitive optical inspection tools in wafer fabs. Sensitivity to yield-critical defects and gap defects is a primary expectation of tool users and is therefore a driving force behind innovation in these tools. As design nodes and defect sizes shrink, detecting critical DOIs has become extremely challenging, even with optimal optical inspection modes on these tools. Therefore, there is a strong need for algorithms to enhance sensitivity. In practice, sensitivity translates to a signal-to-noise ratio (SNR) for a given DOI type. Improving SNR can be achieved by enhancing the signal or reducing noise. The embodiments described herein can be advantageously used to reduce noise in BBP (or other) images using a CNN or other deep learning (DL) models described herein, and, as needed, compute new properties to further enhance sensitivity.

[0020] In some embodiments, the sample is a wafer. The wafer may comprise any wafer known in the semiconductor field. Although some embodiments may be described herein with respect to one or more wafers, the embodiments are not limited to samples in which they may be used. For example, the embodiments described herein may be used in samples such as photomasks, tablets, personal computer (PC) boards, and other semiconductor samples.

[0021] Figure 1 illustrates one embodiment of a system configured to determine information for a sample. In some embodiments, the system includes an imaging subsystem (such as imaging subsystem 100). The imaging subsystem includes and / or is coupled to a computer subsystem (e.g., computer subsystem 36) and / or one or more computer systems 102.

[0022] Generally, the imaging subsystem described herein includes at least one energy source, one detector, and one scanning subsystem. The energy source is configured to generate energy directed to a sample by the imaging subsystem. The detector is configured to detect energy from the sample and respond to the detected energy by generating an output. The scanning subsystem is configured to change the location on the sample where the energy is directed and from which it detects the energy. In one embodiment, as shown in FIG1, the imaging subsystem is configured as a light-based imaging subsystem. In this way, the sample image described herein can be generated by a light-based imaging subsystem.

[0023] In the light-based imaging subsystem described herein, the energy directed to the sample comprises light, and the energy detected from the sample comprises light. For example, in an embodiment of the system shown in FIG1, the imaging subsystem includes an illumination subsystem configured to direct light to sample 14. The illumination subsystem includes at least one light source. For example, as shown in FIG1, the illumination subsystem includes light source 16. The illumination subsystem is configured to direct light to the sample at one or more incident angles (which or the like may include one or more tilt angles and / or one or more normal angles). For example, as shown in FIG1, light from light source 16 is directed at a tilted incident angle through optical element 18 and then through lens 20 to sample 14. The tilted incident angle may include any suitable tilted incident angle that may vary depending on, for example, the characteristics of the sample and the procedures performed on the sample.

[0024] The illumination subsystem can be configured to guide light to the sample at different incident angles at different times. For example, the imaging subsystem can be configured to change one or more characteristics of one or more elements of the illumination subsystem so that light can be guided to the sample at an incident angle different from that shown in FIG1. ​​In one example of this, the imaging subsystem can be configured to move the light source 16, the optical element 18 and the lens 20 so that light is guided to the sample at a different tilted incident angle or a normal (or near-normal) incident angle.

[0025] In some examples, the imaging subsystem can be configured to simultaneously direct light to the sample at more than one incident angle. For example, the illumination subsystem may include more than one illumination channel, one of which may include the light source 16, optical element 18, and lens 20 shown in FIG. 1, and another of the illumination channels (not shown) may include similar elements that may be configured differently or the same, or may include at least one light source and possibly one or more other components, such as those further described herein. If this light is directed to the sample simultaneously with other light, one or more characteristics (e.g., wavelength, polarization, etc.) of the light directed to the sample at different incident angles may be different, such that the light generated by illuminating the sample at different incident angles can be distinguished from each other at (a number of) detectors.

[0026] In another example, the illumination subsystem may include only one light source (e.g., source 16 shown in FIG. 1), and light from the light source may be separated into different optical paths by one or more optical elements (not shown) of the illumination subsystem (e.g., based on wavelength, polarization, etc.). The light in each of these different optical paths may then be directed to the sample. Multiple illumination channels may be configured to direct light to the sample simultaneously or at different times (e.g., when different illumination channels are used to sequentially illuminate the sample). In another example, the same illumination channel may be configured to direct light with different characteristics to the sample at different times. For example, optical element 18 may be configured as a spectral filter, and the properties of the spectral filter may be changed in various different ways (e.g., by replacing one spectral filter with another) so that light of different wavelengths can be directed to the sample at different times. The illumination subsystem may have any other suitable configuration known in the art for sequentially or simultaneously directing light with different or the same characteristics to the sample at different or the same incident angles.

[0027] Light source 16 may include a BBP light source. In this way, the light generated by and directed to the sample by the light source may include broadband light. However, the light source may include any other suitable light source, such as any suitable laser known in this art configured to generate light (some) of any suitable wavelength. The laser may be configured to generate monochromatic or near-monochromatic light. In this way, the laser may be a narrow-band laser. The light source may also include a multicolor light source that generates light of multiple discrete wavelengths or bands.

[0028] Light from optical element 18 can be focused onto sample 14 by lens 20. Although lens 20 is shown as a single refractive optical element in FIG. 1, in practice, lens 20 may include several refractive and / or reflective optical elements that combine to focus light from optical element onto sample. The illumination subsystem shown in FIG. 1 and described herein may include any other suitable optical elements (not shown). Examples of such optical elements include (but are not limited to): several polarizing components, several spectral filters, several spatial filters, several reflective optical elements, several apodizers, several beam splitters, several apertures, and the like, which may include any such suitable optical elements known in the art. In addition, the subsystem can be configured to change one or more of the elements of the illumination subsystem based on the type of illumination to be used for imaging.

[0029] The imaging subsystem may also include a scanning subsystem configured to change the position of light directed to and from the sample on which it detects the light, and may cause a light scan across the sample. For example, the imaging subsystem may include a stage 22 on which the sample 14 is placed during imaging. The scanning subsystem may include any suitable mechanical and / or robotic assembly (which includes the stage 22) that can be configured to move the sample so that light can be directed to and detected from different positions on the sample. Alternatively or additionally, the imaging subsystem may be configured to perform a scan of the light across the sample by one or more optical elements of the imaging subsystem, so that light can be directed to and detected from different positions on the sample. In the example of light scanning across the sample, the light can be scanned across the sample in any suitable manner (such as in a serpentine path or a spiral path).

[0030] The imaging subsystem further includes one or more detection channels. At least one of the detection channels includes a detector configured to detect light originating from the sample attributable to illumination of the sample by the imaging subsystem, and to generate an output in response to the detected light. For example, the imaging subsystem shown in FIG1 includes two detection channels, one formed by a light collector 24, element 26, and detector 28, and the other formed by a light collector 30, element 32, and detector 34. As shown in FIG1, the two detection channels are configured to collect and detect light at different collection angles. In some examples, the two detection channels are configured to detect scattered light, and these detection channels are configured to detect light scattered from the sample at different angles. However, one or more of the detection channels may be configured to detect another type of light from the sample (e.g., reflected light).

[0031] As further shown in Figure 1, the two detection channels are shown positioned in the plane of the paper, and the illumination subsystem is also shown positioned in the plane of the paper. Therefore, in this embodiment, the two detection channels are positioned (e.g., centered) in the plane of incidence. However, one or more of the detection channels may be positioned outside the plane of incidence. For example, the detection channel formed by the light collector 30, element 32, and detector 34 may be configured to collect and detect light scattered from the plane of incidence. Therefore, this detection channel may be collectively referred to as a "side" channel, and this side channel may be centered in a plane substantially perpendicular to the plane of incidence.

