Inspection of patterned features with noise

Abstract

The present invention provides a method and system for detecting defects on a sample. A system includes: an inspection subsystem configured to generate an image of a sample; and one or more computer subsystems configured to detect defect candidates on the sample. Detecting the defect candidate includes identifying a patterned feature in a test image included in the generated image of the sample. Detecting the defect candidate also includes: for at least one pixel in the test image located within the patterned feature, determining the difference between the characteristics of the at least one pixel and the characteristics of other pixels in the test image located within a predetermined window of the at least one pixel. In addition, detecting the defect candidate includes: detecting the defect candidate at the at least one pixel based on the determined difference.

Description

Technical Field

[0001] The present invention generally relates to methods and systems configured for detecting defects on samples such as redistribution layers (RDLs) on back-end packaging wafers. Background Art

[0002] The following description and examples are not admitted to be prior art by virtue of their inclusion in this section.

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

[0004] Inspection processes are used at various steps during the semiconductor manufacturing process to detect defects on the wafer to facilitate higher yields and therefore higher profits in the manufacturing process. Inspection has always been an important part of making semiconductor devices. However, as the size of semiconductor devices decreases, inspection becomes more important to successfully manufacturing acceptable semiconductor devices because even small defects can cause the device to fail.

[0005] Since the beginning of the integrated circuit (IC) industry, wafer inspection has been mainly based on image intensity differences between adjacent dies, because photolithography generally involves printing "almost identical" adjacent dies. As technology evolved, another technique called "golden reference die" or "standard reference die" was developed for inspecting dies with common die defects induced by mask defects. Whether performing die-to-die or die-to-golden die inspection, the common assumption is that every die on the wafer is almost "identical". These traditional inspection methods have been very effective in lithography-based wafer manufacturing processes.

[0006] In recent IC manufacturing processes, a new packaging process is applied, in which a final die unit is composed by connecting two or more diced and tested good sub-dies via redistribution layer (RDL) lines, such as Figure 2 In particular, the final die unit 200 is composed of sub-dies 202 and 204 connected by a plurality of RDL lines 206. In this way, the new back-end advanced wafer level packaging (aWLP) process makes a new die from the sub-dies.

[0007] Because the sub-die are mechanically placed on the die and the RDL lines are placed on top of the sub-die, there can be inconsistent sub-die shifts between adjacent die units. Because of this kind of inconsistency, even if the RDL lines in adjacent die units are perfectly aligned, traditional die-to-die difference calculations will produce a lot of noise in the inspection results. For example, if Figure 3 As shown in , adjacent constituent die units may have inconsistent shifts under RDL lines. Specifically, Figure 3 Two final die units 300 and 302 are shown. Final die unit 300 is formed of sub-dies 304 and 306 connected by RDL lines 308, and final die unit 302 is formed of sub-dies 310 and 312 connected by RDL lines 314. As more clearly shown in exploded view 316 of final die unit 300 and exploded view 318 of final die unit 302, different final die units can have inconsistent shifts between sub-dies under the RDL lines.

[0008] Furthermore, RDL lines are mainly made of metal (aluminum or copper). Defects of interest (DOIs) in RDL lines are line-open defects (disconnected lines) and partial line-open defects (also called “mouse-bite” defects), such as Figure 4 and 5 Specifically, Figure 4 A test image 400, a reference image 402, and a difference image 404 are shown for a complete line opening defect in an RDL line. As can be clearly seen in the circled portion 406 of the test image 400, the line opening defect is evident in the test image. However, as shown in the difference image 404 (generated by subtracting the reference image 402 from the test image 400), the line opening defect cannot be detected in the circled portion 408 corresponding to the circled portion 406 of the test image 400, and the portion of the difference image corresponding to the line opening defect cannot be distinguished from other portions of the difference image, such as the portion shown in the square portion 410. Therefore, the difference image 404 cannot be used to successfully detect the line opening defect.

[0009] Figure 5A reference image 500, a test image 502, and a difference image 504 are shown for another line opening defect on an RDL line. As can be seen from the test image 502, a partial line opening defect located in a circle portion 510 of the test image is apparent in the test image. However, as shown in the difference image 504 (generated by subtracting the reference image 500 from the test image 502), the partial line opening defect cannot be detected in the circle portion 514 corresponding to the circle portion 510 in the test image. In addition, the portion of the difference image corresponding to the partial line opening defect cannot be distinguished from other portions of the difference image (e.g., the portion shown in the square portion 512, which corresponds to the non-defective portion 506 of the reference image and the non-defective portion 508 of the test image). As such, the difference image 504 cannot be used to detect the partial line opening defect visible in the test image 502.

[0010] Therefore, in Figure 4 and 5 In both, it is obvious that the test image contains obvious line opening or partial line opening defects. However, even if the real defect signal exists in the test image, the surface roughness of the metal RDL line makes the inspection based on intensity difference extremely difficult, because the line opening or partial line opening signal in the difference image is completely buried in the noise generated by the random roughness in the corresponding line in the test and reference images.

[0011] Therefore, it would be advantageous to develop systems and methods for detecting defects (e.g., full and / or partial line opening defects) in RDL lines on samples (e.g., back-end aWLP samples) that do not have one or more of the disadvantages described above. Summary of the invention

[0012] The following description of various embodiments should not be viewed in any way as limiting the subject matter of the appended claims.

[0013] One embodiment relates to a system configured to detect defects on a sample. The system includes: an inspection subsystem configured to generate an image of the sample. The system also includes: one or more computer subsystems configured for detecting defect candidates on the sample. Detecting the defect candidate includes identifying a patterned feature in a test image included in the generated image of the sample. Detecting the defect candidate also includes: for at least one pixel in the test image located within the patterned feature, determining a difference between a characteristic of the at least one pixel and the characteristics of other pixels in the test image located within a predetermined window of the at least one pixel. In addition, detecting the defect candidate includes: detecting the defect candidate at the at least one pixel based on the determined difference. The system may be further configured as described herein.

[0014] Another embodiment relates to a computer-implemented method for detecting defects on a sample. The method includes: identifying, by an inspection subsystem, a patterned feature in a test image included in a generated image of the sample. The method also includes: determining, for at least one pixel in the test image that is within the patterned feature, a difference between a characteristic of the at least one pixel and characteristics of other pixels in the test image that are within a predetermined window of the at least one pixel. Additionally, the method includes: detecting a defect candidate at the at least one pixel based on the determined difference. The identifying, the determining, and the detecting are performed by one or more computer subsystems coupled to the inspection subsystem.

[0015] Each of the steps of the method described above may be further performed as described herein. In addition, the embodiments of the method described above may include any other steps of any other method described herein. In addition, the method described above may be performed by any of the systems described herein.

[0016] Another embodiment relates to a non-transitory computer-readable medium storing program instructions executable on one or more computer systems to perform a computer-implemented method for detecting defects on a sample. The computer-implemented method includes the steps of the method described above. The computer-readable medium may be further configured as described herein. The steps of the computer-implemented method may be performed as further described herein. In addition, the computer-implemented method may include any other steps of any other method described herein, and the program instructions may be executed to implement the computer-implemented method. BRIEF DESCRIPTION OF THE DRAWINGS

[0017] Other advantages of the present invention will become apparent to those skilled in the art upon having the benefit of the following detailed description of the preferred embodiments and upon reference to the accompanying drawings, in which:

[0018] Figure 1 and 1a is a schematic diagram illustrating a side view of an embodiment of a system configured as described herein;

[0019] Figure 2 is a schematic diagram illustrating a plan view of one example of a die formed of a plurality of sub-dies with redistribution layer (RDL) lines formed on top of the plurality of sub-dies;

[0020] Figure 3 is a schematic diagram illustrating a plan view of an example of a die constructed from multiple sub-dies having non-uniform shifts relative to each other beneath RDL lines formed thereon;

[0021] Figure 4 and5 Examples of test images, reference images, and difference images for different line opening defects in RDL lines are included;

[0022] Figure 6 is a schematic diagram illustrating a plan view of one example of portions of different patterned features having different orientations;

[0023] Figure 7 contains images illustrating examples of different types of defects in RDL lines;

[0024] Figure 8 is a schematic diagram illustrating a plan view of an example of a patterned feature having a predetermined window overlaid thereon within which differences may be determined as described herein and used to detect defect candidates in the patterned feature;

[0025] Fig. 9 and 10 includes examples of test images for different defects that can be detected as described herein;

[0026] Fig.11 Examples of test images including different patterned features and embodiments of predetermined windows overlaid on the different patterned features, the predetermined windows having different characteristics depending on the orientation of the different patterned features;

[0027] Figures 12 to 14 is a flow chart illustrating an embodiment of steps that may be performed by embodiments described herein for detecting defects on a sample; and

[0028] Fig.15 is a block diagram illustrating one embodiment of a non-transitory computer-readable medium storing program instructions for causing one or more computer systems to perform the computer-implemented methods described herein.

[0029] Although the invention is susceptible to various modifications and alternative forms, specific embodiments thereof are shown in the drawings by way of example and will be described in detail herein. The drawings may not be drawn to scale. However, it should be understood that the drawings and detailed description of the drawings are not intended to limit the invention to the specific forms disclosed, but on the contrary, the invention is intended to cover all modifications, equivalent forms and alternative forms within the spirit and scope of the invention as defined by the appended claims. DETAILED DESCRIPTION

[0030] Turning now to the drawings, it should be noted that the figures are not drawn to scale. In particular, the proportions of some of the elements of the figures are greatly exaggerated to emphasize the characteristics of the elements. It should also be noted that the figures are not drawn to the same scale. The same reference numerals have been used to indicate elements shown in more than one figure that may be similarly configured. Unless otherwise mentioned herein, any of the elements described and shown may include any suitable commercially available elements.

