Diffusion model in scanning charged-particle microscope image quality enhancement

Abstract

An apparatus for determining measurements of a feature on a wafer from a denoised image includes a memory storing a set of instructions and at least one processor configured to execute the set of instructions to cause the apparatus to perform: applying a trained diffusion neural network model to denoise scanning charged-particle microscope (SCPM) images, wherein the trained diffusion neural network model has been trained to denoise a received SCPM image of the wafer; determining an averaged image as an average of the denoised SCPM images; and determining measurements of the feature based on the averaged image.

Description

DIFFUSION MODEL IN SCANNING CHARGED-PARTICLE MICROSCOPE IMAGE QUALITYENHANCEMENTCROSS-REFERENCE TO RELATED APPLICATIONS

[0001] This application claims priority of US application 63 / 702,927 which was filed on 3 October 2024, and which is incorporated herein in its entirety by reference.TECHNICAL FIELD

[0002] The embodiments provided herein relate to image quality enhancement, and more particularly to using a diffusion model to enhance image quality for scanning charged-particle microscope images.BACKGROUND

[0003] In manufacturing processes of integrated circuits (ICs), unfinished or finished circuit components are inspected to ensure that they are manufactured according to design and are free of defects. Inspection systems utilizing optical microscopes or charged particle (e.g., electron) beam microscopes, such as a scanning electron microscope (SEM) can be employed. As the physical sizes of IC components continue to shrink, accuracy and yield in defect detection become more important. Various metrology tools are developed and used to check whether the ICs are correctly manufactured.

[0004] When performing metrology on an IC after manufacturing, features of the IC may be measured from images taken of the IC. To be able to take multiple measurements from an image, a clear image is desired.SUMMARY

[0005] Some embodiments provide an apparatus for performing operations for determining measurements of a feature on a wafer from a denoised image. The apparatus can include a memory storing a set of instructions and at least one processor configured to execute the set of instructions to cause the apparatus to perform: applying a trained diffusion neural network model to denoise scanning charged-particle microscope (SCPM) images, wherein the trained diffusion neural network model has been trained to denoise a received SCPM image of the wafer; determining an averaged image as an average of the denoised SCPM images; and determining measurements of the feature based on the averaged image.

[0006] Other advantages of the embodiments of the present disclosure will become apparent from the following description taken in conjunction with the accompanying drawings wherein are set forth, by way of illustration and example, certain embodiments of the present invention.BRIEF DESCRIPTION OF FIGURES

[0007] The above and other aspects of the present disclosure will become more apparent from the description of exemplary embodiments, taken in conjunction with the accompanying drawings.

[0008] Fig. 1 is an example of a scanning charged-particle microscope (SCPM) image.

[0009] Fig. 2 is a schematic diagram illustrating an example SCPM system, consistent with some embodiments of the present disclosure.

[0010] Fig. 3 is a schematic diagram illustrating an example charged-particle beam tool, consistent with some embodiments of the present disclosure that may be a part of the example SCPM system of Fig. 2.

[0011] Fig. 4 is a schematic diagram illustrating an example multi-beam tool, consistent with embodiments of the present disclosure that may be a part of the example SCPM system of Fig. 2.

[0012] Fig. 5 is a block diagram of an exemplary server, consistent with some embodiments of the present disclosure.

[0013] Fig. 6 is a schematic diagram illustrating an example neural network, consistent with some embodiments of the present disclosure.

[0014] Fig. 7 is an example of a process fortraining a diffusion model using a SCPM image, consistent with some embodiments of the present disclosure.

[0015] Fig. 8 is an example of a denoising process using a trained diffusion model, consistent with some embodiments of the present disclosure.

[0016] Fig. 9 is an example of measurements that may be taken from an averaged raw image and an averaged denoised SCPM image, consistent with some embodiments of the present disclosure.

[0017] Fig. 10 is a flowchart of an example method for using a diffusion model with SCPM images, consistent with embodiments of the present disclosure.DETAILED DESCRIPTION

[0018] Reference will now be made in detail to exemplary embodiments, examples of which are illustrated in the accompanying drawings. The following description refers to the accompanying drawings in which the same numbers in different drawings represent the same or similar elements unless otherwise represented. The implementations set forth in the following description of exemplary embodiments do not represent all implementations. Instead, they are merely examples of apparatuses and methods consistent with aspects related to the disclosed embodiments as recited in the appended claims. For example, although some embodiments are described in the context of utilizing electron beams, the disclosure is not so limited. Other types of charged-particle beams (e.g., including protons, ions, muons, or any other particle carrying electric charges) may be similarly applied. Furthermore, other imaging systems may be used, such as optical imaging, photon detection, x-ray detection, ion detection, etc.

