Method and system for inspection of photolithography reticle

Abstract

A method for operating a photomask inspection system includes training an analytical model of the photomask inspection system using machine learning processing, generating a difference threshold based on the machine learning processing, storing a reference image, and generating a scanned image of the photomask using the photomask inspection system. The method includes generating a relative difference value by comparing the scanned image with the reference image. The method includes determining that the photomask is not defective if the relative difference value is less than the difference threshold. The method also includes determining that the photomask is defective if the relative difference value is greater than the difference threshold.

Description

Rapid inspection method and system for photolithography masks none There is a constant demand for increased computing power in electronic devices, including smartphones, tablets, desktop computers, laptops, and many other types of electronic devices. Integrated circuits provide the computing power for these electronic devices. One way to increase computing power in integrated circuits is to increase the number of transistors and other integrated circuit features that can be included in a given area of ​​a semiconductor substrate. Features in integrated circuits are produced in part by photolithography. Traditional photolithography techniques involve creating a mask (or shadow mask) that outlines the pattern of features to be formed on the integrated circuit die. A photolithography light source illuminates the integrated circuit die through the mask. The size of features that can be produced by photolithography of the integrated circuit die is influenced in part by the wavelength of the light produced by the photolithography light source. Smaller light wavelengths produce smaller feature sizes. As the patterns in the reticle continue to become increasingly dense, any defects or contamination in the reticle can result in defects that are transferred to the pattern formed in the wafer. Therefore, a reticle inspection process is implemented to help ensure that the reticle does not have any defects. If the reticle inspection process detects a defect, the reticle may undergo a lengthy cleaning process during which wafer processing is stopped. Additionally, previously processed wafers may be inspected to ensure that no defects are present in the wafer due to the defective reticle. One potential problem is that the inspection process may incorrectly determine that the reticle is defective, resulting in an expensive and time-consuming stop in wafer processing. none The following disclosure provides many different embodiments or examples for implementing the different features of the provided subject matter. Specific examples of components and configurations are described below to simplify the disclosure. Of course, these are merely examples and are not intended to be limiting. For example, the formation of a first feature above or on a second feature in the following description may include embodiments in which the first and second features are formed in direct contact, and may also include embodiments in which additional features may be formed between the first and second features so that the first and second features are not in direct contact. In addition, the disclosure may repeat reference numbers and / or letters in various examples. This repetition is for the purpose of simplicity and clarity and does not inherently dictate the relationship between the various embodiments and / or configurations discussed. Furthermore, spatially relative terms such as "below," "beneath," "below," "above," and "upper" may be used herein for convenience in describing the relationship of one component or feature to another component(s) or feature(s) as illustrated in the figures. Spatially relative terms are intended to encompass different orientations of the device in use or operation in addition to the orientation depicted in the figures. The device may be otherwise oriented (rotated 90 degrees or at other orientations) and the spatially relative descriptors used herein interpreted accordingly. Terms indicating relative degrees such as "about," "substantially," etc. should be interpreted as what one of ordinary skill in the art would consider given the current technical specifications. Embodiments of the present disclosure provide systems and methods for reliably inspecting photolithography reticles. Embodiments of the present disclosure utilize a reticle inspection system that includes an analytical model trained using machine learning to assist in inspecting photolithography reticles between exposures. The reticle inspection system stores one or more reference images of the reticle. During the inspection process, the reticle inspection system captures new scans of the reticle and compares them to the reference images. The reticle inspection system utilizes the analytical model to ensure that inspection is accurate and efficient, so that the reticle is not incorrectly determined to be defective. In some embodiments, a reticle inspection system may train an analysis model using a training set that includes images or scans of both defective and non-defective reticles. After training, the analysis model may analyze the scan of the reticle to determine whether the reticle is defective. In some embodiments, the analysis model may determine whether the reticle is defective and may also help determine the type of defect. The analysis model may also be used to determine whether the current scan of the reticle should be used to partially or completely overwrite one or more reference images used by the reticle inspection system. In some embodiments, reticle characteristic parameters and reticle inspection system parameters may be provided to a machine learning database used to train the analysis model. The analysis model may then generate simulated reference images to aid in future inspections of the reticle. This may help reduce false defect counts and improve inspection tool quality. Embodiments of the present disclosure provide several benefits. Specifically, the number of false defect counts is significantly reduced or completely eliminated. This results in fewer instances of unnecessary reticle cleaning and wafer inspection. This helps ensure that wafer processing is not stopped, resulting in more processed wafers and better wafer yield over time. Expensive reticle cleaning is also avoided. FIG1 is a block diagram of an EUV photolithography system 100 according to some embodiments. The components of EUV photolithography system 100 cooperate to perform photolithography processes and inspect photolithography reticles. As described in greater detail below, photolithography system 100 utilizes multiple high-NA brightfield reticles to produce a single pattern on a wafer. As used herein, the terms "EUV light" and "EUV radiation" may be used interchangeably. While the description of the figures may primarily focus on EUV reticles, the principles of the present disclosure extend to reticles utilized in photolithography processes other than EUV photolithography. EUV light lithography system 100 includes a droplet generator 102, an EUV light generation chamber 104, a droplet receiver 106, a scanner 108, and a laser 111. Droplet generator 102 outputs droplets into EUV light generation chamber 104. Laser 111 illuminates the droplets with pulses of laser light within EUV light generation chamber 104. The illuminated droplets emit EUV light 117. EUV light 117 is collected by collector 114 and reflected toward scanner 108. Scanner 108 conditions EUV light 117, reflects EUV light 117 from reticle 101, which includes a mask pattern, and focuses EUV light 117 onto wafer 116. EUV light 117 patterns a layer on wafer 116 according to the pattern of reticle 101. Each of these processes is described in more detail below. The droplet generator 102 generates and outputs a stream of droplets. The droplets may comprise titanium nitride, but droplets of other materials may be used without departing from the scope of the present disclosure. The droplets move at a high rate toward the droplet receiver 106. The droplets have an average velocity between 60 m / s and 200 m / s. The droplets have a diameter between 10 μm and 200 μm. The generator may output between 1,000 and 100,000 droplets per second. The droplet generator 102 may generate droplets having different initial velocities and diameters than those described above without departing from the scope of the present disclosure. In some embodiments, the EUV light generation chamber 104 is a laser produced plasma (LPP) EUV light generation system. As droplets pass through the EUV light generation chamber 104 between the droplet generator 102 and the droplet receiver 106, they are illuminated by a laser 111. When the droplets are illuminated by the laser 111, the energy from the laser 111 causes the droplets to form a plasma. The plasma-formed droplets generate EUV light 117. This EUV light 117 is collected by a collector 114 and transmitted to a scanner 108, and then to a wafer 116. In some embodiments, laser 111 is positioned outside EUV light generation chamber 104. During operation, laser 111 outputs pulses of laser light into EUV light generation chamber 104. The pulses of laser light are focused at points where droplets pass on their path from droplet generator 102 to droplet receiver 106. Each pulse of laser light is received by a droplet. When a droplet receives a pulse of laser light, the energy from the laser pulse generates high-energy plasma in the droplet. The high-energy plasma outputs EUV light 117. In some embodiments, laser 111 irradiates the droplet with two pulses. The first pulse flattens the droplet into a disk-like shape. The second pulse causes the droplet to form a high-temperature plasma. The second pulse is significantly more powerful than the first pulse. Laser 111 and droplet generator 102 are calibrated so that the laser emits a pair of pulses, irradiating the droplet with the pair of pulses. The laser can irradiate the droplet in ways other than those described above without departing from the scope of this disclosure. For example, laser 111 can irradiate each droplet with a single pulse or with more than two pulses. In some embodiments, there are two separate lasers. The first laser delivers a flattening pulse. The second laser delivers a plasmatizing pulse. In some embodiments, light output by the droplets is randomly scattered in many directions. The photolithography system 100 utilizes a collector 114 to collect scattered EUV light 117 from the plasma and direct or output the EUV light 117 toward a scanner 108 . The scanner 108 includes scanner optics 110. The scanner optics 110 include a series of optical adjustments to direct EUV light 117 toward the reticle. The scanner optics 110 may include refractive optics such as a lens or a lens system having multiple lenses (e.g., a zone plate). The scanner optics 110 may also include reflective optics such as a single mirror or a mirror system having multiple mirrors. The scanner optics 110 directs ultraviolet light from the EUV light generation chamber 104 toward the reticle 101. FIG1 illustrates a first reticle 101 within a scanner 108. Although not shown, the reticle 101 may be coupled to a holder. The holder holds the reticle 101a during the photolithography process. The photolithography process may also be described as "exposing" because the reticle 101 is exposed to EUV light during the photolithography process. During the EUV exposure process, EUV light 117 is reflected from the reticle 101 back toward further optical features of the scanner optics 110. In some embodiments, the scanner optics 110 includes a projection optics box. The projection optics box can have refractive optics, reflective optics, or a combination of refractive and reflective optics. The projection optics box directs the EUV light 117 toward a wafer 116, such as a semiconductor wafer. EUV light 117 includes a pattern from reticle 101. Specifically, reticle 101 includes a pattern to be defined in wafer 116. After EUV light 117 reflects from reticle 101, it contains the pattern of reticle 101. A layer of photoresist typically covers wafer 116 during EUV lithography. The photoresist helps pattern the surface of semiconductor wafer 116 according to the pattern of the reticle. The pattern of the mask 117 can be extremely dense, with very small features. Therefore, if any defects exist on the mask 117, such defects can cause malformation of the features or pattern on the wafer 116. Various types of defects can occur. One type of reticle defect is contamination of the reticle by debris particles. As previously described, EUV light 117 can be generated by irradiating titanium nitride droplets with an intense laser. This can sometimes result in very fine titanium nitride debris particles traveling from EUV light generation chamber 104 into scanner 108. The titanium nitride debris particles can adhere to the exposure surface of reticle 101. Such debris particles can cause reflected EUV light 117 to carry a pattern including the debris particles. This can result in distorted features on wafer 116. Additionally, other types of debris particles can impact reticle 117. For example, microscopic dust particles or other types of debris particles can enter scanner 108 and accumulate on reticle 101. Other types of reticle defects may include delamination, coating peeling, material concentration, scattering, roughness, coating deformation, oxidation, reduction, structural cross-linking, or other types of defects in the material layers comprising the reticle 101. Each of these defects may result in improper formation of features or patterns in the wafer 116. To ensure that the reticle is not contaminated before exposure, it can be beneficial to perform a wafer inspection process. One possible solution is to have a reticle inspection system capture an image of the reticle. If the reticle inspection system detects a defect, a human can inspect the scan to determine whether the defect actually exists. However, human inspection is expensive, time-consuming, and prone to errors. Another possible solution utilizes one or more reference images corresponding to images of a non-defective reticle for comparison with the current scan of the reticle to detect defects in the reticle based on whether the current scan differs from the one or more reference images. However, because the reticle is used in multiple exposure processes, the state of the reticle may change over time without being defective. For example, there may be a decrease in the reflectivity or brightness of the reticle that does not correspond to an actual defect, as the reticle may continue to properly form features in a wafer with slightly altered reflectivity or brightness. However, such a decrease in reflectivity or brightness may be flagged during comparison with the reference images, leading to an erroneous determination of a defect in the reticle.