[0031] As used herein, the terms "design" and "design data" generally refer to the physical design (layout) of an IC and data derived from the physical design by complex simulations or simple geometric and Boolean operations. The physical design may be stored in a data structure such as a Graphical Data Stream (GDS) file, any other standard machine-readable file, any other suitable file known in the art, and a design database. A GDSII file is one of a class of files used for representation of design layout data. Other examples of such files include GL1 and OASIS files and proprietary file formats such as RDF data proprietary to KLA Corp., Milpitas, Calif. The design may include any other design data or design data proxy described in commonly owned U.S. Patent No. 7,570,796 issued to Zafar et al. on August 4, 2009 and commonly owned U.S. Patent No. 7,676,077 issued to Kulkarni et al. on March 9, 2010, both of which are incorporated by reference as if fully set forth herein. In addition, the design data may be standard cell library data, integrated layout data, design data for one or more layers, derivatives of design data, and all or part of the chip design data.

[0032] "Design" and "design data" described herein also refer to information and data generated by semiconductor device designers in the design process and thus can be used in the embodiments described herein well before the design is printed on any physical wafer. A "design" or "physical design" can also be a design that will ideally be formed on a wafer.

[0033] As the term is used herein, "Nuisance" is a defect that the user does not care about and / or an event that is detected by inspection but is not actually a defect. Nuisances that are detected as events (or "defect candidates") but are not actually defects may be detected due to non-defect noise sources on the sample (e.g., line edge roughness (LER), relatively small critical dimension (CD) variations in patterned features, thickness variations, metal grain noise, etc.) and / or due to marginalities of the inspection subsystem itself or its configuration used for inspection. Therefore, in general, the goal of inspection is not to detect nuisances on a sample (e.g., a wafer).

[0034] The term "defect of interest (DOI)" as used herein is defined as a defect that is detected on a sample and is actually an actual defect on the sample. Therefore, DOI is of interest to users because users generally care about how many and what kind of actual defects there are on the sample being inspected. In some contexts, the term "DOI" is used to refer to a subset of all actual defects on a sample, the subset containing only actual defects of concern to the user. For example, there may be multiple types of DOIs on any given sample, and one or more of the DOIs may be of greater concern to the 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 real defects on a sample.

[0035] One embodiment relates to a system configured to detect defects on a sample. In one embodiment, the sample is a wafer. The wafer may include any wafer known in semiconductor technology. Although some embodiments may be described herein with respect to one or several wafers, the embodiments are not limited to the samples they may be used for. For example, the embodiments described herein may be used for samples such as photomasks, flat panels, personal computer (PC) boards, and other semiconductor samples.

[0036] In one embodiment, the sample includes a metal upper layer. In one such embodiment, the upper layer may be an aluminum layer. The aluminum layer may include any suitable aluminum layer known in the art. The upper layer of the sample may also include a back end of line (BEOL) layer, which may include any BEOL layer known in the art, including the BEOL layers described herein. In another embodiment, the upper layer of the sample may be a redistribution layer (RDL) which may have any suitable configuration known in the art.

[0037] The upper layers of the sample may also include metal lines. For example, BEOL and RDL layers may include metal lines that form various elements of the device being formed on the sample. Such metal lines may generate a significant amount of "grain" noise, which is further described herein. As the term is used herein, "grain" refers to a dislocation in the crystal structure of a metal, such as aluminum or copper. The embodiments described herein are configured for enabling detection of defects on such layers despite the presence of grain noise.

[0038] In some embodiments, the sample is a post-dice sample. A "post-dice" sample may generally be defined as a wafer or other substrate on which multiple devices have been formed (e.g., in different dies or dies) and then separated from each other in one of a variety of ways. A "post-dice" sample may also be a sample that has been separated into multiple dies or dies that have not yet entered a packaging process.

[0039] In another embodiment, the sample includes a high noise layer. As that term is defined herein, a "high noise" layer generally refers to a layer whose noise is the main obstacle in the inspection of the layer. For example, although each wafer layer inspected by any inspection tool may exhibit more or less noise than other layers (and the technology for handling the detection of such noise must generally be used in the inspection of each wafer layer), the main obstacle in successfully inspecting a wafer layer is most often the extremely small size of the defects that must be detected. In contrast, the embodiments described herein are particularly suitable for detecting relatively large ("macro") defects of about 200nm and above in size. Therefore, the main obstacle in this inspection is not necessarily the size of the defects that must be detected (this is because many inspection tool configurations are capable of detecting such large defects on most layers). Instead, the layers described herein will generally exhibit such "high noise" levels in the images generated for the layers that it may become difficult (if not impossible) to detect even such large-sized defects. However, the embodiments described herein have been designed to handle such noise levels via the defect detection described herein.

[0040] The embodiments described herein are designed to be particularly effective for detecting such defects for several different reasons. For example, in conventional front-end lithography wafer fabrication processes, wafer inspection is typically performed by image comparison between adjacent dies due to the valid assumption of identical images from adjacent dies. With reasonable pre-processing steps (e.g., adjacent die image alignment and gray uniformity scaling by histogram remapping), differences greater than a threshold are typically caused by anomalies and therefore possible defects.

[0041] In the new advanced wafer level packaging (aWLP) technology, the composed die units can be made by connecting the sub-dies (which have been tested as good) via RDL lines. Figure 2 An example of such a composed die unit is shown in FIG. The RDL line and the sub-die below the RDL line may have an inconsistent relative spatial shift, as discussed further above. Figure 3 Even if alignment between adjacent dies is performed with respect to the RDL lines, the sub-die structures under the RDL lines may still be misaligned. This misalignment makes the conventional die-to-die difference-based approach substantially difficult due to misalignment noise from the structures under the RDL lines.

[0042] Additionally, since RDL lines are typically metal (copper), the random optical appearance induced by the surface roughness of the RDL line image can make intensity difference based RDL line open defect detection more challenging because the line open defect signal can be completely buried in the surface roughness induced noise. In one such example, a large amount of noise can be detected by inspection due to within the RDL metal line noise that can be caused by excessive metal particles. In another such example, a large amount of noise can be detected by inspection due to RDL metal inter-layer noise caused by transparent dielectric polymers on or below the RDL layer.

[0043] However, the embodiments described herein provide RDL line inspection that is not hampered by severe die-to-die difference noise. Additionally, the embodiments described herein may perform test image only (i.e., not die-to-die) inspection for the RDL lines. In this way, the embodiments described herein may be a single die inspection (SDI) or single image detection (SID) system and method for detecting such defects, which will eliminate the die-to-die noise source in the inspection.

[0044] exist Figure 1 One embodiment of a system configured to detect defects on a sample is shown in FIG. The system includes an inspection subsystem 10 and one or more computer subsystems (e.g., computer subsystem 36 and computer subsystem 102). Inspection subsystem 10 is configured to generate an output (e.g., an image) of the sample. In one embodiment, the inspection subsystem is configured as an optical subsystem. For example, in Figure 1 In an embodiment of the present invention, the inspection subsystem is configured to scan light against or direct light to a physical version of the sample while detecting light from the sample to thereby produce an output of the sample. The inspection subsystem may also be configured to utilize multiple modes to perform scanning (or directing) and detection.

[0045] In one embodiment, the inspection subsystem is configured for macro inspection. Therefore, the system described herein may be referred to as a macro inspection tool. Macro inspection tools are particularly suitable for inspection of relatively noisy BEOL layers (e.g., RDL) and post-die applications to detect defects in the presence of huge noise (e.g., particles on metal lines). Macro inspection tools are defined herein as systems that are not necessarily diffraction-limited and have a spatial resolution of about 200nm to about 2.0 microns and above. This spatial resolution means that the smallest defect that can be detected by such a system has a size greater than about 200nm, which is much larger than the smallest defect that can be detected by the most advanced inspection tools on the market today, hence the name "macro" inspector. Such systems tend to utilize longer wavelengths of light (e.g., about 500nm to about 700nm) compared to the most advanced inspection tools on the market today. These systems can be used when the DOI is relatively large and may also be when a throughput of 100 wafers / hour (wph) or more is required (wafer throughput here refers to the number of 300mm wafers inspected per hour).

[0046] exist Figure 1 In the embodiment of the system shown in FIG. 1 , the inspection subsystem 10 includes an illumination subsystem configured to direct light to the sample 14. The illumination subsystem includes at least one light source. For example, Figure 1 , the illumination subsystem includes a light source 16. The illumination subsystem may be configured to direct light to the sample at one or more incident angles (which may include one or more oblique angles and / or one or more normal angles). Figure 1 , light from light source 16 is directed through optical element 18 and then lens 20 at an oblique angle of incidence to sample 14. The oblique angle of incidence may include any suitable oblique angle of incidence, which may vary depending on, for example, the characteristics of the sample.

[0047] The inspection subsystem may be configured to direct light to the sample at different incident angles at different times. For example, the inspection subsystem may be configured to change one or more characteristics of one or more elements of the illumination subsystem so that the light can be directed at different angles. Figure 1 In one such example, the inspection subsystem may be configured to move light source 16, optical element 18, and lens 20 so that light is directed to the sample at different oblique angles of incidence or normal (or near-normal) angles of incidence.