[0019] Relative dimensions of components in drawings may be exaggerated for clarity. Within the following description of drawings, the same or like reference numbers refer to the same or like components or entities, and only the differences with respect to the individual embodiments are described. As used herein, unless specifically stated otherwise, the term “or” encompasses all possible combinations, except where infeasible. For example, if it is stated that a component may include A or B, then, unless specifically stated otherwise or infeasible, the component may include A, or B, or A and B. As a second example, if it is stated that a component may include A, B, or C, then, unless specifically stated otherwise or infeasible, the component may include A, or B, or C, or A and B, or A and C, or B and C, or A and B and C.

[0020] Electronic devices are constructed of circuits formed on a piece of semiconductor material called a substrate. The semiconductor material may include, for example, silicon, gallium arsenide, indium phosphide, or silicon germanium, or the like. Many circuits may be formed together on the same piece of silicon and are called integrated circuits or ICs. The size of these circuits has decreased dramatically so that many more of them can be fit on the substrate. For example, an IC chip in a smartphone can be as small as a thumbnail and yet may include over 2 billion transistors, the size of each transistor being less than 1 / 1000th the size of a human hair.

[0021] Making these ICs with extremely small structures or components is a complex, timeconsuming, and expensive process, often involving hundreds of individual steps. Errors in even one step have the potential to result in defects in the finished IC, rendering it useless. Thus, one goal of the manufacturing process is to avoid such defects to maximize the number of functional ICs made in the process; that is, to improve the overall yield of the process.

[0022] One component of improving yield is monitoring the chip-making process to ensure that it is producing a sufficient number of functional integrated circuits. One way to monitor the process is to inspect the chip circuit structures at various stages of their formation. Inspection can be carried out using a scanning charged-particle microscope (SCPM). For example, an SCPM may be a scanning electron microscope (SEM). An SCPM can be used to image these extremely small structures, in effect, taking a “picture” of the structures of the wafer. The image can be used to determine if the structure was formed properly in the proper location. If the structure is defective, then the process can be adjusted, so the defect is less likely to recur.

[0023] The working principle of an SCPM (e.g., an SEM) is similar to a camera. A camera takes a picture by receiving and recording intensity of light reflected or emitted from people or objects. An SCPM takes a “picture” by receiving and recording energies or quantities of charged particles (e.g., electrons) reflected or emitted from the structures of the wafer. Typically, the structures are made on a substrate (e.g., a silicon substrate) that is placed on a platform, referred to as a stage, for imaging. Before taking such a “picture,” a charged-particle beam may be projected onto the structures, and when the charged particles are reflected or emitted (“exiting”) from the structures (e.g., from the wafer surface, from the structures underneath the wafer surface, or both), a detector of the SCPM mayreceive and record the energies or quantities of those charged particles to generate an inspection image. To take such a “picture,” the charged-particle beam may scan through the wafer (e.g., in a line- by-line or zig-zag manner), and the detector may receive exiting charged particles coming from a region under charged particle-beam projection (referred to as a “beam spot”). The detector may receive and record exiting charged particles from each beam spot one at a time and join the information recorded for all the beam spots to generate the inspection image. Some SCPMs use a single charged-particle beam (referred to as a “single-beam SCPM,” such as a single-beam SEM) to take a single “picture” to generate the inspection image, while some SCPMs use multiple charged- particle beams (referred to as a “multi-beam SCPM,” such as a multi-beam SEM) to take multiple “sub-pictures” of the wafer in parallel and stitch them together to generate the inspection image. By using multiple charged-particle beams, the SCPM may provide more charged-particle beams onto the structures for obtaining these multiple “sub-pictures,” resulting in more charged particles exiting from the structures. Accordingly, the detector may receive more exiting charged particles simultaneously and generate inspection images of the structures of the wafer with higher efficiency and faster speed.