[0026] Embodiments of the present disclosure utilize a reticle inspection system 103 that includes an analysis model 130 to assist in inspecting the reticle 101. The analysis model 130 is trained using one or more machine learning processes to assist in detecting reticle defects. In some embodiments, the reticle inspection system 103 uses a machine learning process to train the analysis model 130 to detect defects in the reticle 101. The machine learning process may include generating a training set that includes images of reticles with defects and non-defective reticles. During the machine learning process, the analysis model 130 is iteratively trained to correctly identify which images from the training set correspond to defective reticles and which images correspond to non-defective reticles. The training process may include iteratively training the neural network of the analysis model to accurately classify each image from the training set. The training process is complete when the analysis model 130 is able to correctly classify images in the training set within a selected error tolerance. After training analysis model 130, the analysis model can be utilized to help reticle inspection system 130 detect defects in reticle 101. For example, after a selected number of exposures of reticle 101, reticle inspection system 134 can collect scanned images 134 of reticle 101. The scanned images can include optical images, electron microscope images, or other types of scanned images. Scanned images 134 are then processed by analysis model 130. Analysis model 130 determines whether reticle 101 is defective. In some embodiments, the analysis model 130 may return a confidence score indicating how confident the analysis model 130 is that a defect is present or absent. If the confidence score is below a selected threshold, a human may be requested to inspect the scanned image 134. If the confidence score is above a selected threshold, the reticle 101 may be provided to the reticle cleaning system 127. In some embodiments, the analysis model 130 can be utilized to determine what types of defects are present in the reticle 101. During machine learning processing of the analysis model 130, a training set can be assembled that labels the training set images based on whether defects are present and what types of defects are present. A machine learning process can then be performed that trains the analysis model to classify the scanned images as one of a plurality of defect types or as completely defect-free. Some examples of defect classifications may include contamination by droplet debris, contamination by particles other than droplets, absorption degradation, reflectance degradation, coating delamination, material concentration, scattering, roughness, coating deformation, oxidation, reduction, tissue cross-linking, or other types of defects. In some embodiments, reticle inspection system 103 utilizes a reference image to detect defects in reticle 101. Reticle inspection system 103 captures a scanned image 134 of reticle 101 and compares the scanned image to reference image 132. If scanned image 134 differs sufficiently from reference image 132, reticle inspection system 103 may determine that reticle 101 is defective. In some embodiments, analytical model 130 may facilitate the comparison between reference image 132 and scanned image 134. However, as previously described, one problem with conventional approaches is that, over time with repeated exposures, some characteristics of the reticle 101 may change in a manner that does not result in improper functioning of the reticle 101. In other words, some variation, and even some type of degradation, may occur to the reticle 101 without causing the reticle 101 to fail to properly form patterns or features on the wafer 116. In the case where reticle 101 is an EUV reticle, the reticle may be composed of a reflective multilayer on a substrate. The reflective multilayer reflects EUV light. The EUV reticle may also include a layer of an absorbent material on the reflective multilayer. The absorbent material absorbs EUV light. The pattern of the reticle is formed by patterning the absorbent material so that the reflective multilayer is exposed according to the pattern of grooves formed in the absorbent material. When EUV light 117 is incident on the EUV reticle, some of the EUV light will be absorbed by the absorbent material, and some of the EUV light will be reflected by the reflective multilayer. The reflected EUV light 117 includes the pattern of the reticle. However, with numerous exposures, the reflectivity of the reflective multilayer can change. Additionally, the absorption properties of the absorbent material can also change. Consequently, the brightness level reflectivity of reticle 101 can change over time without properly accounting for reticle 101 failures in the pattern formed in wafer 116. However, if a simple comparison is made between reference image 132 corresponding to a new reticle and scanned image 134 corresponding to a reticle 101 that has undergone numerous exposures, the difference in brightness reflectivity can trigger a reticle inspection system to identify reticle 101 as defective. In some embodiments, if the most recent scan image 134 does not correspond to a defective reticle 101, the reticle inspection system 103 overcomes this problem by updating the reference image 132 with the most recent scan image 134. In this way, as the reticle 101 ages and is not defective, the reference image 132 is continuously updated or replaced by the most recent scan image 134 so that there is never a large difference between the new scan image 134 and the reference image 132 unless a true defect has occurred. In some embodiments, reticle inspection system 103 can replace portions of reference image 132 with portions of scanned image 134. Thus, reticle inspection system 103 can selectively replace the entire reference image 132 with the most recent scanned image 134, or can selectively replace one or more portions of reference image 132 with one or more corresponding portions of the most recent scanned image 134. Analysis model 130 can be trained to determine whether a reference image should be updated or replaced by a more recent scanned image 134. In some embodiments, the analysis model 130 can be trained to generate simulated reference images 132. For example, the analysis model 130 can be trained using a training set that includes a large number of scanned images spanning the lifespan of a large number of reticles. The analysis model 130 can be trained to generate reference images based on the number of exposures and other operating parameters associated with the reticle 101. This is because the training process trains the analysis model to understand what a healthy reticle 101 should look like at a given lifespan based on the number of exposures and other operating parameters associated with the reticle 101 and the reticle inspection system 103. For example, the scanned image generated by the reticle inspection system 103 may be affected by various parameters of the reticle inspection system 103. In instances where the reticle inspection system 103 utilizes one or more lasers to scan the reticle 101, the operating parameters of the reticle inspection system 103 may include power supply, beam position, beam stability, beam size, wavelength purity, pulse duration, timing control, laser heating, lens heating, isolator stability, acousto-optic modulation (AOM) stability, acousto-optic tunable filter (AOTF) stability, and other parameters. Other image generation parameters that may affect the generation of the scanned image may include wafer optical-to-electrical conversion efficiency and stability, channel coherence calibration, and dark noise associated with charge-coupled devices, complementary metal-oxide-semiconductor (CMOS) devices, and complementary metal-oxide-semiconductor (CMOS) time-delay integrated circuit devices, among other factors. Parameters associated with the stage storing the reticle 101 in the reticle inspection system 103 may include vibration, X, Y, and Z stability, leveling, stage acceleration, and stage movement speed. Parameters associated with optical modulation of the reticle inspection system 103 may include target temperature, lens transmission, and focus stability. Other factors may include mask usage, EUV movement, exposure die size, dose, material concentration, handling, and storage time. All of these factors and other factors associated with the reticle and with the reticle inspection system 103 may affect the generation of the scanned image 134. Thus, the training process for the analysis model 130 may utilize a large amount of data related to the reticle inspection system 103, the reticle 101, reticle scans, and other data to train the analysis model 130 to understand how a scan of a healthy reticle should appear based on the parameters of the reticle 101, the usage of the reticle 101, and the parameters of the reticle inspection system 103. The training process enables analysis model 130 to generate simulated reference images based on known