[0032] Although Figure 1 illustrates one embodiment of an imaging subsystem comprising two detection channels, the imaging subsystem may comprise a different number of detection channels (e.g., only one detection channel or two or more detection channels). In one example, the detection channel formed by the light collector 30, element 32, and detector 34 may form a side channel as described above, and the imaging subsystem may comprise an additional detection channel (not shown) formed as another side channel positioned on the opposite side of the incident plane. Thus, the imaging subsystem may comprise a detection channel comprising the light collector 24, element 26, and detector 28 and centered in the incident plane and configured to collect and detect light scattered at (a number of) scattering angles normal to or near normal to the sample surface. Thus, this detection channel may be collectively referred to as a "top" channel, and the imaging subsystem may also comprise two or more side channels configured as described above. Therefore, the imaging subsystem may include at least three channels (i.e., one top channel and two side channels), and each of the at least three channels has its own light collector, which is configured to collect light at a different scattering angle than the other light collectors.

[0033] As further described above, each of the detection channels included in the imaging subsystem can be configured to detect scattered light. Therefore, the imaging subsystem shown in Figure 1 can be configured for dark-field (DF) imaging of the sample. However, the imaging subsystem may also, or alternatively, include several detection channels configured for bright-field (BF) imaging of the sample. In other words, the imaging subsystem may include at least one detection channel configured to detect light reflected from the specular surface of the sample. Therefore, the imaging subsystem described herein can be configured for DF imaging only, BF imaging only, or both DF and BF imaging. Although each of the light collectors is shown as a single refractive optical element in Figure 1, each of the light collectors may include one or more refractive optical elements and / or one or more reflective optical elements.

[0034] One or more detection channels may include any suitable detector known in the art, such as a photomultiplier tube (PMT), charge-coupled device (CCD), or time-delay integration (TDI) camera. These detectors may also include non-imaging detectors or imaging detectors. If the detector is a non-imaging detector, each of these detectors may be configured to detect specific characteristics (such as intensity) of the scattered light, but may not be configured to detect these characteristics as they vary depending on their position within the imaging plane. Therefore, the output generated by each detector in the detection channels included in the imaging subsystem may be a signal or data rather than an image signal or image data. In these examples, a computer subsystem (such as computer subsystem 36) may be configured to generate an image of the sample from the non-imaging output of the detector. However, in other examples, the detector may be configured as an imaging detector configured to generate an image signal or image data. Therefore, the imaging subsystem can be configured to generate images in several ways.

[0035] It should be noted that Figure 1 is provided herein to generally illustrate a configuration of one of the imaging subsystems that may be included in the system embodiments described herein. Clearly, the imaging subsystem configuration described herein can be modified to optimize the performance of the imaging subsystem as normally performed when designing a commercial imaging system. Alternatively, the system described herein can be implemented using existing systems such as the commercially available 29xx / 39xx series tools from KLA Corporation of Milpitas, California (e.g., by adding the functionality described herein to an existing detection system). For some such systems, the methods described herein can be provided as optional functionality of the system (e.g., in addition to other system functionalities). Alternatively, the system described herein can be designed "from scratch" to provide a completely new system.

[0036] Computer subsystem 36 may be coupled to the detector of the imaging subsystem in any suitable manner (e.g., via one or more transmission media, or the like, which may include "wired" and / or "wireless" transmission media), so that the computer subsystem can receive the output generated by the detector. Computer subsystem 36 may be configured to perform certain functions (including the steps and functions further described herein) with or without using the output of the detector. Thus, the steps described herein may be performed "on the tool" by a computer subsystem coupled to or as part of an imaging subsystem. Alternatively or additionally, (a number of) computer systems 102 may perform one or more of the steps described herein. Thus, one or more of the steps described herein may be performed "outside the tool" by a computer system not directly coupled to an imaging subsystem. Computer subsystem 36 and (a number of) computer systems 102 may be further configured as described herein.

[0037] Computer subsystem 36 (and other computer subsystems described herein) may also be referred to herein as computer system(s). Each of the computer subsystem(s) or system(s) described herein may take various forms, including a personal computer system, a video computer, a mainframe computer system, a workstation, a network appliance, an internet appliance, or other device. Generally, the term "computer system" may be broadly defined to encompass any device having one or more processors that execute instructions from a memory medium. The computer subsystem(s) or system(s) may also include any suitable processor known in the art, such as a parallel processor. Additionally, the computer subsystem(s) or system(s) may include a computer platform with high-speed processing and software as a stand-alone tool or a network-connected tool.

[0038] If the system comprises more than one computer subsystem, the different computer subsystems can be coupled to each other, enabling the transmission of images, data, information, instructions, etc., between the computer subsystems. For example, computer subsystem 36 can be coupled to (a number of) computer systems 102 (as shown by the dashed lines in Figure 1) via any suitable transmission medium (which may include any suitable wired and / or wireless transmission medium known in this art). Two or more of these computer subsystems can also be effectively coupled via a shared computer-readable storage medium (not shown).

[0039] Although the imaging subsystem has been described above as an optical or light-based imaging subsystem, in another embodiment, the imaging subsystem is configured as an electronic imaging subsystem. In this way, the sample image described herein can be generated by an electronic imaging subsystem. In an electron beam imaging subsystem, the energy directed to the sample contains electrons, and the energy detected from the sample contains electrons. In one embodiment shown in FIG. 1a, the imaging subsystem includes an electron column 122, and the system includes a computer subsystem 124 coupled to the imaging subsystem. The computer subsystem 124 can be configured as described above. Additionally, this imaging subsystem can be coupled to one or more other computer systems in the same manner as described above and shown in FIG. 1.

[0040] As also shown in Figure 1a, the electron column includes an electron beam source 126 configured to generate electrons focused onto the sample 128 by one or more elements 130. The electron beam source may include, for example, a cathode source or an emitter tip, and the one or more elements 130 may include, for example, a gun lens, an anode, a beam-limiting aperture, a gate valve, a beam current selection aperture, an objective lens, and a scanning subsystem, all of which may include any suitable elements known in the art.

[0041] Electrons returning from the sample (e.g., secondary electrons) can be focused onto the detector 134 by one or more elements 132. One or more elements 132 may include, for example, a scanning subsystem, which may be the same scanning subsystem included in (a number of) elements 130.

[0042] The electron column may include any other suitable element known in the art. Additionally, the electron column may be further configured as described in: U.S. Patent No. 8,664,594 to Jiang et al., issued April 4, 2014; U.S. Patent No. 8,692,204 to Kojima et al., issued April 8, 2014; U.S. Patent No. 8,698,093 to Gubbens et al., issued April 15, 2014; and U.S. Patent No. 8,716,662 to MacDonald et al., issued May 6, 2014, which are incorporated herein by reference as if their entirety were described herein.

[0043] Although Figure 1a shows the electron column configured such that electrons are guided to the sample at one angle of incidence and scattered from the sample at another angle, the electron beam can be guided to and scattered from the sample at any suitable angle. Furthermore, the electron beam imaging subsystem can be configured to produce sample outputs using multiple modes, as further described herein (e.g., using different illumination angles, collection angles, etc.). The multiple modes of the electron beam imaging subsystem can differ in the parameters of any output produced by the imaging subsystem.