[0048] The inspection subsystem may be configured to direct light to the sample at more than one incident angle simultaneously. For example, the illumination subsystem may include more than one illumination channel, one of which may include a plurality of illumination channels such as a plurality of illumination channels, and a plurality of illumination channels such as a plurality of illumination channels. Figure 1, and another of the illumination channels (not shown) may include similar elements that may be configured in a different or identical manner, or may include at least a light source and possibly one or more other components (such as those described further 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 angles of incidence may be different, so that light resulting from illumination of the sample at different angles of incidence may be distinguished from each other at a detector.

[0049] In another example, the illumination subsystem may include only one light source (e.g., Figure 1 ), and the light from the light source can be separated into different optical paths (e.g., based on wavelength, polarization, etc.) by one or more optical elements (not shown) of the illumination subsystem. The light in each of the different optical paths can then be directed to the sample. Multiple illumination channels can 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 can be configured to direct light with different characteristics to the sample at different times. For example, in some examples, the optical element 18 can be configured as a spectral filter, and the properties of the spectral filter can be changed in a variety of different ways (e.g., by replacing the spectral filter) so that different wavelengths of light can be directed to the sample at different times. The illumination subsystem can have any other suitable configuration known in the art for directing light with different or the same characteristics to the sample sequentially or simultaneously at different or the same angles of incidence.

[0050] In one embodiment, the light source 16 may include a broadband plasma (BBP) light source. In this way, the light generated by the light source and directed to the sample may include broadband light. However, the light source may include any other suitable light source, such as a laser. The laser may include any suitable laser known in the art, and may be configured to generate light at any one or several suitable wavelengths known in the art. In addition, the laser may be configured to generate monochromatic or nearly monochromatic light. In this way, the laser may be a narrow-band laser. The light source may also include a polychromatic light source that generates light at multiple discrete wavelengths or bands.

[0051] Light from optical element 18 may be focused onto sample 14 by lens 20. Figure 1 In FIG. 2 , lens 20 is shown as a single refractive optical element, but in practice, lens 20 may include several refractive and / or reflective optical elements that in combination focus light from the optical element onto the sample. Figure 1The illumination subsystem shown in and described herein may include any other suitable optical elements (not shown). Examples of such optical elements include, but are not limited to, polarizing components, spectral filters, spatial filters, reflective optical elements, apodizers, beam splitters, stops, and the like, which may include any such suitable optical elements known in the art. Additionally, the inspection subsystem may be configured to alter one or more of the elements of the illumination subsystem based on the type of illumination to be used for inspection.

[0052] The inspection subsystem may also include a scanning subsystem configured to cause light to scan the sample. For example, the inspection subsystem may include a stage 22 on which the sample 14 is disposed during inspection. The scanning subsystem may include any suitable mechanical and / or robotic assembly (including the stage 22) that may be configured to move the sample so that light can scan the sample. Additionally or alternatively, the inspection subsystem may be configured so that one or more optical elements of the inspection subsystem perform some scanning of the sample by light. The light may scan the sample in any suitable manner, such as in a serpentine path or in a spiral path.

[0053] The inspection subsystem further includes one or more detection channels. At least one of the one or more detection channels includes a detector configured to detect light from the sample due to illumination of the sample and to generate an output in response to the detected light. For example, Figure 1 The inspection subsystem shown in FIG. 1 includes two detection channels, one detection channel formed by a collector 24, an element 26 and a detector 28 and the other detection channel formed by a collector 30, an element 32 and a detector 34. Figure 1 , 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 the 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., specularly reflected light).

[0054] like Figure 1 As further shown in , the two detection channels are shown as being positioned in the plane of the paper and the illumination subsystem is also shown as being positioned in the plane of the paper. Thus, in this embodiment, the two detection channels are positioned in (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 collector 30, the element 32, and the detector 34 may be configured to collect and detect light scattered from the plane of incidence. Thus, this detection channel may be generally referred to as a "side" channel, and this side channel may be centered in a plane that is substantially perpendicular to the plane of incidence.

[0055] Although Figure 1An embodiment of an inspection subsystem including two detection channels is shown, but the inspection subsystem may include a different number of detection channels (e.g., only one detection channel or two or more detection channels). In one such example, the detection channel formed by the collector 30, element 32, and detector 34 may form a side channel as described above, and the inspection subsystem may include an additional detection channel (not shown) formed as another side channel positioned on the opposite side of the incident plane. Thus, the inspection subsystem may include a detection channel that includes the collector 24, element 26, and detector 28 and is centered in the incident plane and is configured to collect and detect light at a scattering angle normal to or close to normal to the sample surface. This detection channel may therefore be generally referred to as a "top" channel, and the inspection subsystem may also include two or more side channels configured as described above. In this way, the inspection subsystem can 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 collector, each of the collectors being configured to collect light at a different scattering angle than each of the other collectors.

[0056] As further described above, each of the detection channels included in the inspection subsystem can be configured to detect scattered light. Figure 1 The inspection subsystem shown in can be configured for dark field (DF) inspection of the sample. However, the inspection subsystem may also or alternatively include a detection channel configured for bright field (BF) inspection of the sample. In other words, the inspection subsystem may include at least one detection channel configured to detect light specularly reflected from the sample. Thus, the inspection subsystem described herein may be configured for DF only, BF only, or both DF and BF imaging. Although in Figure 1 Each of the collectors is shown in as a single refractive optical element, but each of the collectors may include one or more refractive optical elements and / or one or more reflective optical elements.

[0057] One or more detection channels may include any suitable detector known in the art. For example, the detector may include a photomultiplier tube (PMT), a charge coupled device (CCD), a time delay integration (TDI) camera, and any other suitable detector known in the art. The detector may also include a non-imaging detector or an imaging detector. If the detector is a non-imaging detector, each of the detectors may be configured to detect a specific characteristic (such as intensity) of the scattered light, but may not be configured to detect such characteristics based on the position in the imaging plane. In this way, the output generated by each of the detectors in each of the detection channels included in the inspection subsystem may be a signal or data, but not an image signal or image data. In such 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, which is configured to generate an image signal or image data. Therefore, the inspection subsystem may be configured to generate an image in several ways.

[0058] Note that in this article Figure 1 The configuration of an inspection subsystem is generally illustrated, which may be included in the system embodiments described herein or may produce outputs used by the system embodiments described herein. Obviously, the inspection subsystem configuration described herein may be changed to optimize the performance of the inspection subsystem, as is typically performed when designing a commercial inspection system. In addition, the system described herein may be implemented using existing systems such as the Altair series of tools commercially available from KLA (e.g., by adding the functionality described herein to the existing system). For some such systems, the embodiments described herein may be provided as optional functionality of the system (e.g., in addition to other functionality of the system). Alternatively, the inspection subsystem described herein may be designed "from scratch" to provide an entirely new inspection subsystem. The inspection subsystem may be further configured as described in U.S. Patent No. 7,782,452, issued to Mehanian et al. on August 24, 2010 (which is generally incorporated by reference as if fully set forth herein).

[0059] The computer subsystem 36 may be coupled to the detector of the inspection subsystem in any suitable manner (e.g., via one or more transmission media, which may include "wired" and / or "wireless" transmission media) so that the computer subsystem can receive the output generated by the detector during scanning of the sample. The computer subsystem 36 may be configured to use the output of the detector to perform a number of functions further described herein.

[0060] Figure 1The computer subsystem shown in (and other computer subsystems described herein) may also be referred to herein as a computer system. Each of the computer subsystems or systems described herein may be in various forms, including a personal computer system, an image computer, a mainframe computer system, a workstation, a network appliance, an Internet appliance, or other devices. In general, 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 or system may also include any suitable processor known in the art, such as a parallel processor. In addition, the computer subsystem or system may include a computer platform with high-speed processing and software as a stand-alone tool or a network-connected tool.

[0061] If the system includes more than one computer subsystem, the different computer subsystems may be coupled to each other so that images, data, information, instructions, etc. may be transmitted between the computer subsystems. For example, computer subsystem 36 may be coupled to computer subsystem 102 (e.g., a computer subsystem 104) via any suitable transmission medium. Figure 1 The transmission medium may include any suitable wired and / or wireless transmission medium known in the art. Two or more of such computer subsystems may also be effectively coupled via a shared computer-readable storage medium (not shown).

[0062] Although the inspection subsystem is described above as being an optical or light-based inspection subsystem, in some embodiments, the inspection subsystem is configured as an electron beam subsystem. Figure 1a In one such embodiment shown in FIG. 1 , the inspection subsystem includes an electronic column 122 coupled to a computer subsystem 124. Figure 1a , the electron column includes an electron beam source 126 configured to generate electrons that are focused by one or more elements 130 onto a sample 128. The electron beam source may include, for example, a cathode source or 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 selecting aperture, an objective lens, and a scanning subsystem, all of which may include any such suitable elements known in the art.

[0063] Electrons returning from the sample (eg, secondary electrons) may be focused to a 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 element 130.

[0064] The electron column may include any other suitable elements known in the art. In addition, the electron column may be further configured as described in the following U.S. Patents: U.S. Patent No. 8,664,594 issued to Jiang et al. on April 4, 2014, U.S. Patent No. 8,692,204 issued to Kojima et al. on April 8, 2014, U.S. Patent No. 8,698,093 issued to Gubbens et al. on April 15, 2014, and U.S. Patent No. 8,716,662 issued to MacDonald et al. on May 6, 2014, which are generally incorporated by reference as if fully set forth herein.

[0065] Although in Figure 1a , the electron beam is shown as being configured so that electrons are directed to the sample at an oblique angle of incidence and scattered from the sample at another oblique angle, but the electron beam can be directed to the sample and scattered from the sample at any suitable angle. In addition, the electron beam subsystem can be configured to use multiple modes to generate images of the sample as further described herein (e.g., at different illumination angles, collection angles, etc.). The multiple modes of the electron beam subsystem can differ in any image generation parameter of the inspection subsystem.