characteristics of reticle 101 and reticle inspection system 103. Thus, prior to an upcoming scan of reticle 101, parameters associated with reticle 101 and reticle inspection system 103 may be provided to trained analysis model 130. Trained analysis model 130 may then generate one or more simulated reference images for use as reference images 132 in inspecting reticle 101. Reticle inspection system 103 may then capture a newly scanned image 134 of reticle 101 and compare the newly scanned image 134 to the simulated reference images. Analysis model 130 may then detect defects in reticle 101 based on the differences between the newly scanned image 134 and the simulated reference images. The comparison may take into account lateral image shift, reflectivity calibration, and similar factors, focus / Z position maps, and defect scan image area analysis. Scan data may be overwritten based on scan image quality analysis such as focus, position shift stability, and reflectivity stability. Reference data can be overwritten for all areas of the mask, a single step, multiple stripes, or various localized regions can be overwritten. In some embodiments, the reticle inspection system 103 may utilize a difference threshold, also known as a specification, to help determine whether a defect exists in the reticle 101. As an example, over time, there may be significant variations in the health practices or reflectivity levels for healthy reticles. The difference threshold can account for potentially large variations in brightness reflectivity when comparing a scanned image to a reference image. As long as the difference between the reference image in the scanned image is less than the difference threshold, the reticle 101 may be considered non-defective. However, if the difference between the reference image and the scanned image is greater than the difference threshold, the reticle inspection system 103 may determine that the reticle 101 has a defect. If the difference falls within the difference threshold, the analysis model 130 determines whether to automatically replace the reference image 132 (either completely or partially) with the current scanned image. EUV system 100 includes reticle storage 121. Reticle storage 121 may include storage and protection pods that enclose and protect reticle 101 when not in use. Reticle 101 may be enclosed in reticle storage 121 immediately after it has been initially manufactured. Reticle 101 remains in reticle storage 121 during transportation from a manufacturing site to a wafer processing site. Reticle storage 121 provides very strong protection from contaminants when reticle 101 is not in use. The reticle 101 may be held in the reticle storage 121 until it is ready for EUV photolithography processing. At this point, the reticle 101 is transferred from the reticle storage 121 to the scanner 108. The reticle storage 121, or a portion of the reticle storage 121, may be carried to the scanner 108. The reticle 101 is then unloaded from the reticle storage onto a holder and subsequently used for double exposure EUV processing. After EUV processing, the reticle 101 is unloaded from the holder into the reticle storage 121. The EUV light lithography system 100 may also include a wafer storage 123. The wafer storage 123 stores wafers 116 when they are not in use. The wafer storage 123 may include storage for wafers 116 that are yet to be transferred to the scanner 108 for patterning. The wafer storage 123 may also include storage for wafers 116 that have already been patterned within the scanner 108. The EUV system 100 includes a transfer system 125. The transfer system 125 may include one or more robotic arms. The one or more robotic arms may transfer the reticle 101 between the scanner 108, the reticle stocker 121, the reticle scanner, and the reticle cleaning station. The one or more robotic arms may also transfer the wafer 116 between the scanner 108 and the wafer stocker 123. In some embodiments, the robotic arm that transfers the wafer 116 is separate from the robotic arm that transfers the reticle 101. The EUV system 100 may include other types of reticle transport systems without departing from the scope of this disclosure. EUV system 100 includes a reticle cleaning system 127. When reticle inspection system 103 determines that a defect or contamination is present in reticle 101, transfer system 125 can transfer reticle 101 from scanner 108 to reticle cleaning system 127. As previously described, the process of cleaning a reticle is time-consuming and expensive. During cleaning, reticle 101 cannot be used to process wafer 116. Therefore, incorrectly determining that reticle 101 is defective is very expensive. Therefore, reticle inspection system 103 of FIG. 1 is very beneficial in reducing or completely eliminating false detections of defects in reticle 101. FIG2 is a simplified cross-sectional view of an EUV reticle 101 according to some embodiments. Reticle 101 includes a substrate 140, a reflective multilayer 142 positioned on substrate 140, a buffer layer 144 positioned on reflective multilayer 142, and an absorber layer 146 positioned on buffer layer 144. Grooves 156 in absorber layer 146 correspond to the pattern of reticle 101 that will be imparted to wafer 116. The substrate 140 includes a low thermal expansion material. The low thermal expansion material substrate 140 is used to minimize image distortion caused by heating of the photomask 101. The low thermal expansion material substrate 140 may include a material having a low defect level and a smooth surface. In one embodiment, the substrate 140 may include SiO 2. The substrate 140 may be doped with titanium dioxide. The substrate 140 may include other low thermal expansion materials besides those described above without departing from the scope of the present disclosure. The photomask 101 includes a reflective multilayer 142. The reflective multilayer 142 is positioned on the substrate 140. The reflective multilayer 142 is used to reflect extreme ultraviolet light during the photolithographic process used by the photomask 101. The reflective properties of the reflective multilayer 142 are described in more detail below. In one embodiment, reflective multilayer 142 operates based on the reflective properties of the interface between two materials. Specifically, when light is incident on an interface between two materials of different refractive indices, reflection of the light occurs. When the difference in refractive index is greater, a greater portion of the light is reflected. One technique for increasing the proportion of reflected light is to deposit multiple layers of alternating materials, including multiple interfaces. The properties and dimensions of the materials can be selected so that constructive interference occurs with light reflected from different interfaces. However, the absorption properties of the materials used for the multiple layers can affect the achievable reflectivity. Thus, reflective multilayer 142 includes a plurality of pairs of layers. Each pair includes a layer of a first material and a layer of a second material. The materials and thicknesses of the layers are selected to promote reflection and constructive interference of EUV light. In one embodiment, each pair of layers includes a molybdenum layer and a silicon layer. In one example, the thickness of the molybdenum layer is between 2 nm and 4 nm. In one example, the thickness of the silicon layer is between 3 nm and 5 nm. The thickness of the layers in the reflective multilayer 142 is selected based on the expected wavelength of the EUV light used in the photolithography process and the expected angle of incidence of the EUV light during the photolithography process. The wavelength of the EUV light is between 1 nm and 20 nm. According to one embodiment, the number of pairs of layers is between 20 and 60 pairs of layers. Other materials, thicknesses, numbers of pairs, and configurations of layers in the reflective multilayer 142 may be utilized without departing from the scope of the present disclosure. Other wavelengths of EUV light may be used without departing from the scope of the present disclosure. In one embodiment, a buffer layer 144 is positioned on the reflective multilayer 142. One purpose of the buffer layer 144 is to protect the reflective multilayer during the etching process of the absorber layer 146. Therefore, the buffer layer 144 includes a material that is resistant to etching by the etching processes that etch the absorber layer 146. The etching process and the materials of the absorber layer are described in more detail below. In one embodiment, buffer layer 144 comprises ruthenium. Buffer layer 144 may include ruthenium compounds, including ruthenium boride and ruthenium silicide. The buffer layer may comprise chromium, chromium oxide, or chromium nitride. Buffer layer 144 may be deposited using a low-temperature deposition process to prevent diffusion of buffer layer 144 into reflective multilayer 142. In one embodiment, buffer layer 144 has a thickness between 2 nm and 4 nm. Other materials, deposition processes, and thicknesses