[0044] The computer subsystem 124 may be coupled to the detector 134 as described above. The detector may detect electrons returning from the surface of the sample, thereby forming an electron beam image (or other output) of the sample. The electron beam image may include any suitable electron beam image. The computer subsystem 124 may be configured to use the output generated by the detector 134 to detect events on the sample, as may be performed as further described herein. The computer subsystem 124 may be configured to perform any additional steps described herein. A system including one of the imaging subsystems shown in FIG1a may be further configured as described herein.

[0045] It should be noted that FIG. 1a is provided herein to generally illustrate a configuration of one of the electron beam imaging subsystems that may be included in the embodiments described herein. As with the optical imaging subsystem described above, the configuration of the electron beam imaging subsystem described herein can be modified to optimize the performance of an imaging subsystem as normally performed when designing a commercial system. Alternatively, the system described herein can be implemented using an existing system such as tools commercially available from KLA (e.g., by adding the functionality described herein to an existing system). For some of these systems, the methods described herein can be provided as optional functionality of the system (e.g., in addition to other system functionalities). Alternatively, the system described herein can be designed "from scratch" to provide a completely new system.

[0046] Although the imaging subsystem has been described above as a light or electron beam imaging subsystem, it may be an ion beam imaging subsystem. This imaging subsystem may be configured as shown in Figure 1a, except that the electron beam source may be replaced by any suitable ion beam source known in the art. Furthermore, the imaging subsystem may include any other suitable ion beam imaging system, such as those included in commercially available focused ion beam (FIB) systems, helium ion microscopy (HIM) systems, and secondary ion mass spectrometry (SIMS) systems.

[0047] As further mentioned above, the imaging subsystem can be configured to have multiple modes. Generally, a "mode" is defined by the values ​​of parameters of the imaging subsystem used to generate the output of the sample. Therefore, different modes can differ in the values ​​of at least one of the imaging parameters of the imaging subsystem (other than the position on the sample generating the output). For example, for a light-based imaging subsystem, different modes can use light of different wavelengths. Modes can differ in the wavelength of the light directed to the sample, as further described herein (e.g., by using different light sources, different spectral filters, etc. for different modes). In another embodiment, different modes can use different illumination channels. For example, as mentioned above, the imaging subsystem can include more than one illumination channel. Thus, different illumination channels can be used in different modes.

[0048] The multiple modes may also differ in illumination and / or collection / detection. For example, as further described above, the imaging subsystem may include multiple detectors. Thus, one detector may be used in one mode and another detector may be used in another mode. Furthermore, the modes may differ from each other in more than one way as described herein (e.g., different modes may have one or more different illumination parameters and one or more different detection parameters). Additionally, the multiple modes may differ in viewing angle, meaning they may have any or both of different incident angles and collection angles that can be achieved as further described above. The imaging subsystem may be configured to scan the sample in different modes in the same scan or different scans, for example, depending on the ability to use multiple modes to scan the sample simultaneously.

[0049] In some examples, the systems described herein can be configured as detection systems. However, the systems described herein can be configured as another type of semiconductor-related quality control system (such as a defect inspection system and a metrology system). For example, embodiments of the imaging subsystems described herein and shown in Figures 1 and 1a can be modified in one or more parameters to provide different imaging capabilities depending on the application they will be used for. In one embodiment, the imaging subsystem is configured as an electron beam defect inspection subsystem. For example, the imaging subsystem shown in Figure 1a can be configured to have a higher resolution when it is intended for defect inspection or metrology rather than detection. In other words, the embodiments of the imaging subsystems shown in Figures 1 and 1a describe some general configurations and various configurations for an imaging subsystem that can be customized in several ways that will be obvious to those skilled in the art to produce imaging subsystems with different imaging capabilities more or less suitable for different applications.

[0050] As mentioned above, the imaging subsystem can be configured to direct energy (e.g., light, electrons) to a physical version of the sample and / or to scan the energy across that physical version of the sample, thereby generating an actual image of the physical version of the sample. In this way, the imaging subsystem can be configured as an "actual" imaging system rather than a "virtual" system. However, a storage medium (not shown) and the computer subsystem(s) 102 shown in FIG. 1 can be configured as a "virtual" system. Specifically, the storage medium and the computer subsystem(s) are not part of the imaging subsystem 100 and do not have any ability to process the physical version of the sample, but can be configured to use a virtual detector for an execution-type detection function, a virtual metrology system for an execution-type metrology function, a virtual defect inspection tool for an execution-type defect inspection function, etc., using the output of the storage detector. Systems and methods configured as "virtual" systems are described in U.S. Patent No. 8,126,255, jointly assigned to Bhaskar et al., issued February 28, 2012; U.S. Patent No. 9,222,895, issued December 29, 2015, to Duffy et al.; and U.S. Patent No. 9,816,939, issued November 14, 2017, which are incorporated herein by reference as if their entirety were described. The embodiments described herein may be further configured as described in these patents. For example, a computer subsystem described herein may be further configured as described in these patents.

[0051] The system includes: a computer subsystem that may include any configuration of any of the computer subsystems or systems described above; and one or more components, which or the like are executed by the computer subsystem. For example, as shown in FIG1, the system may include a computer subsystem 36 and one or more components 104 executed by the computer subsystem. The one or more components may be executed by the computer subsystem as further described herein or in any other suitable manner known in the art. At least a portion of executing one or more components may include inputting one or more inputs (such as images, data, etc.) into one or more components. The computer subsystem may be configured to input any images, data, etc., into one or more components in any suitable manner.

[0052] One or more components include a deep learning (DL) model 106 configured for noise removal from an image of one sample generated by an imaging subsystem. The image may include any of the images described herein generated by any of the imaging subsystems described herein. Although some embodiments are described herein as noise removal from an image, the DL model described herein can be used to remove noise from any number of images of one or more samples generated by an imaging subsystem. The imaging subsystem may include any of the imaging subsystems described herein.

[0053] Generally speaking, "deep learning" (also known as deep structured learning, hierarchical learning, or deep machine learning) is a branch of machine learning based on a set of algorithms that attempt to model higher-order abstractions in data. In a simple case, there may be two sets of neurons: one set of neurons that receive an input signal and another set of neurons that send an output signal. When an input layer receives an input, it passes a modified version of that input to the next layer. In a deep learning-based model, there are many layers between the input and output (and these layers are not composed of neurons but help to think about it in this way), thus allowing the algorithm to use multiple processing layers consisting of multiple linear and / or nonlinear transformations.

[0054] The DL model used in the embodiments described herein can be classified as a generative model. A "generative" model can be broadly defined as a model that is inherently probabilistic. In other words, a "generative" model is not a model that performs forward simulation or a rule-based approach. Instead, a generative model can be learned based on a suitable training dataset (because its parameters can be learned). In one embodiment, the DL model is configured as a deep generative model. For example, the DL model can be configured to have a DL architecture, as the DL model can contain multiple layers that perform several algorithms or transformations.

[0055] In one embodiment, the DL model is configured as a convolutional neural network (CNN). A CNN typically comprises a stack of convolutional layers and pooling layers to extract local features. The convolutional layers may have any suitable configuration known in the art and are typically configured to determine the features (i.e., a feature map) of an image that vary across its location by applying a convolutional function to the input image using one or more filters. The one or more pooling layers may also have any suitable configuration known in the art (e.g., a max pooling layer) and are typically configured to reduce the dimensionality of the feature map generated by the one or more convolutional layers while retaining the most important features. The DL model may have any suitable CNN configuration or architecture known in the art.