[0024] Generating and processing these images to determine whether any defects exist (sometimes as small as the nanometer scale) are computationally intensive. And as the physical sizes of IC components continue to shrink, accuracy and yield in defect detection become more important. To inspect a single wafer, it is not uncommon for an inspection system to generate and process a substantial number of images. For example, if each image taken corresponds to 6 pm x 6 pm portion of a wafer, for a 200mm wafer, it would take over 872 million images to image the entire wafer. If these images are not processed and evaluated efficiently, not surprisingly, yield will be dramatically impacted, thereby affecting the wafer throughput.

[0025] Metrology tools can be used to determine whether the ICs are correctly manufactured by identifying a number of defects on each wafer, including at different levels of detail, such as a pattern level, an image (field of view) level, a die level, a care area level, or a wafer level. Some SCPM images (for example, image 100 as shown in Fig. 1) may not have clear edges. For example, as shown in image 100, there is noise (e.g., the white static-like points) and the lines of the features are not clear (e.g., the lines of the features are fuzzy). With such noise in the image, it may be difficult to accurately detect the edges (e.g., lines) of the features by a scanning tool. If the feature edges cannot be accurately determined, then measurements derived from the images and the detected feature edges may also not be accurate. Accurate measurements derived from the images are useful in detecting wafer defects. If the image 100 could be cleaned, the edges may become clearer so the scanning tool can easily detect the edges (e.g., detect the edges with a greater accuracy). More accurately detecting the edges makes it easier to determine measurements from the image and therefore easier to detect defects on the wafer.

[0026] A prior method of performing image enhancement includes training a machine learning (ML) model on a lower quality SCPM image and a higher quality SCPM image. The trained ML model maybe used to apply an enhancement to the lower quality SCPM image to obtain the higher quality SCPM image. One problem with this approach is that for proper ML model training, the lower quality SCPM image and the higher quality SCPM image need to be properly aligned. Between a time of obtaining the lower quality SCPM image and a time of obtaining the higher quality SCPM image, there may be wafer movement, which may cause the two SCPM images to be misaligned. For example, the SCPM images may not start in the same location or the scan direction of the SCPM tool when creating the images may be different. Either of these situations may result in a misalignment between the lower quality SCPM image and the higher quality SCPM image.

[0027] Wafer defects are generally rare (e.g., defects do not occur everywhere on a wafer), so when the lower quality SCPM image and the higher quality SCPM image are compared, the “right” defect location needs to be found to better align the images. After image alignment, supervised learning for the ML model may be performed. If the SCPM images are not correctly aligned, the supervised learning by the ML model will result in incorrect information. For example, in the situation where there is a defect in the lower quality SCPM image but not in the higher quality SCPM image, a poor alignment of the images may result in a defect being missed. Because both images do not include the defect, the lower quality SCPM image will be filtered out by the ML model (because the goal of applying the ML model is to obtain the higher quality image). As another example, in the situation where there is a defect in the higher quality SCPM image but not in the lower quality SCPM image, a poor alignment of the images may result in a number of “false” defects being detected. In this circumstance, the ML model will create an artificial “forced” defect because of the poor alignment between the images. In both of these examples, the ML model ends up being improperly trained, leading to incorrect or inaccurate results during inference.

[0028] Embodiments of the present disclosure can provide a trained diffusion model to denoise a SCPM image. One of the advantages of the trained diffusion model is that it does not require image alignment for proper model training, which can help with reducing processing times and thereby improve throughput. In particular, the trained diffusion model may be applied to remove noise from SCPM images. Removing the noise from the SCPM images results in clearer images. The denoised SCPM images may be averaged together to obtain an average denoised image. Measurements may be taken based on the average denoised image, such as gauge, critical dimension, and edge placement error. The measurements that may be taken from the average denoised image may have a higher accuracy than measurements taken from images that have not been denoised. Obtaining higher- accuracy measurements helps later metrology and manufacturing processes.