[0066] The computer subsystem 124 may be coupled to a detector 134, as described above. The detector may detect electrons returning from the surface of the sample, thereby forming an electron beam image 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 perform one or more functions further described herein with respect to the sample. The computer subsystem 124 may be configured to perform any additional steps described herein. Figure 1a The system of inspection subsystems shown in may be further configured as described herein.

[0067] Note that in this article Figure 1a The configuration of the electron beam subsystem that may be included in the embodiments described herein is generally illustrated. As with the optical subsystem described above, the electron beam subsystem configuration described herein may be altered to optimize the performance of the inspection subsystem, as is typically performed when designing a commercial inspection system. In addition, the systems described herein may be implemented using existing systems such as the eSxxx and eDR-xxxx series tools commercially available from KLA (e.g., by adding the functionality described herein to the existing system). For some such systems, the embodiments described herein may be provided as optional functionality of the system (e.g., in addition to other functionality of the system). Alternatively, the systems described herein may be designed "from scratch" to provide an entirely new system.

[0068] Although the inspection subsystem is described above as an optical or electron beam subsystem, the inspection subsystem may be an ion beam subsystem. Figure 1a , except that the electron beam source can be replaced with any suitable ion beam source known in the art. In addition, the inspection subsystem can be any other suitable ion beam subsystem, such as ion beam subsystems included in commercially available focused ion beam (FIB) systems, helium ion microscope (HIM) systems, and secondary ion mass spectrometry (SIMS) systems.

[0069] As described above, the inspection subsystem is configured for causing energy (e.g., light, electrons, etc.) to scan a physical version of the sample, thereby generating an output of the physical version of the sample. In this way, the inspection subsystem can be configured as a "real" subsystem, rather than a "virtual" subsystem. However, the storage medium (not shown) and the computer subsystem 102 shown in 1 can be configured as a "virtual" inspection subsystem configured to generate an image of the sample. In particular, the storage medium and the computer subsystem can be configured as a "virtual" inspection system as described in commonly assigned U.S. Patent Nos. 8,126,255, issued to Bhaskar et al. on February 28, 2012, and 9,222,895, issued to Duffy et al. on December 29, 2015 (both of which are generally incorporated by reference as if fully set forth herein) and used as an inspection subsystem in the embodiments described herein. The embodiments described herein can be further configured as described in such patents.

[0070] As further mentioned above, the inspection subsystem may be configured to utilize multiple modes to generate outputs for the sample. In general, a "mode" may be defined by parameter values ​​of the inspection subsystem used to generate outputs (e.g., images) for the sample. Thus, different modes are defined by different values ​​of at least one parameter of the inspection subsystem (in addition to the position on the sample at which the output is generated). For example, in an optical inspection subsystem, different modes may use at least one different wavelength of light for illumination. Modes may differ in illumination wavelength, as described in detail herein (e.g., by using different light sources, different spectral filters, etc. for different modes). In another example, different modes may use different illumination channels of the inspection subsystem. For example, as described above, the inspection subsystem may include more than one illumination channel. Thus, different illumination channels may be used for different modes. Modes may also or alternatively differ in one or more collection / detection parameters of the inspection subsystem. The inspection subsystem may be configured to utilize different modes to scan the sample in the same scan or different scans, for example, depending on the ability to scan the sample using multiple modes simultaneously.

[0071] As described above, the traditional defect detection algorithm based purely on die-to-die strength difference Figures 2 to 5 The main resistance is the excessive nuisance induced by random metal line surface roughness and / or inconsistent spatial displacement, that is, although the difference-based signal can be detected, the noise in the difference signal can be larger than the real line opening defect in terms of grayscale intensity and geometric size. To address the line opening defect, the embodiments described herein are specialized inspection methods and systems for detecting line opening defects on RDL lines that can be performed in a single die based inspection manner.

[0072] Although the embodiments described herein are particularly suitable for detecting line opening defects in RDL lines and some embodiments are described herein with respect to such defects and samples, the embodiments described herein are not limited to such defects and samples. For example, the embodiments described herein may be advantageously used to inspect defects in other relatively high noise patterned features and / or situations where the defects and patterned features have substantially different characteristics in the inspection image.

[0073] The computer subsystem is configured for detecting defect candidates on a sample. As the term is used herein, a "defect candidate" is defined as any event detected on a sample and not filtered out as nuisance. For example, after defect detection, the defect candidates may include all events detected on the sample and may include both actual defects (or DOIs) and nuisances. After nuisance filtering, the defect candidates will include all detected defect candidates that were not eliminated by nuisance filtering. Whether the defect candidates after nuisance filtering include only DOIs depends on how well the nuisance filtering distinguishes between DOIs and nuisances. After performing all nuisance filtering, the remaining defect candidates may be designated as detected defects, even if one or more of the designated defects are actually nuisances.

[0074] Detecting defect candidates includes identifying patterned features in a test image included in a generated image of the sample. The test image may be any test, target, or frame image generated by the inspection subsystem. The test image may be of any suitable size and may be a relatively small patch image, the size of which may vary depending on the configuration of the inspection subsystem and / or computer processing of the output generated by the inspection subsystem.

[0075] In one embodiment, the patterned features include lines on an RDL layer. In another embodiment, the patterned features are formed of metal. For example, as further described herein, a sample for which the embodiments described herein may detect defects may include RDL lines formed of metal. As such, the computer subsystem may be configured to identify the RDL lines in the test image so that the RDL lines may be inspected for line open defects, as further described herein. In a similar manner, the computer subsystem may be configured to identify other patterned features formed of metal for inspection.

[0076] In some embodiments, identifying a patterned feature includes applying a mask to a test image, the mask separating pixels in the test image within the patterned feature from all other pixels in the test image. In other words, identifying a patterned feature in a test image may include identifying pixels in the test image corresponding to the patterned feature. A particularly suitable way to perform this identification is by applying a mask to the test image. The mask may have clearly defined and designated areas corresponding to and not corresponding to the patterned feature. For example, the mask may include a clear area corresponding to the identified patterned feature and an opaque area that does not correspond to the patterned feature (the opaque area may include a patterned feature that is not of interest, an underlying or non-patterned area, etc.). In general, the mask may have any suitable configuration known in the art. Applying the mask to the test image may include overlaying the mask on the test image and then aligning the test image with the mask. This overlay and alignment may be performed in any suitable manner known in the art.

[0077] Although the identification step is described herein with respect to patterned features, the identification step may actually include identifying more than one patterned feature of interest in the same test image, which may have the same characteristics or different characteristics (e.g., two or more RDL lines having the same characteristics, two or more RDL lines at least some of which have different shapes, orientations, sizes, etc. from each other). In other words, the identification step is described with respect to patterned features to simplify and clarify the understanding of the present invention. However, the identification step may be performed for more than one patterned feature simultaneously or sequentially. In a similar manner, although the present invention may be described herein with respect to a test image for clarity and simplicity, the steps described herein may be performed sequentially or simultaneously for more than one test image. The test image may include a test image generated at the same intra-die location in one or more dies on the sample or at different intra-die locations in one or more dies on the sample, which means that the test image may not contain patterned features that are identical to each other. The test image may further include any of the test images described above and generated using one or more modes of the inspection subsystem.

[0078] In one such embodiment, one or more computer subsystems are configured to generate masks according to the design of the sample. In this way, the embodiments described herein can utilize design information for RDL line mask generation. For example, since the Kronos 3D printer commercially available from KLA TM 1080 wafer-level packaging inspection system, so FlexPoint TM Algorithm architecture (which is introduced from the Altair product line to Kronos TM 1080 product for adaptive inspection feature) provides median intensity based segmentation for separating RDL lines and spaces between lines. TM The sensitivity-nuisance tradeoff for the Altair inspection tool is greatly improved, especially in spatial regions where the image intensity is typically relatively quiet, so that bridges between RDL lines are easily detected. However, the inventors have also found that it can be difficult to produce clear line masks due to RDL LER. Some 89xx tools commercially available from KLA will include a feature that allows the user to generate FlexPoint based on design information (e.g., GDS files) that can be provided by the user. TM Features of masks (for RDL lines / spaces). Such masks are advantageously faster, cleaner, and less affected by RDL LER than masks generated from sample images. TM Mask, in addition to bridges in the RDL space, RDL line openings can also be detected by the embodiments described herein. FlexPoint using design information TM Mask generation may be performed as described in commonly assigned U.S. Patent Application No. 16 / 901,102 filed by Zhang et al. on June 15, 2020 (which is incorporated by reference as if fully set forth herein), and the features will be implemented as part of DesignWise TM Features may be used on 89xx tools commercially available from KLA.The embodiments described herein may be further configured as described in this patent application.

[0079] For handling detection of line opening defects in RDL lines, embodiments described herein may benefit from the relatively quick and easy generation of design-based RDL line and space masks. TM An immediate advantage of mask generation is that embodiments do not rely on pure total image intensity to perform patch-to-mask alignment (PMA), which is a FlexPoint TM The components in the features used to accurately place the mask on the image. Therefore, the displacement from the RDL to the previous layer (such as Figure 3Blurring or confusion of the image (as shown in ) will not affect the alignment process.