[0056] When used herein, "noise removal" does not refer to any reduction in noise that may occur by transforming an image for other purposes (such as transforming an image from one space (or image type) to another space (or image type) or by changing the resolution of an image). In other words, "noise removal" as described herein is not intended to change any characteristics of an image other than its noise characteristics. Noise removal does not transform any image from a relatively noisy type of image (such as a sample image) to a relatively quiet type of image (such as a design image), which requires a change in image type. Noise removal also does not transform any image from a low-resolution image to a high-resolution image, which requires a change in at least one image characteristic (resolution) other than noise. In this way, although some image transformation algorithms or methods (which may or may not be based on DL) may produce an image that can be considered to have less noise than the original image, such image transformation algorithms and methods do not preserve all the original image characteristics other than noise. In one example of this, the image from which noise has been removed by the embodiments described herein may be an optical inspection image with a resolution and pixel size generated by a BBP tool, and the resulting noise-removed image may also be an optical inspection image with the same resolution and pixel size but with less noise than the original image.

[0057] While image transformations such as those described above (e.g., changing an image from one type to another or from one resolution to another) may be useful for many applications, such transformations are not the target of the noise removal described herein. On the other hand, it is important to note that there is no reason why the noise removal described herein cannot be combined with other such image transformations. For example, an image with noise removed as described herein, when input to a model that transforms the noise-removed image into an image of a different type or with a different resolution, can provide a transformed image of better quality than if the original image were input to the transformation. In this way, the noise removal performed as described herein can be performed upstream of other image transformation procedures, where the noise-removed image is used as input to those procedures. The noise removal and information determination steps described herein can also be used in different ways with other image transformation steps. For example, the noise-removed image described herein can be used for scrambling, and further image transformations can then be performed on the defects remaining after scrambling using either the noise-removed image or the original image. In this way, the noise removal and information determination steps described herein can be integrated into a testing or other quality control procedure in various ways.

[0058] The embodiments described herein may have two steps: training and inference. In one embodiment, a computer subsystem is configured to train a DL model using a training set comprising multiple noisy implementations of the same image and an average loss function. For example, a CNN model may be used to train sampled data from a BBP thermal scan of layer A (i.e., the sample layer to be inspected). Training may be accomplished by mapping multiple noisy implementations of the same image in a CNN network. Test images and reference images (or similar to test images) may be used as noisy implementations of the same image. These images may or may not use defect detection. In other words, these images may be images in which a defect is detected by a defect detection algorithm or images in which the presence of a defect is unknown because defect detection has not yet been performed on the image. Generally, two or more (e.g., three) images generated at the same design location on a sample can be used as training inputs, and the training outputs of each of these images can be generated from the input images, for example, via perturbation, averaging, or some other image noise reduction technique known to have sufficient image noise reduction capabilities. In this way, training outputs, or noise-reduced versions of training inputs, can be generated from the training inputs using one of the methods known to provide at least some noise reduction capabilities. When used with an average loss function, the CNN model converges to an average behavior, thus learning a noise-reducing function.

[0059] For training purposes, relatively low-signal events of appropriate sample size can also be sampled across samples. In this context, "relatively low-signal events" can be broadly defined as images that are primarily noise rather than primarily defects. In other words, the image is primarily composed of relatively low-signal noise, rather than being dominated by objects that could be much larger defect signals. This sampling can be performed in various ways, including by using filtering or sampling methods based on one of the signals in the image through the computer subsystem described herein.

[0060] Figure 2 illustrates one embodiment of the steps that can be performed in a training process. As shown in step 200, the system can perform a thermal scan of a sample. Generally, a thermal scan involves scanning a sample and then applying a thermal threshold to the output generated during the scan. The thermal threshold is referred to in this way because it is at, close to, or even within the noise floor of the scan output. In this way, a thermal scan will detect many events, most of which may not be defects but rather noise or obstruction points. Therefore, thermal scans are generally not performed for production purposes in which the procedure is well understood and only specific known types of defects are of concern. Instead, thermal scans are typically used to discover what kinds of defects are on a layer and / or generate a large amount of data that can be used to set up a detection procedure. In this case, a thermal scan can generate many images that are relatively noisy and therefore suitable for training the DL model described herein. Furthermore, many images saved from a thermal scan may not be those with relatively strong defect signals (because the events detected by such a thermal scan may include many relatively low-signal events). Such images, which may contain a defect but are usually a low-signal defect, are particularly suitable for the training described herein.

[0061] While a thermal scan is a particularly easy and advantageous method for generating training images for the embodiments described herein, it is not essential. For example, if a production-type scan produces sufficiently suitable images (images consisting mainly of noise), the scan can be used to generate training images for a noise-removed DL model. Furthermore, the embodiments described herein do not necessarily require the generation of training images; instead, such training images can be acquired from a storage medium previously stored in that medium by an imaging subsystem or a different system or method. Moreover, the training images do not necessarily have to be images on which detection is performed prior to their use for training. Additionally, it may be preferable to generate training images using the same imaging subsystem parameters used in the procedure in which noise-removed images are generated and then used to determine information about the samples, but this is also not essential.

[0062] The training image can also be generated using a sample for which information is determined by using an image with noise removed by a trained DL model. However, as further described herein, in some examples, a noise-removing DL model trained on a sample having a layer formed thereon can also be used to remove noise from images of different samples that may or may not have the same layer formed thereon. In this way, the training image can be generated by, or not by, scanning an image of a physical sample with noise removed by a DL model.

[0063] At least some images for training can also be generated without an imaging subsystem. For example, an actual image generated by scanning a sample with an imaging subsystem can be used to generate artificial images, which can be used for training together with or in place of the actual images. Generating artificial images may involve perturbing the actual images or modifying the actual images with a filter, function, etc., to add artificial or different noise to the images or change other noise characteristics of the actual images. In this way, although a hot scan can be a fast, simple and reliable way to generate training images for the noise-removing DL model described herein, a hot scan is by no means the only way to generate a training set.

[0064] As shown in step 202, the computer subsystem can sample the defects detected by the thermal scan performed in step 200 for training. The number of defects sampled for training can vary considerably and may depend on the configuration of the DL model and the number of training samples typically required to train it. The sampled defect images can also be divided into different subsets for different stages of training (such as training, testing, and validation). Typically, training a DL model requires a large number of training samples, and one advantage of using a thermal scan to generate training images is that a thermal scan almost guarantees the generation of a sufficient number of samples that can be used for training.

[0065] As described above, multiple noisy implementations of the same image are preferably used for training. In this way, the training sample may include multiple images generated at corresponding locations on the sample (such as the same location in multiple grains, fields, etc. on the sample). "Multiple noisy implementations of the same image" as used herein is generally defined as at least two images generated at (or for) multiple locations on a sample printed with the same design portion (i.e., corresponding design coordinates), and the at least two images are "noisy" because they represent images generated in a procedure performed on the sample before any noise removal type function is performed on the sample. In this way, multiple noisy implementations of the same image may be generated in a thermal scan of a sample or a normal production scan of a sample. Multiple noisy implementations may also include one or more test images and / or one or more reference images, wherein such test images and reference images are defined by a procedure performed on the sample using the images, whether the procedure is a grain-to-grain type detection procedure in which the test and reference images are derived from one of the different adjacent grains on the sample or another type of detection or quality control procedure described herein.