[0029] Fig. 2 illustrates an exemplary scanning charged-particle microscope (SCPM) system 200 consistent with some embodiments of the present disclosure. SCPM system 200 may be used for imaging. For example, SCPM system 200 may use an electron beam for imaging. As shown in Fig. 2, SCPM system 200 includes a main chamber 201, a load / lock chamber 202, a beam tool 204, and an equipment front end module (EFEM) 206. Beam tool 204 is located within main chamber 201. EFEM206 includes a first loading port 206a and a second loading port 206b. EFEM 206 may include additional loading port(s). First loading port 206a and second loading port 206b receive wafer front opening unified pods (FOUPs) that contain wafers (e.g., semiconductor wafers or wafers made of other material(s)) or samples to be inspected (the terms “wafers” and “samples” may be used interchangeably). A “lot” is a plurality of wafers that may be loaded for processing as a batch.

[0030] One or more robotic arms (not shown) in EFEM 206 may transport the wafers to load / lock chamber 202. Load / lock chamber 202 is connected to a load / lock vacuum pump system (not shown) which removes gas molecules in load / lock chamber 202 to reach a first pressure below the atmospheric pressure. After reaching the first pressure, one or more robotic arms (not shown) may transport the wafer from load / lock chamber 202 to main chamber 201. Main chamber 201 is connected to a main chamber vacuum pump system (not shown) which removes gas molecules in main chamber 201 to reach a second pressure below the first pressure. After reaching the second pressure, the wafer is subject to inspection by beam tool 204. Beam tool 204 may be a single-beam system or a multi -beam system.

[0031] A controller 209 is electronically connected to beam tool 204. Controller 209 may be a computer that may execute various controls of SCPM system 200. While controller 209 is shown in Fig. 2 as being outside of the structure that includes main chamber 201, load / lock chamber 202, and EFEM 206, it is appreciated that controller 209 may be a part of the structure.

[0032] In some embodiments, controller 209 may include one or more processors (not shown). A processor may be a generic or specific electronic device capable of manipulating or processing information. For example, the processor may include any combination of any number of a central processing unit (or “CPU”), a graphics processing unit (or “GPU”), an optical processor, a programmable logic controller, a microcontroller, a microprocessor, a digital signal processor, an intellectual property (IP) core, a Programmable Logic Array (PLA), a Programmable Array Logic (PAL), a Generic Array Logic (GAL), a Complex Programmable Logic Device (CPLD), a Field- Programmable Gate Array (FPGA), a System On Chip (SoC), an Application-Specific Integrated Circuit (ASIC), and any type circuit capable of data processing. The processor may also be a virtual processor that includes one or more processors distributed across multiple machines or devices coupled via a network.

[0033] In some embodiments, controller 209 may further include one or more memories (not shown). A memory may be a generic or specific electronic device capable of storing codes and data accessible by the processor (e.g., via a bus). For example, the memory may include any combination of any number of a random-access memory (RAM), a read-only memory (ROM), an optical disc, a magnetic disk, a hard drive, a solid-state drive, a flash drive, a security digital (SD) card, a memory stick, a compact flash (CF) card, or any type of storage device. The codes may include an operating system (OS) and one or more application programs (or “apps”) for specific tasks. The memory may also be avirtual memory that includes one or more memories distributed across multiple machines or devices coupled via a network.

[0034] Fig. 3 illustrates an example imaging system 300 consistent with some embodiments of the present disclosure. Beam tool 204 of Fig. 3 may be configured for use in SCPM system 200. Beam tool 204 may be a single beam apparatus or a multi-beam apparatus. As shown in Fig. 3, beam tool 204 includes a motorized sample stage 301, and a wafer holder 302 supported by motorized sample stage 301 to hold a wafer 303 to be inspected. Beam tool 204 further includes an objective lens assembly 304, a charged-particle detector 306 (which includes charged-particle sensor surfaces 306a and 306b), an objective aperture 308, a condenser lens 310, a beam limit aperture 312, a gun aperture 314, an anode 316, and a cathode 318. Objective lens assembly 304, in some embodiments, may include a modified swing objective retarding immersion lens (SORIL), which includes a pole piece 304a, a control electrode 304b, a deflector 304c, and an exciting coil 304d. Beam tool 204 may additionally include an Energy Dispersive X-ray Spectrometer (EDS) detector (not shown) to characterize the materials on wafer 303.