may be used for buffer layer 144 without departing from the scope of this disclosure. Absorber layer 146 is positioned on buffer layer 144. The material of absorber layer 146 is selected to have a high absorption coefficient for the wavelength of EUV radiation to be used in photolithography processes with reticle 101. In other words, the material of absorber layer 146 is selected to absorb EUV radiation. In one embodiment, the absorber layer 146 has a thickness between 40 nm and 100 nm. In one embodiment, the absorber layer 146 comprises a material selected from the group consisting of chromium, chromium oxide, titanium nitride, tantalum nitride, tantalum, titanium, aluminum-copper, palladium, tantalum boron nitride, tantalum boron oxide, aluminum oxide, molybdenum, or other suitable materials. Other materials and thicknesses may be used for the absorber layer 146 without departing from the scope of the present disclosure. In one embodiment, the absorbent layer 146 includes a first absorbent layer 148 and a second absorbent layer 150. The first absorbent layer 146 is positioned on the buffer layer 144. The second absorbent layer 150 is positioned on the first absorbent layer 148. In one embodiment, first absorber layer 148 comprises tantalum boron nitride. Second absorber layer 150 comprises tantalum boron oxide. The thickness of the first absorber layer is between 30 nm and 80 nm. The thickness of the second absorber layer 150 is between 1 nm and 40 nm. Absorber layer 146 may comprise different materials, thicknesses, and numbers of layers than those described above without departing from the scope of the present disclosure. In one embodiment, absorber layer 146 comprises only a single absorber layer. Thus, absorber layer 146 may be an absorber layer. The pattern of the grooves 156 corresponds to the pattern to be formed in the wafer 116 (or in one case corresponds to the reverse of the pattern). As previously described, the reflectivity, brightness, composition, and other parameters of the mask 101 may change over time after repeated exposures. As previously described, the mask inspection system 103 is able to take such changes into account when inspecting the mask 101. Figure 3 is a block diagram of the mask inspection system 103 of Figure 1 according to some embodiments. As described with respect to Figure 1, the mask inspection system 103 is used to inspect the mask 101. As previously described, the mask inspection system 103 uses machine learning to detect defects in the mask, decide whether to replace the reference image 132 with the most recent scanned image 134, and generate a simulated reference image. In one embodiment, the mask inspection system 103 includes an analysis model 130 and a training module 160. The training module 160 trains the analysis model 130 using machine learning processing. As previously described, the machine learning process trains the analysis model 130 to detect defects in the reticle 101, determines whether to detect defects in the reticle, determines whether to replace the reference image 132 with the most recently scanned image 134, and generates a simulated reference image. Although the training module 160 is shown as separate from the analysis model 130, in practice, the training module 160 may be part of the analysis model 130. Furthermore, although the analysis model 130 is shown as a single analysis model, in practice, the analysis model 130 may include multiple different analysis models to perform the functions described herein. The reticle inspection system 103 includes or stores training data 162. The content of the training data 162 depends on the functionality assigned to the analysis model 130. In some embodiments, the reticle inspection system 103 uses the machine learning process to train the analysis model 130 to detect defects in the reticle 101. The machine learning process may include generating the training data 162, which includes images of reticles with defects and reticles without defects. During the machine learning process, the analysis model 130 is trained through iterations to correctly identify which images from the training set correspond to defective reticles where the images correspond to non-defective reticles. In some embodiments, the analysis model 130 can be utilized to determine what types of defects are present in the reticle 101. During machine learning processing of the analysis model 130, the training set data 162 can include labeled training set images based on whether a defect is present and what type of defect is present. A machine learning process can then be performed that trains the analysis model 130 to classify the scanned images as one of a plurality of defect types or as completely defect-free. Some examples of defect classifications may include contamination by droplet debris, contamination by particles other than droplets, absorption degradation, reflectance degradation, coating delamination, material concentration, scattering, roughness, coating deformation, oxidation, reduction, tissue cross-linking, or other types of defects. In some embodiments, the analysis model 130 can be trained to generate simulated reference images 132. For example, the analysis model 130 can be trained using a training set of data comprising a large number of scanned images spanning the lifespan of a large number of reticles. The analysis model 130 can be trained to generate reference images based on the number of exposures and other operating parameters associated with the reticle 101. This is because the training process trains the analysis model to understand what a healthy reticle 101 should look like at a given lifespan based on the number of exposures and other operating parameters associated with the reticle 101 and the reticle inspection system 103. The training set of data 162 can include parameters such as beam power, beam position, beam stability, beam size, wavelength purity, pulse duration, timing control, laser heating, lens heating, isolator stability, acousto-optic modulation (AOM) stability, acousto-optic tunable filter (AOTF) stability, and other parameters. Training data 162 may include wafer-to-cell conversion efficiency and stability, channel-to-channel calibration, and dark noise associated with charge-coupled devices (CCDs), complementary metal-oxide-semiconductor (CMOS) devices, and complementary metal-oxide-semiconductor (CMOS) time-delay integrated circuits (TDICs), among other factors. Training data 162 may include stage data such as vibration, X, Y, and Z stability, leveling, stage acceleration, and stage motion speed. Training data 162 may include target temperature, lens transmission, and focus stability. Training data may include mask usage, EUV movement, exposure die size, dose, material concentration, handling, and storage time. The training process enables the analysis model 130 to generate simulated reference images based on known characteristics of the reticle 101 and the reticle inspection system 103. In one embodiment, the analysis model 130 comprises a neural network. However, other types of analysis models or algorithms may be used without departing from the scope of the present disclosure. In some embodiments, the reticle inspection system 103 includes processing resources 168, memory resources 170, and communication resources 172. The processing resources 168 may include one or more controllers or processors. The processing resources 168 are used to execute software instructions, process data, perform reticle inspection processing, perform signal processing, read data from memory, write data to memory, and perform other processing operations. The processing resources 168 may include physical processing resources 168 located at the location or facility of the photolithography system 100. The processing resources may include virtual processing resources 168 remote from the location of the photolithography system 100 or the facility where the photolithography system 100 is located. The processing resources 168 may include cloud-based processing resources, including processors and servers accessed via one or more cloud computing platforms. In some embodiments, memory resources 170 may include one or more computer-readable memories. Memory resources 170 are used to store software instructions associated with the functionality of the reticle inspection system and its components, including but not limited to the analysis model 130. Memory resources 170 may store data associated with the functionality of the reticle inspection system 103 and its components. The data may include training set data 162, current processing condition data, and any other data associated with the operation of the reticle inspection system 103 or any of its components. Memory resources 170 may include physical memory resources located at the site or facility of the photolithography system 100. Memory resources may include virtual memory resources located remotely from the site or facility of the photolithography system 100. Memory resources 170 may include cloud-based memory resources accessed via one or more cloud computing platforms. In some