[0066] Although multiple noisy implementations of the same image can be used to perform training, this does not mean that training can be performed on only one image of the sample. For example, training can be performed using multiple noisy implementations of a first image generated at a first design coordinate, multiple noisy implementations of a second image generated at a second design coordinate, etc. In this way, the DL model can learn to remove noise from images generated at different locations in the design of the sample.

[0067] As shown in step 204, the computer subsystem may train the DL model using two or more similar noisy implementations, as further described herein. In this step, the computer subsystem trains the DL model to learn the noise removal capabilities that will then be applied to other images, which may include images of the same sample and, as needed, images of other samples (which may or may not have the same type, as further described herein). Therefore, the result of the training performed in step 204 is a trained noise removal model 206, which can then be used by the computer subsystem to remove noise from images as described above. In other words, after training using multiple similar noisy implementations of the same image, the computer subsystem can input other sample images into the DL model, which will remove noise from those input images, which can be used to determine information as further described herein.

[0068] The DL model may be trained by or without a computer subsystem and / or by one of several components executed by the computer subsystem. For example, another method or system may train the DL model and then store it for use as one of the components executed by the computer subsystem. In either case, training may involve feeding training inputs into the DL model and changing one or more parameters of the DL model until the DL model converges to an average behavior. Training may involve changing one or more trainable parameters of the DL model. One or more trained parameters of the DL model may include one or more weights of any layer of the DL model with trainable weights. In one such example, the weights may include the weights of convolutional layers rather than the weights of pooling layers.

[0069] Once a model is trained as described above, it can be applied to images of other defects in layers on a sample on which it generates an image, or to images generated by thermal scanning of a different sample with the same layer. In other words, once a noise removal model is trained, it can be used not only for images of the sample on which the model was trained, but also for other samples on which the same layer has been formed and which have been imaged by an imaging subsystem. In one embodiment of this, multiple noisy implementations generate the same image for an additional sample, and the sample and the additional sample have different layers formed on them. For example, preliminary data generated by the inventors supports the possibility that a model trained on an image of a sample having (e.g.) a layer A formed thereon can also function on an image of a sample having (e.g.) a layer B formed thereon (i.e., adequately remove noise). In other words, even if the designs of the two samples are different, making layers A and B different, there may be sufficient similarity between the designs and layers (such as the material and size of the patterned features formed on layers A and B) so that a noise removal model trained on the image of layer A can be used to remove noise from the image of layer B. Furthermore, based on preliminary results from the inventors, a noise removal model trained on a single layer of data exhibits improved sensitivity when applied to a different layer of data without any additional training. Using a noise removal model from one layer to another for any algorithm is believed to be its first detection method in optical inspection (such as BBP detection). Even if results show that a noise removal model trained on one layer cannot adequately remove noise from the image of another layer, a noise removal model trained on layer A can be used as an initial DL model configuration for retraining on layer B in a master network or transfer learning application.

[0070] In another embodiment, the image is a test image in which a defect is detected before noise removal. In this way, the embodiment can remove noise from the image after detection. In contrast, in the process currently used by the inspection tool, there is no post-detection step of removing noise from the image. In one of this embodiments, the defect is detected by thermal scanning of a sample, which can be performed as described above. Noise removal of an image in which a defect is detected in a thermal scan of a sample can be particularly useful for a variety of reasons. For example, due to the reasons described above (i.e., substantially low threshold values ​​used for detection), a thermal scan often produces an incredible number of detected defects. Therefore, in most examples, the defect population generated by a thermal scan must be processed to filter out non-defects (disruptive points or noise) (for defect detection or inspection procedure settings) from the actual defects. By inputting an image in which a defect is detected by a scan into a noise-removing deep learning (DL) model described herein, the noise-removed image generated by the DL model can produce much more useful images for separating defects from noise or obstructions, or for determining additional information about detected defects, as further described herein. Although the noise-removed image in the embodiments described herein may include an image in which a defect is detected by a thermal scan, this is not necessary. For example, any image in which a defect is detected (regardless of how the defect was detected) can be input into the DL model described herein for noise removal.

[0071] Figure 3 illustrates one embodiment of the steps that can be performed in an inference flow using the embodiments described herein. In this embodiment, the system can perform a thermal scan, as shown in step 300, but the inference flow shown in this figure can be performed for any image generated in any type of sample scan. The output of the thermal scan may be included as a test image 302 and a reference image 304, respectively, shown in Figure 3. Both the test image and the reference image can be input to a noise removal (inference) step. For example, as shown in Figure 3, the test image 302 can be input to the noise removal step 306, and the reference image 304 can be input to the noise removal step 308. Although the test and reference image noise removal are shown as different steps in Figure 3, these steps can be performed by the same DL model configured as described herein. The test and reference images can be input to the DL model simultaneously, such as in multiple input channels, or sequentially on an image-by-image basis, depending on the configuration of the DL model. The output of noise removal step 306 is a noise-removed test image 310, and the output of noise removal step 308 is a noise-removed reference image 312. The computer subsystem can then subtract the noise-removed reference image from the noise-removed test image, as shown in step 314, to generate a noise-removed differential image 316. In this way, in some embodiments, the output generated by inference is a hot scan of one of the noise-removed test, reference, and differential images.

[0072] As shown in Figure 3, a noise-removed differential image can be generated by subtracting a noise-removed reference image from a noise-removed test image. In some examples, it is also possible or alternatively suitable to input a differential image generated by subtracting an un-noise-removed reference image from an un-noise-removed test image into a DL model for noise removal. In other words, a differential image generated from the original image can also be or alternatively input into the noise-removed DL model and used to determine information as described above.

[0073] In some embodiments, the computer subsystem is configured to detect defects on a sample based on test images of the sample generated by the imaging subsystem and to input the test images into a DL model for noise removal. The test images input into the DL model by the computer subsystem contain only those test images in which one or more defects are detected. In this way, the images input into the DL model for noise removal may not contain all images generated by a single scan. Alternatively, the noise-removed images may contain only the test images in which at least one defect is detected. In these examples, the noise-removed images may also contain reference and (possibly) differential images of those test images, rather than reference or differential images of any test images in which no defect was detected. By performing noise removal after defect detection, the number of noise-removed images can be significantly reduced, which can reduce the computation time and overhead of noise removal. The steps of this embodiment may be performed in other ways as described herein.

[0074] In a further embodiment, noise removal is performed prior to defect detection using the image. For example, contrary to some embodiments described herein where noise removal of tile images is performed after detection, noise removal using the embodiments described herein can be performed on the tile data before detection. In this way, noise removal can be performed by inputting images (and possibly all images) generated by scanning one of a sample into a DL model configured as described herein. It should also be noted that although inference noise removal can be performed at different times (before or after detection), these are not mutually exclusive. In one of these examples, image noise removal can be performed both before and after defect detection, depending on whether the additional computation time / expense is considered worthwhile, which may be advantageous in some instances.