[0035] A primary charged-particle beam 320 (or simply “primary beam 320”), such as an electron beam, is emitted from cathode 318 by applying an acceleration voltage between anode 316 and cathode 318. Primary beam 320 passes through gun aperture 314 and beam limit aperture 312, both of which may determine the size of charged-particle beam entering condenser lens 310, which resides below beam limit aperture 312. Condenser lens 310 focuses primary beam 320 before the beam enters objective aperture 308 to set the size of the charged-particle beam before entering objective lens assembly 304. Deflector 304c deflects primary beam 320 to facilitate beam scanning on the wafer. For example, in a scanning process, deflector 304c may be controlled to deflect primary beam 320 sequentially onto different locations of top surface of wafer 303 at different time points, to provide data for image reconstruction for different parts of wafer 303. Moreover, deflector 304c may also be controlled to deflect primary beam 320 onto different sides of wafer 303 at a particular location, at different time points, to provide data for stereo image reconstruction of the wafer structure at that location. Further, in some embodiments, anode 316 and cathode 318 may generate multiple primary beams 320, and beam tool 204 may include a plurality of deflectors 304c to project the multiple primary beams 320 to different parts / sides of the wafer at the same time, to provide data for image reconstruction for different parts of wafer 303.

[0036] Exciting coil 304d and pole piece 304a generate a magnetic field that begins at one end of pole piece 304a and terminates at the other end of pole piece 304a. A part of wafer 303 being scanned by primary beam 320 may be immersed in the magnetic field and may be electrically charged, which, in turn, creates an electric field. The electric field reduces the energy of impinging primary beam 320 near the surface of wafer 303 before it collides with wafer 303. Control electrode 304b, being electrically isolated from pole piece 304a, controls an electric field on wafer 303 to prevent microarching of wafer 303 and to ensure proper beam focus.

[0037] A secondary charged-particle beam 322 (or “secondary beam 322”), such as secondary electron beams, may be emitted from the part of wafer 303 upon receiving primary beam 320. Secondary beam 322 may form a beam spot on sensor surfaces 306a and 306b of charged-particle detector 306. Charged-particle detector 306 may generate a signal (e.g., a voltage, a current, or the like) that represents an intensity of the beam spot and provide the signal to an image processing system 350. The intensity of secondary beam 322, and the resultant beam spot, may vary according to the external or internal structure of wafer 303. Moreover, as discussed above, primary beam 320 may be projected onto different locations of the top surface of the wafer or different sides of the wafer at a particular location, to generate secondary beams 322 (and the resultant beam spot) of different intensities. Therefore, by mapping the intensities of the beam spots with the locations of wafer 303, the processing system may reconstruct an image that reflects the internal or surface structures of wafer 303.

[0038] Imaging system 300 may be used for inspecting a wafer 303 on motorized sample stage 301 and includes beam tool 204, as discussed above. Imaging system 300 may also include an image processing system 350 that includes an image acquirer 360, storage 370, and controller 209. Image acquirer 360 may include one or more processors. For example, image acquirer 360 may include a computer, server, mainframe host, terminals, personal computer, any kind of mobile computing devices, and the like, or a combination thereof. Image acquirer 360 may connect with a detector 306 of beam tool 204 through a medium such as an electrical conductor, optical fiber cable, portable storage media, IR, Bluetooth, internet, wireless network, wireless radio, or a combination thereof. Image acquirer 360 may receive a signal from detector 306 and may construct an image. Image acquirer 360 may thus acquire images of wafer 303. Image acquirer 360 may also perform various post-processing functions, such as generating contours, superimposing indicators on an acquired image, and the like. Image acquirer 360 may perform adjustments of brightness and contrast, or the like of acquired images. Storage 370 may be a storage medium such as a hard disk, cloud storage, random access memory (RAM), other types of computer readable memory, and the like. Storage 370 may be coupled with image acquirer 360 and may be used for saving scanned raw image data as original images, post-processed images, or other images assisting of the processing. Image acquirer 360 and storage 370 may be connected to controller 209. In some embodiments, image acquirer 360, storage 370, and controller 209 may be integrated together as one control unit.

[0039] In some embodiments, image acquirer 360 may acquire one or more images of a sample based on an imaging signal received from detector 306. An imaging signal may correspond to a scanning operation for conducting charged particle imaging. An acquired image may be a single image including a plurality of imaging areas. The single image may be stored in storage 370. The single image may be an original image that may be divided into a plurality of regions. Each of the regions may include one imaging area containing a feature of wafer 303.