embodiments, communication resources may include resources that enable the reticle inspection system 103 to communicate with components associated with the photolithography system 100. For example, the communication resources 172 may include wired and wireless communication resources that enable the reticle inspection system 103 to receive data associated with the photolithography system 100. The communication resources 172 may enable the reticle inspection system 103 to communicate with remote systems. The communication resources 172 may include one or more networks, such as a wired network, a wireless network, the Internet, or an intranet, or may facilitate communication via such one or more networks. The communication resources 172 may enable components of the reticle inspection system 103 to communicate with each other. In some embodiments, analysis model 130 is implemented by processing resources 168, memory resources 170, and communication resources 172. Reticle inspection system 103 can be a distributed reticle inspection system having components and resources remotely located from each other and from photolithography system 100. In some embodiments, the reticle inspection system 103 includes a data integration server 166. The data integration server 166 collects data to be provided to the analysis model 130. For example, the data integration server 166 may aggregate and store the training set data 162. The data integration server 166 may receive data from one or more controllers implemented based on the processing resources 168 and the memory resources 170. In some embodiments, the reticle inspection system 103 includes a scanning resource 174. The scanning resource 174 may include components, processes, and systems utilized to generate the scanned image 134 of the reticle 101. This may include cameras, lasers, charge-coupled devices, complementary metal-oxide-semiconductor (CMOS) devices, stages, motors, sensors, and other components. FIG4 is a flow chart of a method 400 according to some embodiments. Method 400 may utilize the components, processes, and systems described with respect to FIG1 through FIG3 and in subsequent figures. At 402, an EUV reticle is exposed during a photolithography process. At 404, the reticle is scanned by the reticle inspection system 103. At 406, scan results are output by the reticle inspection system 103. At 408, the analysis model 130 analyzes the scan results output according to one or more processes. The processes utilized by the analysis model may include one or more of semi-automatic defect detection, fully automated defect detection, reference image replacement, and reference image simulation. If no defects are detected, the reticle 101 is exposed to the photolithography process again at 402. If defects are detected, the reticle 101 may be cleaned. FIG5 is a flow chart of a method 500 according to some embodiments. Method 500 may utilize the components, processes, and systems described with respect to FIG1 through FIG4 and in subsequent figures. At 502, reticle 101 is scanned by reticle inspection system 103. At 504, reticle inspection system 103 falsely detects a defect. At 506, training data is collected for use in one or more machine learning processes. Although FIG5 illustrates training data collected after a false defect has been detected, in practice, the training data and corresponding machine learning processes are performed previously. At 508, machine learning processes are performed to generate thresholds different from those previously described. The scanned image is then compared to a reference image. If the difference between the scanned image and the reference image is outside the difference threshold, the reticle is sent for cleaning at 510. If the difference between the scanned image and the reference image is within the difference threshold, the analysis model performs the processes at 512 or 514. At 512, the reference image is fully or partially overwritten with the most recently scanned image as previously described. At 514, one or more simulated reference images are generated. This may replace an old reference or set a new reference. FIG6 is a flow chart of a method 600 according to some embodiments. Method 600 may utilize the components, processes, and systems described with respect to FIG1 through FIG5 and in subsequent figures. At 602, the reticle inspection system 103 scans the reticle 101 and outputs the scan results. At 604, the reticle inspection system 103 performs a machine defect determination based on the output scan results. If the machine defect determination definitively determines that a true defect exists in the reticle 101, then at 606, the wafer lot processed by the reticle is halted, the lot is verified, the reticle 101 is repaired, and a root cause investigation is performed to determine the source of the defect. If the machine defect determination is inconclusive, then at 608, a human judge may review the scan results. If the human judge determines that a defect exists, processing proceeds to 606 as previously described. If the human judge determines that no true defect exists, then at 610, the wafer lot is released, and at 612, the analysis model 130 compares the reference image to the scanned image. At 614, the analysis model determines whether the comparison is within a difference threshold. If the comparison is within the difference threshold, then at 616, the analysis model performs a check of the ratio of the false defective area to the total area. If the ratio is greater than the threshold ratio, the entire reference image is overwritten with the most recent scanned image. If the ratio is less than or equal to the threshold ratio, then at 620, the reference image is partially overwritten with a portion of the scanned image. From 618 to 620, the method returns to 602. At 614, if the comparison is outside the difference threshold, then at 622, the expert can verify the results. If no real defect is detected, then the process proceeds to 616. If a real defect is detected at 622, then the process proceeds to the original restoration step at 624. The original restoration step corresponds to step 606. FIG7 is a flow chart of a method 700 according to some embodiments. Method 700 may utilize the components, processes, and systems described with respect to FIG1 through FIG6 and in subsequent figures. At 702, an initial setup for reticle inspection is performed. At 704, a training set is assembled as previously described. Machine learning processing is performed to train an analysis model. At 706, a threshold difference is generated based on the machine learning processing. If the scan result is within the threshold, the method proceeds to 712. If the scan result is outside the threshold difference, then at 708, an expert may review the result. If the result is abnormal, the reticle may be sent to reticle cleaning at 710. If the result is normal, the process returns to 706. At 712, a reticle pattern is simulated. The reticle pattern corresponds to the pattern intended to be carried by reticle 101. At 714, a simulated image representing a defect-free condition is generated. At 716, a simulated image representing a defect condition is generated. At 718, reticle inspection system 103 outputs the scan results to the analysis model. The analysis model also receives simulated reference images of defects and non-defects. At 720, the analysis model analyzes the scan results. If the analysis model is 100% aligned with the human judgment at 722, processing can proceed to 724 or 726. At 724, a comparison is performed between the scanned image and the reference image. At 726, an automatic defect determination is performed. If no defects are present at 728, processing proceeds to 730. At 730, a determination is made to determine whether a portion of the previous reference image should be overwritten. At 732, the reference image is partially or completely overwritten and provided to the analysis model. FIG8 is a flow chart of a method 800 according to some embodiments. Method 800 may utilize the components, processes, and systems described with respect to FIG1 through FIG7 and in subsequent figures. At 802, an initial setup for reticle inspection is performed. At 804, a training set is assembled as previously described. Machine learning processing is performed to train an analysis model. At 806, a threshold difference is generated based on the machine learning processing. If the scan result is within the threshold, the method proceeds to 812. If the scan result is outside the threshold difference, then at 808, an expert can verify the result. If the result is abnormal, the reticle can be sent to reticle cleaning at 810. If the result is normal, the process returns to 806. At 811 , the characteristics of future scans are predicted using an analytical model based on machine learning processing. At 812 , a reticle pattern is simulated. The reticle pattern corresponds to the pattern intended to be carried by reticle 101 . At 814 , a simulated image representing a defect-free condition is generated. At 816 , a simulated image representing a defect condition