[0075] In some embodiments, the image is a test image of one of the samples, and the DL model is configured to remove noise from a reference corresponding to the test image. For example, as shown in FIG3, both the test image and the reference image can be input into a noise removal step. These images can then be used to generate a noise-removed differential image as shown in FIG3 and / or used in any other manner described herein. The inventors believe that the embodiments described herein are first known methods for removing noise from both the test image and the reference image. For example, there are various current methods for removing noise from a reference image, which are generally generated for the purpose of attempting to generate an optimal reference for defect detection (or as a byproduct). However, due to concerns about unintentionally reducing or eliminating defect signals (which can lead to missed defects and inaccurate detection results), the test image is usually not removed from noise, especially before defect detection. Therefore, a novel feature of the embodiments described herein is believed to be that they provide a method or algorithm for removing noise from test images generated by optical inspection tools. For example, some current methods use a linear combination of images similar to the test image to generate a reference image, resulting in a reference image with less noise compared to the linearly combined images. In contrast, the embodiments described herein propose a solution using deep learning capable of learning non-linear relationships. Therefore, the embodiments described herein can generate better noise-removed images for both the test image and the reference image.

[0076] In one embodiment of this, the reference is an additional image of the sample generated by the imaging subsystem at a location on the sample corresponding to a location on the sample from which the test image was generated. For example, in many inspection procedures, a test image is compared with another image generated at a corresponding location on the sample. This other image can then be used as a reference image in inspections (such as grain-to-grain inspection, cell-to-cell inspection, etc.). Unlike other reference images not generated from the sample (e.g., a reference generated from a database), a reference image generated from a sample may have many of the same noise characteristics as the test image or may contain enough noise that could cause detection problems unless the noise is removed. This is why many different ways of generating a reference image have been created. Typically, the current method of generating a reference image requires images to be acquired at multiple locations on the same designed position on the sample. The need for multiple images for reference image generation obviously causes problems such as increased image acquisition time, computation time for generating the reference, reference storage and reusability issues, and reference image quality problems attributable to things such as inaccurate image alignment, sample variations, etc.

[0077] Unlike current methods, the embodiments described herein can advantageously generate a suitable reference image from only a single image acquired at a location corresponding to the location where the test image is generated. For example, an image from a grain, cell, etc., adjacent to the grain, cell, etc., where a test image is generated can be input into the DL model described herein, which, when properly trained, can generate a substantially noise-free reference image. Therefore, the embodiments described herein can generate suitable reference images for inspection and other quality control types of procedures much faster and more quickly than currently used methods and systems. Furthermore, since the DL model described herein can be trained to reduce any and all types of noise regardless of its characteristics (linear, nonlinear, etc.), the embodiments described herein can generate significantly noise-free reference images regardless of the noise characteristics of the original image, thereby providing greater flexibility and robustness than currently used methods and systems for reference image generation.

[0078] The computer subsystem is configured to use the image (with noise removed) to determine the sample information. The determined information and the method of using (a number of) noise-removed images for information determination may vary depending on the procedure performed on the sample. The information determination step may be performed by the computer subsystem using an algorithm, which may be part of one or more components executed by the computer subsystem or may be separate from those components.

[0079] In another embodiment, the determination information includes subtracting a noise-removed reference from a noise-removed test image to generate a differential image, and detecting defects on the sample based on this differential image. In this way, defect detection can be performed based on the differential image generated from the corresponding noise-removed test and reference images, as shown in FIG3. Therefore, this defect detection can only be performed after the test and reference images have been noise-removed. Thus, in this embodiment, noise removal can be performed before defect detection and for images where defects have not yet been checked. Performing defect detection on a noise-removed image can be more computationally intensive than performing noise removal after defect detection (where noise removal is performed only for images where at least one defect has been detected). However, in some way, the computation can be offset because the defect group detected using the noise-removed image can be much smaller than the defect group detected using the pre-noise-removed image. For example, if defect detection is performed on the image after noise removal rather than the original image, the detected defect population input to other steps such as scrambling point filtering and attribute extraction (both of which can be performed as described herein) can be much smaller, making those steps less computationally intensive.

[0080] Generally, a noise-removed image can be used for defect detection in the same manner as any other image. In other words, a noise-removed image is not specific to a defect detection algorithm or method, and defect detection using a noise-removed image can be performed using any suitable defect detection algorithm or method known in this art. In one example, defect detection may involve generating a differential image as described above and applying a threshold value to the differential image. Any pixel in the differential image that has a value higher than the threshold value can be identified as a defect, and all other pixels can be not identified as a defect. Of course, this may be the simplest way to perform defect detection and is included herein only as a non-limiting example.

[0081] Therefore, in some embodiments, the information for determining a sample may include information on defects detected on the sample. In such examples, the information may include, for example, one type of detected defect, the location of a detected defect relative to one or more of the sample image, the sample, the imaging subsystem, and a design of the sample, and any other information generated for the defect by a defect detection method or algorithm. As further described herein, this information may be output and / or stored by a computer subsystem.

[0082] In a further embodiment, a defect is detected in the test image before noise removal, and the determination information includes information from the noise-removed test image minus the noise-removed reference image to generate a reference image and information for determining the defect based on the reference image. For example, even if defect detection has been performed based on an image without noise removal, a noise-removed test and reference image can be generated as shown in FIG3, and a differential image can be generated using them. This noise-removed differential image can then be used to determine defect information, such as whether the defect is a disturbance point or an actual defect, one or more attributes of the defect, etc., which can be performed as further described herein. In this way, the difference image without noise removal can be used for defect detection, and then the difference image with noise removal can be used for additional defect functions (such as filtering, classification, attribute extraction, etc.). The difference image with noise removal can also be used in this manner when defect detection is performed using the difference image with noise removal.

[0083] In another embodiment, the information for sample determination includes one or more attributes of a defect detected in the image or the noise-removed image. For example, after inference (noise removal), a set of predetermined attributes can be calculated from at least one (or the owner) of the noise-removed test, reference, and differential images. In the current defect detection process, training, inference, and attribute calculation can each occur on the tool or analysis station after detection or offline. In other words, training the DL model for noise removal, performing noise removal, and calculating attributes from at least one of the noise-removed images can each be performed before or after detection, and on or off the tool. All three steps can occur together or separately in the order described above. Therefore, the embodiments described herein can compute attributes of the possibly improved signal-to-noise ratio (SNR) on the noise-removed image. Defect attributes may include any suitable defect attributes that can be determined from one or more of the noise-removed image and / or its alignment with other information of the sample (such as design data) as described herein, such as classification, size, shape, etc. (in addition to the reported defect location). Defect attributes can be determined by a computer subsystem using any suitable method or algorithm known in this art. These attributes can be written to batch results and can be added to or overwritten by existing attributes.

[0084] In some embodiments, a defect is detected in an image or a noise-removed image, and the information of the determined sample includes a first part of the attribute of the defect determined from the noise-removed image and a second part of the attribute of the defect determined from the image. In other words, as shown in FIG4, which is further described below, the attribute extraction or calculation step performed for any one or more defects detected on a sample can calculate only a predetermined subset of all determined defect attributes from any one or more noise-removed images that may contain any one of the noise-removed images for any defect. For example, some defect attributes may be relatively insensitive to noise, meaning that noise present in an image does not adversely affect the accuracy of those defect attributes determined from the image, or that attribute calculations performed on different examples of the same image but with different noise characteristics will produce the same attribute results. However, other defect attributes may be more sensitive to noise, meaning that the determination of those defect attributes varies with the difference in noise in the image. Therefore, depending on the sensitivity of a defect attribute to noise, the computer subsystem or a user can choose which images(s) (with or without noise removal) to use to determine the defect attribute. Thus, the proposed process for extracting defect attributes from noise-removed images, further described herein, can be executed in parallel with the current process of attribute extraction after detection using the original images.