[0040] Consistent with some embodiments of this disclosure, a computer-implemented method of using a diffusion model to enhance image quality for scanning charged-particle microscope images may include obtaining training data that includes an inspection image of a fabricated integrated circuit (IC) and design layout data of the IC. The obtaining operation, as used herein, may refer to accepting, taking in, admitting, gaining, acquiring, retrieving, receiving, reading, accessing, collecting, or any operation for inputting data. An inspection image, as used herein, may refer to an image generated as a result of an inspection process performed by a scanning charged-particle microscope apparatus (e.g., system 200 of Fig. 2 or system 300 of Fig. 3). For example, an inspection image may be an SCPM image generated by image processing system 350 in Fig. 3. A fabricated IC in this disclosure may refer to an IC manufactured on a sample (e.g., a wafer) in a semiconductor manufacturing process (e.g., a photolithography process). For example, the fabricated IC may be manufactured in a die of the sample. Design layout data of an IC, as used herein, may refer to data representing a designed layout of the IC. In some embodiments, the design layout data may include a design layout fde in a GDS format (e.g., a GDS layout fde). The design layout fde may be visualized (also referred to as “rendered”) to be a 2D image (referred to as a “rendered image” herein) that presents the layout of the IC. The rendered image may include various geometric features (e.g., vertices, edges, comers, polygons, holes, bridges, vias, or the like) of the IC.

[0041] In some embodiments, the design layout data of the IC may include an image (e.g., the rendered image) rendered based on GDS clip data of the IC. GDS clip data of an IC, as used herein, may refer to design layout data of the IC that is to be fabricated in a die, which is of the GDS format. In some embodiments, the design layout data of the IC may include only a design layout file (e.g., the GDS clip data) of the IC. In some embodiments, the design layout data of the IC may include only the rendered image of the IC. In some embodiments, the design layout data of the IC may include only a golden image of the IC. In some embodiments, the design layout data may include any combination of the design layout file, the golden image, and the rendered image of the IC.

[0042] Fig. 4 illustrates a schematic diagram of an example multi-beam beam tool 204 (also referred to herein as apparatus 204) and an image processing system 490 that may be configured for use in EBI system 200 (Fig. 2), consistent with embodiments of the present disclosure.

[0043] Beam tool 204 comprises a charged-particle source 402, a gun aperture 404, a condenser lens 406, a primary charged-particle beam 410 emitted from charged-particle source 402, a source conversion unit 412, a plurality of beamlets 414, 416, and 418 of primary charged-particle beam 410, a primary projection optical system 420, a motorized wafer stage 480, a wafer holder 482, multiple secondary charged-particle beams 436, 438, and 440, a secondary optical system 442, and a charged- particle detection device 444. Primary projection optical system 420 can comprise a beam separator 422, a deflection scanning unit 426, and an objective lens 428. Charged-particle detection device 444 can comprise detection sub-regions 446, 448, and 450.

[0044] Charged-particle source 402, gun aperture 404, condenser lens 406, source conversion unit 412, beam separator 422, deflection scanning unit 426, and objective lens 428 can be aligned with a primary optical axis 460 of apparatus 204. Secondary optical system 442 and charged-particle detection device 444 can be aligned with a secondary optical axis 452 of apparatus 204.

[0045] Charged-particle source 402 can emit one or more charged particles, such as electrons, protons, ions, muons, or any other particle carrying electric charges. In some embodiments, charged- particle source 402 may be an electron source. For example, charged-particle source 402 may include a cathode, an extractor, or an anode, wherein primary electrons can be emitted from the cathode and extracted or accelerated to form primary charged-particle beam 410 (in this case, a primary electron beam) with a crossover (virtual or real) 408. For ease of explanation without causing ambiguity, electrons are used as examples in some of the descriptions herein. However, it should be noted that any charged particle may be used in any embodiment of this disclosure, not limited to electrons. Primary charged-particle beam 410 can be visualized as being emitted from crossover 408. Gun aperture 404 can block off peripheral charged particles of primary charged-particle beam 410 to reduce Coulomb effect. The Coulomb effect may cause an increase in size of probe spots.