is generated. At 818 , a scan comparison is performed. At 820 , the scan data is output. At 822 , if there is a 100% match with the human determination result, processing proceeds to 824 . At 824 , automated defect determination is performed. If a false defect is present, the result is provided to the training set. FIG9 illustrates a graph indicating reflectivity measurements of a reticle over time according to some embodiments. FIG9 illustrates a difference threshold as previously described. FIG9 also illustrates a relative difference. The relative difference corresponds to the difference between the current scan image and the current reference image. The reference image can be updated with the most recent scan results after each inspection. The combination of the difference threshold and the relative difference can help ensure that a reticle is not erroneously determined to be defective. Graph 900 corresponds to a reticle whose reflectivity increases over time. Graph 902 corresponds to a reticle whose reflectivity decreases over time. If the relative difference is within the difference threshold, no defect is detected. If the relative difference is outside the difference threshold, a defect is detected. As previously described, the characteristics of the reticle over time can change. Additionally, the characteristics of the reticle inspection system can change over time. The use of a difference threshold can help ensure that these variations do not lead to erroneous determinations of defects in the reticle. FIG10 is a graph 1000 illustrating reflectivity and usage level of a reticle according to some embodiments. Reticle characteristics may vary depending on usage conditions or usage level, such as exposure time, dose, exposure area, or other parameters. Trend results can be separated by stage. At usage level S1, a reference image is collected. At usage level S2, a scan failure occurs. At usage level S3, a reference image is collected. At usage level S4, a scan failure occurs. At usage level S5, a reference image is collected. FIG11 is a graph 1100 of Z offset / focus (curve 1102) and reflectivity (curve 1104) over time, according to some embodiments. In highly reactive materials, extreme instabilities in properties and performance can occur. Furthermore, some materials with external treatments will exhibit similar behavior. The properties of the reticle are affected by variations in elemental concentrations. Unmasked material treatments can be used in future scan characterization production and reference simulations, resulting in improved performance of the reticle inspection system 103. In graph 1100, reference images were collected at t1, t2, and t4. Scan failures were recorded at t3 and t5. FIG. 12 illustrates principles associated with generating a scanned image of a reticle according to some embodiments. FIG. 12 illustrates three patterns (as seen in a top view of reticle 101) each having grooves A1, A2, and A3 of varying heights (not shown in the top view). FIG. 12 also illustrates features B1, B2, and B3 of varying heights. The reticle inspection system 103 identifies positional changes for focusing on each pattern. In the example of FIG. 12, a zero offset corresponds to a preset offset of the objective lens. A level of 100 nm corresponds to a 100 nm increase in the objective lens to focus on patterns A1 and B1. Patterns A2 and B2 are at the preset level. Patterns A3 and B3 are at a level corresponding to a 100 nm decrease in height. Optimal focus can be selected for scanning a particular pattern. These patterns can be formed in the EUV alignment area. In some potential solutions, the image in focus can be evaluated by leveling the objective lens. However, according to some embodiments herein, the patterns are designed to have different heights. Therefore, it is not necessary to scan a particular pattern several times to find the focus. The focus can be found by a single scan of patterns of different heights. For example, patterns with heights of -100 nm, -50 nm, -0 nm, +50 nm, and +100 nm. Patterns with different heights can be arranged in groups. For example, a horizontal pattern with a height from -100 nm to +100 nm. A cubic pattern can have a height from -100 nm to +100 nm. FIG13 is a top view of reticle 101 according to some embodiments. The pattern area is the central area of ​​reticle 101. However, alignment marks, focus areas (e.g., A1-B3), and shadow areas are arranged around the perimeter of the pattern area. Reticle inspection system 103 can utilize a six-point shadow calibration that uses various purple areas and pattern areas in the shadow calibration. This can be utilized by reticle inspection system 103 when inspecting reticles. In some potential solutions, the shadows of the collected images are calibrated using shadow and alignment points. However, the shadow areas may not have been exposed, but the EUV dose in the alignment areas is proportionally different from that in the patterned areas. Calibration may not fully correct for brightness variations in the patterned areas. According to some embodiments, points 1 and 2 are virtually infinite; these points can help define the camera response relative to target brightness. The grayscale of points 3, 4, and 6 can be used to set the performance function after exposure. Point 1 has the same pattern design as point 3. Point 2 has the same pattern design as point 4. Because points 5 and 6 are identical pattern designs on different areas with ratiometric dose exposure, the function indicates that the pattern area brightness performance and the alignment area performance should have the same slope and different constant values. FIG14 illustrates graphs 1400 and 1402 associated with the shading calibration described with reference to FIG13 , according to some embodiments. Graph 1400 represents camera calibration. Because points 1 and 2 are infinite, point 1 can be calibrated to the darkest possible value and point 2 can be defined as the brightest possible value. Through this analog-to-digital conversion, the signal responses at points 3 and 4 can be obtained. The ratios of point 3 to the recipe setting and point 4 to the recipe setting can be calculated. In graph 1402, points 3 and 4 are set to the darkest and brightest values ​​to set the function representing brightness performance after exposure. The ratio calculated over the most recent segment helps define the target response in the new function. Additionally, the signals at points 5 and 6 can be read under this condition. Because the pattern area and alignment area have proportional dose exposures, the function representing pattern area performance should provide the same slope as the alignment area function. Points 5 and 6 represent the same pattern design but in different exposure areas. The analog signal at point 6 is considered the X on the function after the digital signal at point 5 is Y. A constant value for the function can then be calculated, and the pattern area function can be set. The function representing the correction result can be calculated by fitting point 5 on the pattern area function to the digital signal of the recipe setting target on the alignment area function. Curve 1404 corresponds to the alignment area function. Curve 1403 corresponds to the pattern area function. FIG15 is a flow chart of a method 1500 according to some embodiments. Method 1500 may utilize the processes, components, and systems described with respect to FIG1 through FIG14 . At 1502, method 1500 includes training an analysis model for a reticle inspection system using a machine learning process. An example of an analysis model is analysis model 130 of FIG1 . An example of a reticle inspection system is the reticle inspection system of FIG1 . At 1504, method 1500 includes generating a difference threshold based on the machine learning process. At 1506, method 1500 includes storing a reference image. An example of a reference image is reference image 132 of FIG1 . At 1508, method 1500 includes generating a first scanned image of a reticle using the reticle inspection system. At 1510, method 1500 includes generating a first relative difference value by comparing the first scanned image to the reference image. At 1512, method 1500 includes determining that the reticle is not defective if the first relative difference value is less than the difference threshold. At 1514 , method 1500 includes determining that the reticle is defective if the first relative difference value is greater than a difference threshold value. FIG16 is a flow chart of a method 1600 according to some embodiments. The method 1600 may utilize the processes, components, and systems described with respect to FIG1 through FIG15 . At 1602, the method 1600 includes training an analysis model of a reticle inspection system using a machine learning process to detect whether a reticle is contaminated with debris particles and to classify the type of debris particles. An example of a reticle inspection system is the reticle inspection system 103 of FIG1 . An example of an analysis model is the analysis model 130 of FIG1 . At 1604, the method 1600 