[0085] Figure 4 illustrates one embodiment of the steps that can be performed for attribute calculation. In this process, a noise-removed test image 400, a noise-removed reference image 402, and a noise-removed differential image 404 (all of which can be generated as described herein) can be input to the attribute extraction step 406, where only a predetermined subset of all possible defective attributes is calculated. The predetermined subset of defective attributes may only include those attributes that are affected by image noise. Other defective attributes can be calculated normally from the un-noise-removed image. The result of this step may include a noise-removed background attribute 408 and a noise-removed signal attribute 410, which may be any background or signal attribute of interest that can be determined from the noise-removed image, including any of those determined in the currently used inspection procedure.

[0086] Although Figure 4 illustrates inputting noise-removed test, reference, and differential images into the attribute extraction step for any defect detected on a sample, not all such images need to be input. For example, only noise-removed test images or only noise-removed differential images can be input into the attribute extraction step. In other examples, especially when the attribute extraction step involves determining more than one defect attribute and each attribute determination may require different input images, inputting all available images into the attribute extraction step shown in Figure 4 may be most convenient.

[0087] In an additional embodiment, the information for determining a sample includes determining whether a defect detected in an image is a scrambler. Determining whether a defect detected in an image is a scrambler can be performed using either the determined attributes and / or a noise-reduced image as described herein. For example, attributes determined from any noise-reduced image as described herein (possibly in combination with attributes determined from the original image) can be input to a scrambler event filter (NEF) in the same manner as any other defect attribute. In this way, any attribute determined as described herein can be used with any currently used NEF. Preliminary studies performed by the inventors show improved scrambler suppression when using attributes determined as described herein to perform scrambler filtering, compared to determining attributes only from images that have not yet been noise-reduced as described herein. Furthermore, preliminary data indicate that the embodiments described herein outperform existing algorithms in terms of scrambler suppression.

[0088] Generally, the determination information may include generating one or more types of test results for a sample. Therefore, in essence, the determination information step may have multiple output channels, each for a different type of information. The outputs from the multiple channels can then be combined into a single test result file for the sample (e.g., a KLARF file generated by some KLA testing tools). In this way, for any given location on the sample, the test result file may contain multiple types of information.

[0089] Although Figures 2 through 4 describe and illustrate steps that can be performed for one inspection application, such similar steps can be performed in a similar manner for other applications described herein (such as metrology and defect inspection). In these examples, the input image may differ from those shown in Figures 2 through 4 and may be replaced by a suitable image generated by another imaging subsystem (such as an imaging subsystem included in a metrology or defect inspection tool). Furthermore, the information used for determining the image after noise removal may be different and may vary depending on the procedure performed on the sample. For example, instead of determining defect attributes as shown in Figure 4, the computer subsystem may be configured to perform metrology (such as patterned structure attribute calculation) or a step in performing metrology (such as patterned structure recognition) using a noise-removed test image.

[0090] Generally speaking, even if metrology and defect inspection tools produce images with significantly less noise than those produced by inspection tools, there may be instances where metrology and defect inspection procedures produce images that are noisier than usual (e.g., due to unexpected sample variations or tool drift). In this way, even though the noise removal described herein may not be commonly used in such procedures, it can be used as an alternative method when images are noisier than expected and may even help to understand why images are noisier than usual (e.g., through inspection of subsequently noise-removed images).

[0091] Unlike inspection procedures, a defect inspection procedure typically revisits discrete locations on a sample where a defect has been detected. An imaging subsystem configured for defect inspection can generate sample images as described herein, which can be input into a DL model as described herein for noise removal. The DL model can be trained and configured for image noise removal, as further described herein. For defect inspection applications, the computer subsystem can also be configured to use any suitable defect inspection method or algorithm on any suitable defect inspection tool to determine defect or sample information from (some) noise-removed images, possibly combined with any other information determined by the defect inspection procedure or from (some) noise-removed images. Although the various inputs and outputs may differ for defect inspection applications compared to inspection, the same DL model can be used for both defect inspection and inspection (after appropriate training). The DL model can be trained and configured in other ways as described above.

[0092] As described above, in some embodiments, the imaging subsystem can be configured for sample metrology. In one such embodiment, determination information includes determining one or more characteristics of a sample structure in an input image. For example, the DL model described herein can be configured to remove noise from an image of a sample generated by a metrology tool, which can then be used to determine metrology information of the sample. The metrology information may include any metrology information of interest, which may vary depending on the structure on the sample. Examples of such metrology information include (but are not limited to) critical dimensions (CD), such as linewidth and other dimensions of the sample structure. The sample image may include any image generated by any metrology tool, which may have a configuration such as that described herein or any other suitable configuration known in this art. In this way, the embodiments described herein can advantageously use a sample image generated by a metrology tool and removed from noise as described herein for predicting metrology information of the sample and any or more sample structures contained in the sample image. For metrology applications, the computer subsystem can also be configured to use any suitable metrology method or algorithm on any suitable metrology tool to determine sample information from images(s) after noise removal, possibly combined with any other information determined by the metrology procedure or from images(s) after noise removal. While the various inputs and outputs may differ for metrology applications compared to detection, the same DL model can be used for both metrology and detection (after appropriate training). The DL model can be trained and configured in other ways as described above.

[0093] The computer subsystem can also be configured to generate results containing determined information, such results may include any of the results or information described herein. The results of the determined information may be generated by the computer subsystem in any suitable manner. All embodiments described herein can be configured to store the results of one or more steps of the embodiments in a computer-readable storage medium. The results may include any of the results described herein and may be stored in any manner known in the art. The results containing determined information may have any suitable form or format (such as a standard file type). The storage medium may include any storage medium described herein or any other suitable storage medium known in the art.

[0094] After the results have been stored, they may be accessed in a storage medium and used by any of the methods or system embodiments described herein, formatted for display to a user, or used by another software module, method, or system to perform one or more functions on the sample or another sample of the same type. For example, results generated by a computer subsystem may include information on any defects detected on the sample, such as the location of the bounding box of the detected defect, detection scores, information on defect classification (such as category tags or IDs), any defect attributes determined from any image, predicted sample structural measurements, dimensions, shape, etc., or any such suitable information known in the art. This information may be used by the computer subsystem or another system or method to perform additional functions on the sample and / or the detected defects, such as sampling defects for defect inspection or other analysis, determining a root cause of the defect, etc.

[0095] These functions also include (but are not limited to) modifying a procedure (such as a process or step) that has been or will be performed on a sample in a feedback or feedforward manner. For example, the computer subsystem can be configured to determine one or more changes to a procedure that has been performed on a sample and / or will be performed on a sample based on determination information. The changes to the procedure may include any suitable changes to one or more parameters of the procedure. In one example, the computer subsystem preferably determines such changes so that defects can be reduced or prevented on other samples to which the modified procedure is performed, defects can be corrected or eliminated on the sample in another procedure performed on the sample, defects can be compensated for in another procedure performed on the sample, etc. The computer subsystem can determine such changes in any suitable manner known in the art.

[0096] These changes may then be sent to a semiconductor manufacturing system (not shown) or a storage medium (not shown) accessible to both the computer subsystem and the semiconductor manufacturing system. The semiconductor manufacturing system may or may not be part of the system embodiments described herein. For example, the imaging subsystem and / or computer subsystem described herein may be coupled to the semiconductor manufacturing system via one or more common elements (such as a housing, a power supply, a sample handling device or mechanism, etc.). The semiconductor manufacturing system may include any semiconductor manufacturing system known in the art, such as a lithography tool, an etching tool, a chemical mechanical polishing (CMP) tool, a deposition tool, and the like.