[0046] Source conversion unit 412 can comprise an array of image-forming elements and an array of beam-limit apertures. The array of image-forming elements can comprise an array of micro-deflectors or micro-lenses. The array of image-forming elements can form a plurality of parallel images (virtual or real) of crossover 408 with a plurality of beamlets 414, 416, and 418 of primary charged-particle beam 410. The array of beam-limit apertures can limit the plurality of beamlets 414, 416, and 418. While three beamlets 414, 416, and 418 are shown in Fig. 4, embodiments of the present disclosure are not so limited. For example, in some embodiments, the apparatus 204 may be configured to generate a first number of beamlets. In some embodiments, the first number of beamlets may be in a range from 1 to 1000. In some embodiments, the first number of beamlets may be in a range from 200-500. In some embodiments, an apparatus 204 may generate 400 beamlets.

[0047] Condenser lens 406 can focus primary charged-particle beam 410. The electric currents of beamlets 414, 416, and 418 downstream of source conversion unit 412 can be varied by adjusting the focusing power of condenser lens 406 or by changing the radial sizes of the corresponding beam-limit apertures within the array of beam-limit apertures. Objective lens 428 can focus beamlets 414, 416, and 418 onto a wafer 430 for imaging, and can form a plurality of probe spots 470, 472, and 474 on a surface of wafer 430.

[0048] Beam separator 422 can be a beam separator of Wien filter type generating an electrostatic dipole field and a magnetic dipole field. In some embodiments, if they are applied, the force exerted by the electrostatic dipole field on a charged particle (e.g., an electron) of beamlets 414, 416, and 418 can be substantially equal in magnitude and opposite in a direction to the force exerted on the charged particle by magnetic dipole field. Beamlets 414, 416, and 418 can, therefore, pass straight through beam separator 422 with zero deflection angle. However, the total dispersion of beamlets 414, 416,and 418 generated by beam separator 422 can also be non-zero. Beam separator 422 can separate secondary charged-particle beams 436, 438, and 440 from beamlets 414, 416, and 418 and direct secondary charged-particle beams 436, 438, and 440 towards secondary optical system 442.

[0049] Deflection scanning unit 426 can deflect beamlets 414, 416, and 418 to scan probe spots 470, 472, and 474 over a surface area of wafer 430. In response to the incidence of beamlets 414, 416, and 418 at probe spots 470, 472, and 474, secondary charged-particle beams 436, 438, and 440 may be emitted from wafer 430. Secondary charged-particle beams 436, 438, and 440 may comprise charged particles (e.g., electrons) with a distribution of energies. For example, secondary charged-particle beams 436, 438, and 440 may be secondary electron beams including secondary electrons (energies < 50 eV) and backscattered electrons (energies between 50 eV and landing energies of beamlets 414, 416, and 418). Secondary optical system 442 can focus secondary charged-particle beams 436, 438, and 440 onto detection sub-regions 446, 448, and 450 of charged-particle detection device 444. Detection sub-regions 446, 448, and 450 may be configured to detect corresponding secondary charged-particle beams 436, 438, and 440 and generate corresponding signals (e.g., voltage, current, or the like) used to reconstruct an inspection image of structures on or underneath the surface area of wafer 430.

[0050] The generated signals may represent intensities of secondary charged-particle beams 436, 438, and 440 and may be provided to image processing system 490 that is in communication with charged-particle detection device 444, primary projection optical system 420, and motorized wafer stage 480. The movement speed of motorized wafer stage 480 may be synchronized and coordinated with the beam deflections controlled by deflection scanning unit 426, such that the movement of the scan probe spots (e.g., scan probe spots 470, 472, and 474) may orderly cover regions of interest on the wafer 430. The parameters of such synchronization and coordination may be adjusted to adapt to different materials of wafer 430. For example, different materials of wafer 430 may have different resistance-capacitance characteristics that may cause different signal sensitivities to the movement of the scan probe spots.

[0051] The intensity of secondary charged-particle beams 436, 438, and 440 may vary according to the external or internal structure of wafer 430, and thus may indicate whether wafer 430 includes defects. Moreover, as discussed above, beamlets 414, 416, and 418 may be projected onto different locations of the top surface of wafer 430, or different sides of local structures of wafer 430, to generate secondary charged-particle beams 436, 438, and 440 that may have different intensities. Therefore, by mapping the intensity of secondary charged-particle beams 436, 438, and 440 with the areas of wafer 430, image processing system 490 may reconstruct an image that reflects the characteristics of internal or external structures of wafer 430.