includes generating a scanned image of the reticle using the reticle inspection system. An example of a scanned image is the scanned image 134 of FIG1 . At 1606, the method 1600 includes providing the scanned image to the analysis model. At 1608, the method 1600 includes determining whether the reticle is contaminated with debris particles based on the machine learning process by processing the scanned image using the analysis model. At 1610, method 1600 includes, if the reticle is contaminated with debris particles, determining the type of debris particles using an analytical model. Embodiments of the present disclosure provide systems and methods for reliably inspecting photolithography reticles. Embodiments of the present disclosure utilize a reticle inspection system that includes an analytical model trained using machine learning to assist in inspecting photolithography reticles between exposures. The reticle inspection system stores one or more reference images of the reticle. During the inspection process, the reticle inspection system captures new scans of the reticle and compares them to the reference images. The reticle inspection system utilizes the analytical model to ensure that inspection is accurate and efficient, so that the reticle is not incorrectly determined to be defective. In some embodiments, a reticle inspection system may train an analysis model using a training set that includes images or scans of both defective and non-defective reticles. After training, the analysis model may analyze the scan of the reticle to determine whether the reticle is defective. In some embodiments, the analysis model may determine whether the reticle is defective and may also help determine the type of defect. The analysis model may also be used to determine whether the current scan of the reticle should be used to partially or completely overwrite one or more reference images used by the reticle inspection system. In some embodiments, reticle characteristic parameters and reticle inspection system parameters may be provided to a machine learning database used to train the analysis model. The analysis model may then generate simulated reference images to aid in future inspections of the reticle. This may help reduce false defect counts and improve inspection tool quality. Embodiments of the present disclosure provide several benefits. Specifically, the number of false defect counts is significantly reduced or completely eliminated. This results in fewer instances of unnecessary reticle cleaning and wafer inspection. This helps ensure that wafer processing is not stopped, resulting in more processed wafers and better wafer yield over time. Expensive reticle cleaning is also avoided. In one embodiment, a method for operating a reticle inspection system includes training an analysis model of the reticle inspection system using a machine learning process, generating a difference threshold based on the machine learning process, storing a reference image, and generating a first scan image of the reticle using the reticle inspection system. The method includes generating a first relative difference value by comparing the first scan image to the reference image. The method includes determining that the reticle is not defective if the first relative difference value is less than the difference threshold. The method includes determining that the reticle is defective if the first relative difference value is greater than the difference threshold. In one embodiment, a reticle inspection system includes scanning resources for capturing scanned images of a reticle; memory resources for storing reference images; and an analysis model trained using machine learning processing to generate a difference threshold, generating a relative difference value by comparing the scanned image with the reference image, and overwriting at least a portion of the reference image if the relative difference value is less than the difference threshold. In one embodiment, a method includes training an analysis model of a reticle inspection system using machine learning to detect whether a reticle is contaminated with debris particles and classify the type of debris particles. The method includes generating a scanned image of the reticle using the reticle inspection system and providing the scanned image to the analysis model. The method includes processing the scanned image and determining, using the analysis model based on the machine learning processing, whether the reticle is contaminated with debris particles, and, if so, determining, using the analysis model, the type of debris particles. The foregoing summarizes the features of several embodiments so that those skilled in the art can better understand the aspects of the present disclosure. Those skilled in the art will appreciate that they can readily use this disclosure as a basis for designing or modifying other processes and structures for implementing the same purposes and / or achieving the same advantages of the embodiments described herein. Those skilled in the art will also recognize that such equivalent constructions do not depart from the spirit and scope of the present disclosure, and that they can make various changes, substitutions, and modifications without departing from the spirit and scope of the present disclosure. 100: EUV lithography system 101: Reticle 102: Droplet generator 103: Reticle inspection system 104: EUV generation chamber 106: Droplet receiver 108: Scanner 111: Laser 116: Wafer 117: EUV light 121: Reticle storage 123: Wafer storage 125: Transfer system 127: Reticle cleaning system 130: Analysis model 132: Reference image 134: Scanned image 140: Substrate 142: Reflective multilayer 144: Buffer layer 146: Absorption layer 148: First absorption layer 150: Second absorption layer 156: Groove 160: Training module 162: Training set data 166: Data integration server 168: Processing resources 170: Memory resources 172: Communication resources 174: Scanning resources 400: Methods 402, 404, 406, 408: Step 500: Methods 502, 504, 506, 508, 510, 512, 514: Step 6 00: Method 602, 604, 606, 608, 610, 612, 614, 616, 618, 620, 622, 624: Step 700: Method 702, 704, 706, 708, 710, 712, 714, 716, 718, 720, 722, 724, 726, 728, 730, 732: Step 800: Method 802, 804, 806, 808, 810, 811, 812 ,814,816,818,820,822,824,826:Step 900:Chart 902:Chart 1000:Chart 1100:Chart 1102:Curve 1104:Curve 1400,1402:Chart 1500:Method 1502,1504,1506,1508,1510,1512,1514:Step 1600:Method 1602,1604,1606,1608,1610:Step The aspects of the present disclosure are better understood from the following detailed description when read in conjunction with the accompanying drawings. It should be noted that, in accordance with standard practice in the industry, various features are not depicted to scale. In fact, the dimensions of various features may be arbitrarily increased or decreased for clarity of discussion. FIG. 1 is a block diagram of an extreme ultraviolet (EUV) photolithography system according to some embodiments. FIG. 2 is a cross-sectional view of an EUV mask according to some embodiments. FIG. 3 is a block diagram of a reticle inspection system according to some embodiments. FIG. 4 is a flow chart of a method for operating a reticle inspection system according to some embodiments. FIG. 5 is a flow chart of a method for operating a reticle inspection system according to some embodiments. FIG. 6 is a flow chart of a method for operating a reticle inspection system according to some embodiments. FIG. 7 is a flow chart of a method for operating a reticle inspection system according to some embodiments. FIG. 8 is a flow chart of a method for operating a reticle inspection system according to some embodiments. FIG. 9 includes a graph illustrating reflectivity of a reticle according to some embodiments. FIG. 10 includes graphs illustrating reflectivity of a photomask according to some embodiments. FIG. 11 includes graphs illustrating reflectivity and Z-offset of a reticle according to some embodiments. FIG. 12 illustrates a reticle pattern and focus level associated with a reticle inspection process according to some embodiments. FIG. 13 is a top view of a photomask according to some embodiments. FIG. 14 includes diagrams associated with calibrating a reticle inspection system in accordance with some embodiments. FIG. 15 is a flow chart of a method for operating a reticle inspection system according to some embodiments. FIG. 16 is a flow chart of a method for operating a reticle inspection system according to some embodiments. Domestic deposit information (please note in the order of deposit organization, date, and number) None Foreign deposit information (please note in the order of deposit country, organization, date, and number) None 100: Extreme Ultraviolet Lithography System 101: Mask 102: Droplet Generator 103: Mask inspection system 104: Extreme ultraviolet light generation chamber 106: Droplet Receiver 108: Scanner 111: Laser 116: Wafer 117: Extreme Ultraviolet Light 121: Mask storage 123: Wafer Storage 125: Delivery System 127:Mask cleaning system 130:Analysis Model 132: Reference Image 134: Scanning images