[0080] In another such embodiment, the mask separates pixels within portions of the patterned features having different orientations into different regions, and the determination and detection steps described further herein are performed separately for the different regions. For example, RDL line portions having different orientations can be easily assigned different regional mask IDs, so different inspection algorithms and associated parameters can be applied. The designed RDL line / space mask can be overlaid on top of a median reference die (MRD) image that can be generated in any suitable manner known in the art. Different colors or other suitable markings can be used to represent different regional IDs. Pixels that do not have a colored mask or other marking may belong to uninspected regions, i.e., regions of no interest or areas of no concern.

[0081] like Figure 6 As shown in , with design information, in addition to the spaces between RDL lines, continuous RDL lines can be easily separated into different regions with different orientations (e.g., horizontal, vertical, 45 degree diagonal, and 135 degree diagonal). Figure 6 Examples of different patterned features 600, 602, 604, and 606 are shown, where portions of the patterned features having different orientations are separated into different regions. Patterned feature 600 includes portions 608, 612, and 616 having a horizontal orientation, portion 610 having a 45 degree diagonal orientation, and portion 614 having a 135 degree diagonal portion. Patterned feature 602 includes portions 618, 622, and 626 having a horizontal orientation, portion 620 having a 45 degree diagonal orientation, and portion 624 having a 135 degree diagonal orientation. Patterned feature 604 includes portions 628, 632, and 636 having a vertical orientation, portion 630 having a 135 degree diagonal orientation, and portion 634 having a 45 degree diagonal orientation. Patterned features 606 include portions 638, 642, and 646 having a vertical orientation, a portion 640 having a 135 degree diagonal orientation, and a portion 644 having a 45 degree diagonal orientation.

[0082] Portions of patterned features having the same orientation may be separated into the same region, and portions of patterned features having different orientations may be separated into different regions. In one such example, pixels in portions 608, 612, and 616 of patterned feature 600 may be separated into a first region, pixels in portion 610 of patterned feature 600 may be separated into a second region, and pixels in portion 614 of patterned feature 600 may be separated into a third region. Figure 6Pixels in portions of other patterned features shown in , where regions are generated separately for each patterned feature.

[0083] Pixels in portions of multiple patterned features having the same orientation may be separated into the same region. For example, pixels in portions 608, 612, and 616 of patterned feature 600 and pixels in portions 618, 622, and 626 of patterned feature 602 may be separated into a first region. Pixels in portion 610 of patterned feature 600, pixels in portion 620 of patterned feature 602, pixels in portion 634 of patterned feature 604, and pixels in portion 644 of patterned feature 606 may be separated into a second region. Pixels in portion 614 of patterned feature 600, pixels in portion 624 of patterned feature 602, pixels in portion 630 of patterned feature 604, and pixels in portion 640 of patterned feature 606 may be separated into a third region. Pixels in portions 628, 632, and 636 of patterned feature 604 and pixels in portions 638, 642, and 646 of patterned feature 606 can be separated into a fourth region.

[0084] Different region IDs can be defined for different line segments of the entire RDL line, and can be defined by Figure 6 The different region IDs may be represented by different fill patterns as shown in or in another manner (e.g., using different mask colors). One or more computer subsystems may utilize DesignWise TM Features are easily created and use masks as described herein. Setup of the mask can be performed during recipe setup time.

[0085] For at least one pixel in the test image that is located within a patterned feature, detecting a defect candidate includes determining a difference between characteristics of the at least one pixel and characteristics of other pixels in the test image that are located within a predetermined window of the at least one pixel and detecting the defect candidate at the at least one pixel based on the determined difference. These steps explore and utilize structural continuity / discontinuity in the RDL lines / spaces and RDL line / space specific noise statistics to detect defects such as line opening defects. The goal in these steps can be to detect line opening defects with the aid of an RDL line mask as described above. As described above, due to the RDL line roughness, the difference intensity from die to die or other test to reference comparisons will produce excessive noise. However, by looking at the test image itself, it is obvious that a line opening defect signal exists. For example, Figure 7Test images are included that clearly show line opening defects. In particular, test image 700 shows a relatively small line opening defect 702. Test image 704 shows a faded line defect 706 and a relatively large missing line defect 708 (in both the diagonal segment of the line and another horizontal segment of the line) that are not visible in reference image 710. The roughness of the lines in these images is also very clear.

[0086] The embodiments described herein use the properties of structural continuity / discontinuity to detect line open defects. There are two observations of the RDL line structure related to the steps described herein. One observation is that for pixels in the line open defect area, those pixels "belong to" the space that typically has a lower and / or more uniform intensity than the line in the BF pattern. The second observation is that for pixels close to the edge of the RDL line, the intensity typically has a drop, i.e., the intensity is lower than the intensity of pixels in the interior of the line but higher than the intensity of pixels in the space.

[0087] As described above, in one embodiment, the mask separates pixels within portions of the patterned features having different orientations into different regions, and in some such embodiments, the determination and detection steps described further herein are performed separately for the different regions. TM With the convenience of a mask (based on native inspection pixels), the embodiments described herein can easily identify with + / - 1 pixel accuracy whether a pixel in a frame belongs to a horizontal RDL line, a vertical RDL line, a 135 degree RDL line, a 45 degree RDL line, a space between RDL lines, a space between RDL lines with different orientations, a don't care area, etc. The steps described herein can then be performed separately for different areas of lines and spaces with different orientations, possibly using one or more different parameters (e.g., different predetermined window sizes, different thresholds, different sensitivities, etc.). For example, a first predetermined window size and shape and a first threshold or sensitivity can be used to inspect areas of lines with a vertical orientation, and a second predetermined window size and shape and a second threshold or sensitivity can be used to inspect areas of lines with a horizontal orientation. Other parameters of the steps described herein can be similarly varied. Parameters can also be varied for other patterned structures and orientations described herein.

[0088] In one such embodiment, one or more computer subsystems are configured for independently determining one or more characteristics of a predetermined window for determination steps performed separately for different regions. For example, a predetermined window size for determining spatial discontinuities may be defined in an inspection recipe for different RDL line segments. Each type of line segment may have its own specific window size definition. The window size definition may be determined in a recipe setup phase.

[0089] Fig.11 Different predetermined window size definitions for different RDL line segments are shown. The predetermined windows are shown in these images by the areas outlined in white. For example, image 1100 shows the predetermined window size definition for horizontal line segments. Image 1102 shows the predetermined window size definition for vertical line segments. Image 1104 shows the predetermined window size definition for 135 degree diagonal RDL line segments, and image 1106 shows the predetermined window size definition for 45 degree diagonal RDL line segments.

[0090] In one embodiment, determining the difference includes determining a range of a characteristic of at least one pixel and other pixels within a predetermined window, the range being the difference between a maximum value and a minimum value of the characteristic of at least one pixel and other pixels within the predetermined window, and detecting a defect candidate includes applying a threshold to the range. Two embodiments of a method for single die based inspection are described herein. This method uses a spatial continuity metric based on a spatial range window (i.e., a predetermined window). Based on the above two observations, for each pixel on an RDL line, a spatial window centered on the pixel on the RDL line segment can be thought of and the range (the difference between the maximum intensity pixel and the minimum intensity pixel) is calculated.

[0091] Figure 8 Demonstrates how this method can successfully detect defects on RDL lines. Specifically, Figure 8 An example 800 of using range as a metric to detect structural discontinuities due to line opening defects in horizontal RDL lines, and an example 810 of using range as a metric to detect structural discontinuities of line opening defects in diagonal RDL lines. Example 800 includes an image of a line 802 and a space 804 and two predetermined windows 806 and 808 overlaid on different portions of one of the lines. The predetermined windows 806 and 808 are simply two instances of the same predetermined window centered on different pixels within one of the lines. Example 810 includes an image of a line 812 and a space 814 and two predetermined windows 816 and 818 overlaid on one of the lines. As above, the predetermined windows 816 and 818 are simply two instances of the same predetermined window centered on different pixels within one of the lines. The range values ​​determined for pixels centered in the predetermined windows 806 and 816 will be relatively low because the pixels in these predetermined windows are all located in relatively low intensity and quiet spaces. In contrast, the range values ​​determined in the predetermined windows 808 and 818 are relatively high. For example, based on the brightness level setting of the inspection subsystem used to inspect the image, the range may be a maximum value of bright line structures and dark space background.

[0092] Figure 8It is shown that with this predetermined window, the embodiments described herein can detect RDL discontinuities at a relatively low computational cost. Of course, the sensitivity with which such RDL line opening defects are detected depends on the window size. If the window size is too large, it may miss some substantially small line openings. If the window size is too small, it may introduce nuisances. However, this detection method itself is not necessarily aimed at achieving substantially higher capture rates with zero or very low nuisance rates. Rather, this detection method can be used as a "cheap and fair" RDL line opening detection method. The other steps described herein can be performed to separate the detected defect candidates into nuisances and defects.

[0093] In another such embodiment, the one or more computer subsystems are further configured for determining whether a defect candidate is a defect based on one or more characteristics of the defect candidate. For example, this detection method may be combined with some morphological post-processing to remove relatively small line-open nuisances and / or with assigning a final defect classification through iDO (which is a defect classification module available on some tools commercially available from KLA), or with deep learning (DL) based binning to remove excessive nuisances and retain true line-open DOIs. This morphological post-processing and defect classification or binning may be performed as further described herein.

[0094] In one such embodiment, defect candidates include complete openings in patterned features. For example, the first method described above is relatively simple and fast and will detect fewer nuisances than other currently used methods and systems for RDL line inspection, but its performance is good for complete line openings and not necessarily for partial line openings.

[0095] The second method is based on the difference of pixel intensity relative to the average intensity in the same corresponding RDL line segment in the patterned feature and possibly in each die frame (which may be referred to herein simply as "Single Die Diff-to-Mean" or "SDDifftMean"). For example, in another embodiment, determining the difference includes determining the difference between the characteristic of at least one pixel and the average of the characteristics of at least one pixel and other pixels within a predetermined window. The average and the difference between the characteristic of the pixel and the average may be determined in any suitable manner.