[0097] The embodiments of the various systems described above can be combined together to form a single embodiment.

[0098] Another embodiment relates to a computer-implemented method for determining information about a sample. The method includes removing noise from an image of a sample generated by an imaging subsystem by inputting the image into a DL model configured to perform noise removal. The DL model is included in one or more components executed by a computer subsystem. The method also includes determining information about the sample from the noise-removed image. These input and determination steps are performed by the computer subsystem.

[0099] The steps of the method may be performed as further described herein. The method may also include any other steps that may be performed by the system, computer subsystem, (a number of) components, and / or DL ​​model described herein. The computer subsystem may be configured according to any of the embodiments described herein (e.g., computer subsystem 36). One or more components and DL models may also be configured according to any of the embodiments described herein. The method may be performed by any of the system embodiments described herein.

[0100] An additional embodiment relates to a non-transitory computer-readable medium storing program instructions that can be executed on a computer system to perform a computer-implemented method for determining information of a sample. One such embodiment is illustrated in FIG. 5. Specifically, as shown in FIG. 5, the non-transitory computer-readable medium 500 includes program instructions 502 executable on (a number of) computer systems 504. The computer-implemented method may include (a number of) steps of (a number of) methods described herein.

[0101] Program instructions 502 for implementing methods such as those described herein may be stored on a computer-readable medium 500. The computer-readable medium may be a storage medium, such as a magnetic disk or optical disc, a magnetic tape, or any other suitable non-transitory computer-readable medium known in the art.

[0102] Program instructions can be implemented in any of the following ways, including program-based technologies, component-based technologies, and / or object-oriented technologies. For example, program instructions may be implemented using ActiveX controls, C++ objects, JavaBeans, Microsoft Foundation Classes ("MFC"), SSE (Streaming SIMD Extensions), or other technologies or methodologies as needed.

[0103] (Several) computer systems 504 may be configured according to any of the embodiments described herein.

[0104] In view of this description, those skilled in the art will understand further modifications and alternative embodiments of various aspects of the invention. For example, methods and systems for determining information about a sample are provided. Therefore, this description should be understood only as illustrative and for the purpose of teaching those skilled in the art how to practice the invention in general. It should be understood that the forms of the invention shown and described herein should be considered as the present preferred embodiments. As those skilled in the art will understand after benefiting from this description of the invention, elements and materials can be substituted for those illustrated and described herein, components and procedures can be reversed, and certain features of the invention can be utilized independently. Changes may be made to the elements described herein without departing from the spirit and scope of the invention as described in the following claims. [Simplified Explanation of the Diagram]

[0014] Those skilled in the art will appreciate further advantages of the invention upon the following detailed description of preferred embodiments and upon reference to the accompanying drawings, in which: Figures 1 and 1a are schematic side views illustrating embodiments of a system configured as described herein; Figure 2 is a flowchart illustrating steps for training a deep learning model, which may be performed by embodiments described herein; Figure 3 is a flowchart illustrating steps for generating a noise-removed image, which may be performed by embodiments described herein; Figure 4 is a flowchart illustrating steps for determining information of a sample from one or more noise-removed images, which may be performed by embodiments described herein; and Figure 5 is a block diagram illustrating an embodiment of a non-transitory computer-readable medium storing program instructions for causing a computer system to perform a computer-implemented method described herein. While various modifications and alternatives to the invention are possible, specific embodiments of the invention are shown by way of example in the drawings and are described in detail herein. The drawings may not be to scale. However, it should be understood that the detailed description of the drawings and the like is not intended to limit the invention to the specific forms disclosed. On the contrary, the invention will cover all modifications, equivalents and substitutions that fall within the spirit and scope of the invention as defined by the appended claims.

Claims

1. A system for determining information about a sample through configuration, comprising: A computer subsystem; and one or more components, or the like, executed by the computer subsystem; wherein the one or more components include a deep learning model configured to remove noise from an image of a sample generated by an imaging subsystem; and wherein the computer subsystem is configured to determine information about the sample from the noise-removed image, wherein a defect is detected in the image or the noise-removed image, and wherein the information for determining the sample includes determining only a first portion of an attribute of the defect from the noise-removed image, determining only a second portion of an attribute of the defect from the image, and determining the defect from the first portion and the second portion of the attributes.

2. The system of request item 1, wherein the deep learning model is further configured as a convolutional neural network.

3. The system of request item 1, wherein the image is a test image in which a defect was detected before the noise removal.

4. The system of claim 3, wherein the defect is detected by thermal scanning of one of the samples.

5. The system as requested in item 1, wherein the noise removal is performed before defect detection is performed using the image.

6. The system of claim 1, wherein the image is a test image of the sample, and wherein the deep learning model is further configured to remove noise from a reference corresponding to the test image.

7. The system of claim 6, wherein the reference is an additional image of the sample generated by the imaging subsystem at a location on the sample corresponding to a location on the sample from which the test image was generated.

8. The system of claim 6, wherein determining the information of the sample further includes subtracting the noise-removed reference from the noise-removed test image to generate a differential image and detecting defects on the sample based on the differential image.

9. The system of claim 6, wherein the defect is detected in the test image prior to the noise removal, and wherein the information for determining the defect further includes additional information for determining the defect by subtracting the noise-removed reference from the noise-removed test image to generate a differential image and based on the differential image.

10. The system of claim 1, wherein the computer subsystem is further configured to detect defects on the sample based on test images of the sample generated by the imaging subsystem and to input such test images into the deep learning model for noise removal, and wherein the test images input into the deep learning model by the computer subsystem include only those test images in which one or more of such defects are detected.

11. The system of request item 1, wherein the information for determining the defect includes determining whether the defect is a disturbance point.

12. The system of claim 1, wherein the computer subsystem is further configured to train the deep learning model using one of a plurality of noisy implementations including the same image and an average loss function.

13. The system of claim 12, wherein the plurality of noisy implementations of the same image are generated for an additional sample, and wherein the sample and the additional sample have different layers formed thereon.

14. The system of claim 1, wherein the imaging subsystem is a light-based imaging subsystem.

15. The system of claim 1, wherein the imaging subsystem is an electronically based imaging subsystem.

16. The system of claim 1, wherein the sample is a wafer.

17. A non-transitory computer-readable medium storing program instructions, which can be executed on a computer system to perform a computer-implemented method for determining information of a sample, wherein the computer-implemented method includes: The image of a sample generated by an imaging subsystem is input into a deep learning model configured to perform noise removal, wherein the deep learning model is included in one or more components executed by the computer system; and information about the sample is determined from the noise-removed image, wherein a defect is detected in the image or the noise-removed image, and the information about the sample is determined including determining only a first part of the attributes of the defect from the noise-removed image, determining only a second part of the attributes of the defect from the image, and determining the defect from the first part and the second part of the attributes.

18. A computer-implemented method for determining information of a sample, comprising: The image of a sample generated by an imaging subsystem is input into a deep learning model configured to perform noise removal, wherein the deep learning model is included in one or more components executed by a computer subsystem; and information about the sample is determined from the noise-removed image, wherein the input and determination are performed by the computer subsystem, wherein a defect is detected in the image or the noise-removed image, and the information about the sample includes determining only one part of the attributes of the defect from the noise-removed image, determining only one second part of the attributes of the defect from the image, and determining the defect from the first part and the second part of the attributes.

Images