Claims

1. A method for inspecting a photomask, comprising the following steps: training an analysis model of a photomask inspection system using a machine learning process; generating a difference threshold based on the machine learning process; storing a reference image; generating a first scan image of the photomask using the photomask inspection system; generating a first relative difference value by comparing the first scan image with the reference image; if the first relative difference is less than or equal to the difference threshold, determining that the photomask is not defective and overwriting at least a portion of the reference image with the first scan image; if the first relative difference value is greater than the difference threshold, determining that the photomask is defective; after overwriting at least a portion of the reference image with the first scan image, performing a photomask process on the photomask; generating a second scan image of the photomask using the photomask inspection system; generating a second relative difference value by comparing the second scan image with the reference image; if the second relative difference value is less than or equal to the difference threshold, determining that the photomask is not defective; and if the second relative difference value is greater than the difference threshold, determining that the photomask is defective.

2. The method for inspecting a photomask as described in claim 1 includes the following steps: if the first relative difference value is less than or equal to the difference threshold value, then the entire reference image is overwritten with the first scan image.

3. The method for inspecting a photomask as described in claim 1, comprising the following steps: generating a ratio of an error defect area of ​​the first scan image to a total pattern area of ​​the first scan image based on comparing the first scan image with the reference image; and comparing the ratio with a threshold ratio.

4. The method for inspecting a photomask as described in claim 3, comprising the following steps: if the ratio is greater than the critical ratio, then overwriting an entire portion of the reference image with the first scan image; and if the ratio is less than or equal to the critical ratio, then overwriting only a portion of the reference image with the defective area of ​​the first scan image.

5. The method for inspecting a photomask as described in claim 2, comprising the following steps: If the second relative difference value is less than or equal to the difference threshold value, then at least a portion of the reference image is overwritten with the second scan image.

6. The method for inspecting a photolithography mask as described in claim 1, wherein the step of training the analysis model includes the following steps: training the analysis model to generate a complex simulated reference image.

7. A method for inspecting a photomask, comprising the following steps: training an analysis model of a photomask inspection system using a machine learning process; generating a difference threshold based on the machine learning process; storing a reference image; generating a first scan image of the photomask using the photomask inspection system; generating a first relative difference value by comparing the first scan image with the reference image; determining that the photomask is not defective if the first relative difference is less than or equal to the difference threshold; determining that the photomask is defective if the first relative difference value is greater than the difference threshold; generating a first simulated reference image using the analysis model and storing the first simulated reference image as the reference image before generating the first scan image; performing a photomask process on the photomask if the first relative difference value is less than or equal to the difference threshold; generating a second simulated reference image after performing the photomask process; and replacing the reference image with the second simulated reference image.

8. An inspection system for a photomask, comprising: a scanning resource for capturing a scanned image of a photomask; a memory resource for storing a reference image; and an analysis model trained with a machine learning process to generate a difference threshold, generating a relative difference value by comparing the scanned image with the reference image, and if the relative difference value is less than or equal to the difference threshold, overwriting at least a portion of the reference image, generating a first simulated reference image using the analysis model, storing the first simulated reference image as the reference image before generating the scanned image, performing a photomask process on the photomask if the relative difference value is less than or equal to the difference threshold, generating a second simulated reference image after performing the photomask process, and replacing the reference image with the second simulated reference image.

9. The inspection system for photolithography masks as described in claim 8, wherein the difference threshold is a mask reflectivity difference threshold or a mask brightness difference threshold.

10. A method for inspecting a photomask, comprising the following steps: training an analysis model of a photomask inspection system using a machine learning process to detect whether a photomask is contaminated by a debris particle and classifying a type of the debris particle; generating a scanned image of the photomask using the photomask inspection system; providing the scanned image to the analysis model; determining, based on the machine learning process, whether the photomask is contaminated by a debris particle by processing the scanned image; if the photomask is contaminated by a debris particle, determining a type of the debris particle using the analysis model; generating a first simulated reference image using the analysis model and storing the first simulated reference image as a reference image before generating the scanned image; if the photomask is not defective, performing a photomask process on the photomask; generating a second simulated reference image after performing the photomask process; and replacing the reference image with the second simulated reference image.

Images