[0096] In one such embodiment, the predetermined window encompasses the entirety of the patterned features in the test image. For example, unlike the first approach, the predetermined window within which the difference is determined may include the entirety of the patterned features in the test image. In particular, an average value of a characteristic (e.g., intensity) of all pixels in the identified patterned features may be determined, and then a difference between the characteristic and the average value may be determined for any one pixel in the patterned features and used in the detection steps described herein. When the entirety of the patterned features in the test image has a single orientation (e.g., Figure 8 This embodiment may be suitable when the line shown in Example 800 is shown in FIG.

[0097] In another such embodiment, the predetermined window encompasses the entirety of the patterned feature in the test image and one or more other entire patterned features in the test image, and the one or more other entire patterned features are of the same type and orientation as the patterned feature. For example, if the test image includes more than one patterned feature of the same type and orientation (e.g., Figure 8 ), then an average of the characteristics of all pixels in all those patterned features in the test image may be determined, and then the difference between the characteristic of any one pixel in those patterned features and the average may be determined and used in the detection steps described herein. Determining the average from more than one (or all) of the patterned features of the same type and orientation in the test image may provide advantages for the detection steps described herein, for example, the average may be more stable and less responsive to noise in the patterned features. Additionally, determining the average from only one or several patterned features in the test image (and not from any other images generated for the sample, such as other test images or reference images) may advantageously reduce the effects of variations within or across non-defective samples on the average.

[0098] The predetermined window may be similarly varied for different portions of the patterned features that have different orientations and are separated into different pixel regions. For example, an average value may be determined from all pixels in horizontally oriented segments of the RDL lines in the test image, another average value may be determined from all pixels in vertically oriented segments of the RDL lines in the test image, and so on. The average value used to determine the difference for any one pixel may then be selected based on the segment of the RDL line in which the pixel is located. In other words, if a pixel in the test image is located in a vertically oriented segment of the RDL line, then the average value determined from all (or at least some) of the vertically oriented segments of the RDL lines in the test image may be selected and used to determine the difference for the pixel.

[0099] In another such embodiment, detecting defect candidates includes determining the standard deviation from the mean and characteristics of at least one pixel and other pixels within a predetermined window, determining the signal strength by dividing the difference by the standard deviation, and applying a threshold to the signal strength. The concept is relatively simple: the mean intensity and the standard deviation of the intensity can be calculated for each set of line segments in each frame (e.g., horizontal, vertical, 135 degree diagonal, and 45 degree diagonal RDL segments). Then, the test is simply to find the relative intensity of each pixel, as described in the following equation for the corresponding line segment:

[0100] Signal_intensity=(pixel_intensity-mean_intensity) / standard_deviation.

[0101] If the signal intensity is greater than a specified threshold (specific to polarity, dark defects and light defects have different thresholds), then the pixel is anomalous. This approach is conceptually simple but can be relatively computationally expensive. For example, this approach may include two frame pixel visit passes, a first pass to collect noise statistics (e.g., mean and standard deviation) in each region or group, and a second pass for relative intensity calculations and anomaly detection. The mean and standard deviation may additionally be determined in any suitable manner known in the art.

[0102] In one such embodiment, the one or more computer subsystems are further configured for determining the mean and standard deviation from the median reference image generated for the test image. For example, the RDL spatial segmentation mask and the underlying MRD image can be used to automatically calculate the corresponding parameters of the average intensity range in the spatial segmentation. The MRD image can be generated in any suitable manner known in the art. This step can be performed during prescription setup time.

[0103] In an optional step performed depending on throughput optimization, after the mask is generated from the RDL line segment design polygon (meaning the design of the RDL polygon has been aligned with the RDL line structure in the MRD image), the mean and standard deviation of the pixel intensity of the horizontal, vertical, 45 degree and 135 degree diagonal RDL line segments are calculated, and similar calculations can be performed for those pixels in the spatial segments between the RDL line segments. These steps can be performed during the prescription setup time.

[0104] In another such embodiment, one or more computer subsystems are configured for determining the mean and standard deviation from the test image. For example, the mean and standard deviation may be determined from a single test image as described above and will provide the same advantages described above. In this way, the embodiments described herein may provide true single die image (SDI) defect detection.

[0105] In some embodiments, identifying a patterned feature includes applying a mask to a test image (the mask separating pixels in the test image that are within the patterned feature from all other pixels in the test image) and pre-processing the results of applying the mask to identify one or more pixels beyond one or more edges of the patterned feature and removing the identified one or more pixels from pixels within the patterned feature. In another embodiment, identifying a patterned feature includes applying a mask to a test image (the mask separating pixels in the test image that are within the patterned feature from all other pixels in the test image) and post-processing the results of applying the mask to identify one or more pixels beyond one or more edges of the patterned feature and removing the identified one or more pixels from pixels within the patterned feature. In these embodiments, a mask may be applied to the test image (as further described above) and pre-processing or post-processing (or even both) may be applied to the results of the mask application to account for any marginality in the steps. For example, due to system limitations, the mask placement accuracy may be approximately + / - 1.5 pixels for most systems. As further described below Fig. 9 As can be seen in , at the zoom level, there may be extra spatial pixels included in the RDL line segment. Those extra spatial pixels will contribute to excessive nuisance detection. Therefore, more robust pre-processing or post-processing will be beneficial for mitigating the nuisance detection. There are two ways to address these extra spatial pixels: (1) pre-processing the RDL line segment mask to remove spatial pixels in the line segment; and (2) post-processing to remove excess spatial pixels with lower intensity. Either way, the operation involves more expensive computations than the first method described herein. The pre-processing step and the post-processing step may be additionally performed in any suitable manner known in the art.

[0106] Specifically, Fig. 9 Two test images 900 and 908 generated for a sample are shown. Test image 900 includes two RDL lines 902 and 904 with a line open defect 906 in RDL line 904. Test image 908 also includes two RDL lines 910 and 912, neither of which includes a line open defect. However, the relative signal strengths determined for these images as described herein will result in the detection of both the line open DOI in test image 900 and the nuisance at the RDL line edge 914. In other words, the SDDifftMean method for RDL line open detection provides effective detection of line open and partial line open defects but may also detect excessive nuisance of pixels at the RDL line edge. Therefore, the pre-processing steps and post-processing steps described above may be important nuisance reduction techniques for the embodiments described herein.

[0107] In another embodiment, defect candidates include complete or partial openings in patterned features. For example, the SDDifftMean method is capable of detecting complete line opening defects as well as relatively partial line openings (e.g., Fig.10 Specifically, the SDDifftMean method is able to detect complete line opening defects (e.g. Fig. 9 The line opening defect 906 shown in the test image 900 of FIG. 9 ) and the partial line opening defect (eg Fig.10 1000) shown in the circle portion 1002 of the test image 1000 shown in FIG. 1004. In this way, the relative signal strength determined by the SDDifftMean method can detect partial line openings.

[0108] In some embodiments, one or more computer subsystems are configured to determine whether a defect candidate is a defect by performing a morphological operation on the result of the detection step. For example, the embodiments described herein may use morphological operations to remove relatively few candidate pixel groups. In the two single-die-based RDL line opening candidate detection methods described herein, it is inevitable that both methods will not provide only DOI capture. Compared with the traditional die-to-die difference-based inspection method, the two methods provide a better DOI capture rate to nuisance rate. The traditional die-to-die difference-based inspection method usually overwhelms the inspection with huge nuisance detection, making the detection of the real DOI impossible. The new single-die-based method (based on spatial discontinuity or based on relative signal strength) will give a much better capture rate at a much lower nuisance rate. However, for production-worthy wafer inspection, it is preferred to further reduce nuisance. One observation of this nuisance is that it is usually spread sporadically and has a relatively small pixel count group. In this way, morphological operations can be performed to remove relatively small-sized nuisances. This nuisance filtering based on morphological operations can reduce the amount of substantial nuisance at the inspection post-processing stage.Analyzing the morphological properties of defect candidates and using those morphological properties to separate possible nuisance from possible DOI may additionally be performed in any suitable manner.

[0109] In an additional embodiment, one or more computer subsystems are configured for determining whether a defect candidate is a defect by grading the defect candidate based on a test image. In another embodiment, one or more computer subsystems are configured for determining whether a defect candidate is a defect by applying a deep learning (DL) based classifier to a test image. For example, an embodiment may use iDO based grading or DL ​​based grading to further remove nuisance. In one such example, there are some structures that are relatively large in spatial size but can be visually identified as nuisances due to their appearance, for example, RDL line edge roughness to partial line openings, or line bends. iDO intra-line grading may be used to filter out possible nuisances at each parallel processing unit level. If the DOI and nuisance are visually distinguishable, DL grading using defect candidate tiles may have better nuisance reduction results than the iDO classifier. DL-based classification may be performed as described in U.S. Patent No. 10,607,119 issued to He et al. on March 31, 2020, which is incorporated by reference as if fully set forth herein. DL-based nuisance filtering may be performed as described in U.S. Patent Application No. 16 / 663,283 filed by Huang et al. on October 24, 2019, which is incorporated by reference as if fully set forth herein. The embodiments described herein may be further configured as described in such patents and patent applications.

[0110] In one embodiment, the determination and detection steps are performed without a reference image. In other words, detecting defect candidates may include a single image defect detection, rather than requiring two die images as in die-to-die inspection or test and reference images as in die-to-reference inspection. For example, metal particles are a major nuisance in back-end wafer inspection. Since metal particles are randomly distributed across a sample (e.g., a wafer) (e.g., across all the dies on the wafer), the metal particles cannot be offset by die-to-die differences. In particular, the difference image produced by die-to-die subtraction of the metal layer will include particle noise that cannot be offset by the subtraction and can be detected as an event. In fact, methods based on die-to-die differences can amplify the noise from metal particles, thereby making it impossible to detect DOI. Therefore, currently used die-to-die defect methods for detection and classification will not be applicable to such samples. As such, the embodiments described herein may be configured for detecting and classifying defects in one die without using any reference die (whether from an adjacent die or a standard reference die). Embodiments described herein can detect DOI candidates and separate true DOI from metal particle nuisances through DL or through single die defect patch images in another manner further described herein.

[0111] Fig.12 The diagram illustrates one embodiment of the steps that may be performed for detecting defects as described herein. The steps may be performed in two different stages: a setup stage 1200 and a runtime stage 1202. During the setup stage, one or more computer subsystems may use design data 1204 (e.g., a GDS file) as input for a rendering and placement step 1206, which may be performed in any suitable manner known in the art. In the setup stage, one or more computer subsystems may also perform a reference generation step 1208, which may include generating an MRD or another type of reference for sample inspection. The results of the reference generation step and the rendering and placement step may be used by the computer subsystem for a segmentation mask generation step 1210, which may include generating one of the segmentation masks described herein. The computer subsystem may also use the reference generation step 1208 output to perform an anchor point finding step 1212, in which an anchor point is selected in the generated reference. The anchor point may then be used in the inspection process to align the image generated by the inspection subsystem with the common reference.

[0112] At runtime of the inspection, the recipe parameters may be sent down to each CPU core that is running the inspection. In the runtime phase, additional recipe parameters associated with different RDL line / space segmentations may be sent down to the inspection computer and the mask for use in the inspection. In the runtime phase, the one or more computer subsystems use the output of the anchor point finding step to perform a tile to mask alignment step 1214 in which the image produced by the inspection subsystem is aligned with the mask so that patterned features in the image can be identified as further described herein by placing the mask step 1216 on the scanned image. As shown in step 1218, the one or more computer subsystems then perform a segmentation-based die-to-die (DtD) differential intensity-based inspection or a single die (SD) inspection as described herein. For example, in the run-time stage, the segmentation-based DtD / SD target intensity-based inspection step may include traditional die-to-die inspection using a segmentation mask or performing a new segmentation mask-guided single-die inspection using one of the following two methods further described herein: (1) a spatial window for structural continuity; and / or (2) the difference between pixel intensity in the same segment and its average intensity may be normalized by the standard deviation of the same segment.

[0113] Fig.13 The diagram illustrates an embodiment of steps that may be performed by the embodiments described herein for single die RDL line opening detection based on "spatial continuity". Fig.13, the steps are performed in two stages: a detection path stage 1300 and a post-processing stage 1302. The one or more computer subsystems perform a pre-processing step 1304, which may include any of the pre-processing steps described herein. After completing the pre-processing steps, the detection path can determine whether the pixel cycle is complete, starting as shown in step 1306. Performing a "pixel cycle" means looping through (visiting) each pixel in a working frame, tile image, test image, etc. If the pixel cycle is complete, then the one or more computer subsystems proceed to the post-processing stage.

[0114] If it is determined that the pixel cycle is not complete, the one or more computer subsystems determine whether the pixel is on RDL line n, as shown in step 1308. If the pixel is not on the RDL line, the one or more computer subsystems repeat step 1306. If it is determined that the pixel is on the RDL line, the computer subsystem determines whether the intensity of the pixel is within range (n), as shown in step 1310. The pixel intensity within the range can be defined as an intensity greater than the minimum value of the range setting and less than the maximum value of the range setting. If the pixel is not within range (n), the one or more computer subsystems return to step 1306. If it is determined that the pixel intensity is within range (n), the one or more computer subsystems determine whether the single-die pixel window range is less than a threshold, as shown in step 1312. A single-die pixel window range less than a threshold means that the maximum intensity of the neighboring pixels - the minimum intensity of the neighboring pixels is less than the threshold setting. If the single-die pixel window range is not less than the threshold, the one or more computer subsystems return to step 1306. If it is determined that the single-die pixel window range is less than the threshold, the one or more computer subsystems mark the pixel as defective, as shown in step 1314. After step 1314, the one or more computer subsystems return to step 1306.

[0115] When the one or more computer subsystems determine in step 1306 that the pixel cycle is complete, the one or more computer subsystems generate a defective blob in step 1316 of the post-processing stage 1302. The one or more computer subsystems then determine in step 1318 whether the defective blob cycle is complete. If it is determined that the defective blob cycle is not complete, the one or more computer subsystems determine whether the number of defective pixels in the blob is greater than a threshold, as shown in step 1320. If the number of defective pixels in the blob is not greater than the threshold, the one or more computer subsystems return to step 1318. If the number of defective pixels in the blob is greater than the threshold, the one or more computer subsystems generate a defect, as shown in step 1322. After the one or more computer subsystems generate the defect, the one or more computer subsystems return to step 1318. When the one or more computer subsystems determine that the defective blob cycle is complete, the one or more computer subsystems perform a post-processing step 1324, which may include any of the post-processing steps described herein. The one or more computer subsystems then perform a return results step 1326, in which the results of any of the steps performed by the computer subsystem are output as a test results archive or another suitable archive.

[0116] Fig.14 The diagram illustrates an embodiment of steps that may be performed by the embodiments described herein for single die RDL line opening detection based on “normalized relative intensity and average value”. Fig.14 , the steps may be performed in two stages: a noise collection path stage 1400 and a detection path stage 1402. Before starting the noise collection path, the one or more computer subsystems perform preprocessing 1404, which may include any of the preprocessing steps described herein. After completing the preprocessing steps, the one or more computer subsystems begin the noise collection path by determining whether the first pixel cycle is completed, as shown in step 1406. If the first pixel cycle is not completed, the one or more computer subsystems determine whether the pixel is on RDL line n or space m, as shown in step 1408. If it is determined that the pixel is not on RDL line n or space m, the one or more computer subsystems return to step 1406. If it is determined that the pixel is on RDL line n or space m, the one or more computer subsystems collect the mean and standard deviation (std) of the intensity of the segments of line (n) and / or space (m), as shown in step 1410. After the one or more computer subsystems perform step 1410, the one or more computer subsystems return to step 1406.

[0117] When the one or more computer subsystems determine that the first pixel cycle is complete in step 1406, the one or more computer subsystems begin detection path 1402 by determining whether the second pixel cycle is complete in step 1412. If the one or more computer subsystems determine that the second pixel cycle is not complete, the one or more computer subsystems determine whether the pixel is on RDL line n or space m, as shown in step 1414. If it is determined that the pixel is not on RDL line n or space m, the one or more computer subsystems return to step 1412. If it is determined that the pixel is on RDL line n or space m, the one or more computer subsystems calculate the signal intensity of the pixel using (pixel_intensity-mean[n]) / std[n], as shown in step 1416. After the one or more computer subsystems calculate the signal intensity of the pixel, the one or more computer subsystems determine whether the signal intensity of the pixel is greater than a threshold, as shown in step 1418. If the signal intensity of the pixel is not greater than the threshold, the one or more computer subsystems return to step 1412. If the pixel signal strength is greater than the threshold, the computer subsystem marks the pixel as defective, as shown in step 1420 .

[0118] When the one or more computer subsystems determine that the second pixel cycle is complete in step 1412, the one or more computer subsystems begin a post-processing phase 1422. The post-processing phase performed in this embodiment may be similar to Fig.13 The post-processing stages are the same as those shown in .

[0119] Described above and in Figures 12 to 14 Embodiments of the steps shown in the Figures may be performed by any of the system embodiments described herein. In addition, embodiments of the steps described and shown in these Figures may be performed as further described herein and combined with any other steps described herein.

[0120] There are some differences and similarities between the first and second methods described herein that can be used to select which of the methods is best for any particular use case. For example, the first pixel range-based method requires only one detection pass to access each pixel in the frame, which makes it computationally less expensive than the second method, which may include two passes of frame pixel access, one for noise statistics collection (mean and standard deviation of intensity in each segment) and one for inspection. In addition, the first pixel range-based method detects full line opening defects at a relatively low cost, while the second average-based method can detect RDL partial line opening defects in addition to full line opening defects. To reduce the cost of the second average-based method, an embodiment may calculate the mean and standard deviation on the MRD or another suitable reference in the setup phase, as described herein. In this way, an embodiment may avoid performing two passes for each pixel, but using this reference may be less able to adapt to wafer-scale local process variations in different dies. The post-processing stage of the two methods may be the same. The post-processing steps in both methods may include any of the post-processing steps described herein, such as iDO-based nuisance reduction and DL-grading-based nuisance reduction.

[0121] The embodiments described herein have several advantages over other currently used systems and methods for detecting defects on RDL lines and other noisy patterned features. For example, the embodiments described herein may utilize single die strength-based features for detecting (partial) line opening defects, which overcome the problem of excessive noise caused by the relatively rough metal surface of the RDL lines in die-to-die strength differences. In addition, the use of design-based masks allows the embodiments to address inconsistent shifts between the RDL lines and the sub-die structures below the RDL lines. The embodiments described herein may also help improve the sensitivity of capturing RDL line openings by existing back-end inspection tools for aWLP use cases.

[0122] The embodiments described herein differ from currently used systems and methods for detecting defects on RDL lines and other noisy patterned features in several important ways to achieve the advantages described above. For example, the embodiments described herein enable single die intensity-based inspections that avoid excessive noise in die-to-die difference inspections. In addition, the embodiments described herein provide new spatial continuity-based metrics for RDL line opening detection. The embodiments described herein also provide new relative intensity-based metrics for RDL line opening detection. In addition, the embodiments described herein can be configured to use design polygon information to simplify and facilitate the separation of different segments of RDL lines (e.g., horizontal, vertical, 135 degree diagonal, 45 degree diagonal) into different regions, which is particularly important for situations where traditional intensity-based mask generation becomes impossible. The embodiments described herein can also use DL-based grading to better distinguish between nuisance and DOI by removing nuisance that passes through intensity-based detection methods.

[0123] One or more computer subsystems may be configured to generate results of detected defect candidates and / or detected defects. The results of defect candidates and / or defects may include any of the results described herein, such as information of the bounding box of the detected defect (e.g., location, etc.), detection scores, information about defect classification (e.g., class labels or IDs, etc.), or any such suitable information known in the art. The results of defect candidates and / or defects may be generated by the computer subsystem in any suitable manner. The results of defect candidates and / or defects may have any suitable form or format, such as a standard file type. The computer subsystem may generate results and store the results so that the results may be used by the computer subsystem and / or another system or method to perform one or more functions of the sample or another sample of the same type. Such functions include, but are not limited to, changing a process such as a manufacturing process or a step to be performed on a sample in a feedback manner, changing a process such as a manufacturing process or a step to be performed on a sample in a feedforward manner, etc.

[0124] Each of the embodiments of each of the systems described above may be combined together into a single embodiment.

[0125] Another embodiment relates to a computer-implemented method for detecting defects on a sample. The method includes: identifying, by an inspection subsystem, a patterned feature in a test image included in a generated image of the sample. The method also includes: determining, for at least one pixel in the test image that is within the patterned feature, a difference between a characteristic of the at least one pixel and characteristics of other pixels in the test image that are within a predetermined window of the at least one pixel. Additionally, the method includes: detecting a defect candidate at the at least one pixel based on the determined difference. The identifying, determining, and detecting steps are performed by one or more computer subsystems coupled to the inspection subsystem.

[0126] Each of 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, and / or inspection subsystem described herein. The inspection subsystem and one or more computer subsystems may be configured according to any of the embodiments described herein (e.g., inspection subsystem 10 and computer subsystems 36 and 102, respectively). In addition, the method described above may be performed by any of the system embodiments described herein.

[0127] Additional embodiments relate to a non-transitory computer-readable medium storing program instructions executable on one or more computer systems to perform a computer-implemented method for detecting defects on a sample. Fig.15 One such embodiment is shown in FIG. Fig.15 , non-transitory computer readable medium 1500 includes program instructions 1502 executable on a computer system 1504. The computer-implemented method may include any of the steps of any of the methods described above.

[0128] Program instructions 1502 implementing methods such as those described herein may be stored on computer readable medium 1500. The computer readable medium may be a storage medium such as a magnetic or optical disk, tape, or any other suitable non-transitory computer readable medium known in the art.

[0129] Program instructions may be implemented in any of a variety of ways, including procedural, component-based, and / or object-oriented techniques, among others. For example, program instructions may be implemented using ActiveX controls, C++ objects, JavaBeans, Microsoft Foundation Classes ("MFC"), SSE (Streaming SIMD Extensions), or other techniques or methods as desired.

[0130] Computer system 1504 may be configured according to any of the embodiments described herein.

[0131] In view of this description, other modifications and alternative embodiments of various aspects of the present invention will be apparent to those skilled in the art. For example, systems and methods for detecting defects on samples are provided. Therefore, this description should be considered as merely illustrative and for the purpose of teaching those skilled in the art to implement the general manner of the present invention. It should be understood that the forms of the present invention shown and described herein should be considered as currently preferred embodiments. As will be fully apparent to those skilled in the art after benefiting from this description of the present invention, elements and materials may replace those illustrated and described herein, parts and processes may be reversed, and the specific features of the present invention may be utilized independently. Changes may be made to the elements described herein without departing from the spirit and scope of the present invention as described in the appended claims.

Claims

1. A system configured to detect defects on a sample, comprising: an inspection subsystem configured to generate an image of the sample; and One or more computer subsystems configured to detect defect candidates on the sample, wherein detecting the defect candidates comprises: identifying patterned features in a test image included in a generated image of the sample; determining, for at least one pixel in the test image that is within the patterned feature, a difference between a characteristic of the at least one pixel and the characteristics of other pixels in the test image that are within a predetermined window of the at least one pixel; and detecting a defect candidate at the at least one pixel based on the determined difference; The determining step and the detecting step of a defect candidate at the at least one pixel are performed without a reference image.

2. The system of claim 1, wherein the patterned features comprise lines located on a redistribution layer.

3. The system of claim 1, wherein the patterned features are formed of metal.

4. The system of claim 1, wherein the identifying comprises applying a mask to the test image, the mask separating pixels in the test image that are within the patterned features from all other pixels in the test image.

5. The system of claim 4, wherein the one or more computer subsystems are further configured for generating the mask according to a design of the sample.

6. The system of claim 4, wherein the mask separates pixels within portions of the patterned features having different orientations into different regions, and wherein the determining steps and the detecting the defect candidate at the at least one pixel are performed separately for the different regions.

7. The system of claim 6, wherein the one or more computer subsystems are further configured for independently determining one or more characteristics of the predetermined window for the determining steps performed separately for the different regions.

8. The system of claim 1 , wherein determining the difference comprises determining a range of the characteristic of the at least one pixel and the other pixels within the predetermined window, wherein the range is a difference between a maximum value and a minimum value of the characteristic of the at least one pixel and the other pixels within the predetermined window, and wherein detecting the defect candidate comprises applying a threshold to the range.

9. The system of claim 8, wherein the defect candidate comprises a complete opening in the patterned feature.

10. The system of claim 8, wherein the one or more computer subsystems are further configured for determining whether the defect candidate is a defect based on one or more characteristics of the defect candidate.

11. The system of claim 1, wherein determining the difference comprises determining a difference between the characteristic of the at least one pixel and an average of the characteristics of the at least one pixel and the other pixels within the predetermined window.

12. The system of claim 11, wherein the predetermined window encompasses the entirety of the patterned feature in the test image.

13. The system of claim 11, wherein the predetermined window encompasses the entirety of the patterned feature in the test image and one or more other entire patterned features in the test image, and wherein the one or more other entire patterned features are of the same type and orientation as the patterned feature.

14. The system of claim 11, wherein detecting the defect candidate comprises determining a standard deviation from the mean and the characteristic of the at least one pixel and the other pixels within the predetermined window, determining a signal strength by dividing the difference by the standard deviation, and applying a threshold to the signal strength.

15. The system of claim 14, wherein the one or more computer subsystems are further configured for determining the mean and the standard deviation from the test image.

16. The system of claim 11, wherein the identifying comprises: applying a mask to the test image, the mask separating pixels in the test image that are within the patterned features from all other pixels in the test image; and pre-processing the results of applying the mask to identify one or more pixels beyond one or more edges of the patterned features and remove the identified one or more pixels from the pixels within the patterned features.

17. The system of claim 11, wherein the identifying comprises: applying a mask to the test image, the mask separating pixels in the test image that are within the patterned feature from all other pixels in the test image; and post-processing the results of applying the mask to identify one or more pixels that are beyond one or more edges of the patterned feature and remove the identified one or more pixels from the pixels within the patterned feature.

18. The system of claim 11, wherein the defect candidate comprises a full or partial opening in the patterned feature.

19. The system of claim 1, wherein the one or more computer subsystems are further configured for determining whether the defect candidate is a defect by performing a morphological operation on a result of the detection.

20. The system of claim 1, wherein the one or more computer subsystems are further configured for determining whether the defect candidate is a defect by ranking the defect candidate based on the test image.

21. The system of claim 1, wherein the one or more computer subsystems are further configured for determining whether the defect candidate is a defect by applying a deep learning based classifier to the test image.

22. The system of claim 1, wherein the inspection subsystem is further configured for macro inspection.

23. The system of claim 1, wherein the inspection subsystem is further configured as an electron beam subsystem.

24. The system of claim 1, wherein the inspection subsystem is further configured as an optical subsystem.

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

26. A non-transitory computer readable medium storing program instructions executable on one or more computer systems to perform a computer implemented method for detecting defects on a sample, wherein the computer implemented method comprises: identifying, by the inspection subsystem, patterned features in the test image contained in the generated image of the sample; determining, for at least one pixel in the test image that is within the patterned feature, a difference between a characteristic of the at least one pixel and the characteristics of other pixels in the test image that are within a predetermined window of the at least one pixel; and detecting a defect candidate at the at least one pixel based on the determined difference; The determining step and the detecting step of a defect candidate at the at least one pixel are performed without a reference image.

27. A computer-implemented method for detecting defects on a sample, comprising: identifying, by the inspection subsystem, patterned features in the test image contained in the generated image of the sample; determining, for at least one pixel in the test image that is within the patterned feature, a difference between a characteristic of the at least one pixel and the characteristics of other pixels in the test image that are within a predetermined window of the at least one pixel; and detecting a defect candidate at the at least one pixel based on the determined difference, wherein the identifying, the determining, and the detecting are performed by one or more computer subsystems coupled to the inspection subsystem; The determining step and the detecting step of a defect candidate at the at least one pixel are performed without a reference image.

Images