[0052] In some embodiments, image processing system 490 may include an image acquirer 492, a storage 494, and a controller 496. Image acquirer 492 may comprise one or more processors. For example, image acquirer 492 may comprise a computer, server, mainframe host, terminals, personalcomputer, any kind of mobile computing devices, or the like, or a combination thereof. Image acquirer 492 may be communicatively coupled to charged-particle detection device 444 of beam tool 204 through a medium such as an electric conductor, optical fiber cable, portable storage media, IR, Bluetooth, internet, wireless network, wireless radio, or a combination thereof. In some embodiments, image acquirer 492 may receive a signal from charged-particle detection device 444 and may construct an image. Image acquirer 492 may thus acquire inspection images of wafer 430. Image acquirer 492 may also perform various post-processing functions, such as generating contours, superimposing indicators on an acquired image, or the like. Image acquirer 492 may be configured to perform adjustments of brightness and contrast of acquired images. In some embodiments, storage 494 may be a storage medium such as a hard disk, flash drive, cloud storage, random access memory (RAM), other types of computer-readable memory, or the like. Storage 494 may be coupled with image acquirer 492 and may be used for saving scanned raw image data as original images, and postprocessed images. Image acquirer 492 and storage 494 may be connected to controller 496. In some embodiments, image acquirer 492, storage 494, and controller 496 may be integrated together as one control unit.

[0053] In some embodiments, image acquirer 492 may acquire one or more inspection images of a wafer based on an imaging signal received from charged-particle detection device 444. An imaging signal may correspond to a scanning operation for conducting charged particle imaging. An acquired image may be a single image comprising a plurality of imaging areas. The single image may be stored in storage 494. The single image may be an original image that may be divided into a plurality of regions. Each of the regions may comprise one imaging area containing a feature of wafer 430. The acquired images may comprise multiple images of a single imaging area of wafer 430 sampled multiple times over a time sequence. The multiple images may be stored in storage 494. In some embodiments, image processing system 490 may be configured to perform image processing steps with the multiple images of the same location of wafer 430.

[0054] In some embodiments, image processing system 490 may include measurement circuits (e.g., analog-to-digital converters) to obtain a distribution of the detected secondary charged particles (e.g., secondary electrons). The charged-particle distribution data collected during a detection time window, in combination with corresponding scan path data of beamlets 414, 416, and 418 incident on the wafer surface, can be used to reconstruct images of the wafer structures under inspection. The reconstructed images can be used to reveal various features of the internal or external structures of wafer 430, and thereby can be used to reveal any defects that may exist in the wafer.

[0055] In some embodiments, the charged particles may be electrons. When electrons of primary charged-particle beam 410 are projected onto a surface of wafer 430 (e.g., probe spots 470, 472, and 474), the electrons of primary charged-particle beam 410 may penetrate the surface of wafer 430 for a certain depth, interacting with particles of wafer 430. Some electrons of primary charged-particle beam 410 may elastically interact with (e.g., in the form of elastic scattering or collision) the materialsof wafer 430 and may be reflected or recoiled out of the surface of wafer 430. An elastic interaction conserves the total kinetic energies of the bodies (e.g., electrons of primary charged-particle beam 410) of the interaction, in which the kinetic energy of the interacting bodies does not convert to other forms of energy (e.g., heat, electromagnetic energy, or the like). Such reflected electrons generated from elastic interaction may be referred to as backscattered electrons (BSEs). Some electrons of primary charged-particle beam 410 may inelastically interact with (e.g., in the form of inelastic scattering or collision) the materials of wafer 430. An inelastic interaction does not conserve the total kinetic energies of the bodies of the interaction, in which some or all of the kinetic energy of the interacting bodies convert to other forms of energy. For example, through the inelastic interaction, the kinetic energy of some electrons of primary charged-particle beam 410 may cause electron excitation and transition of atoms of the materials. Such inelastic interaction may also generate electrons exiting the surface of wafer 430, which may be referred to as secondary electrons (SEs). Yield or emission rates of BSEs and SEs depend on, e.g., the material under inspection and the landing energy of the electrons of primary charged-particle beam 410 landing on the surface of the material, among others. The energy of the electrons of primary charged-particle beam 410 may be imparted in part by its acceleration voltage (e.g., the acceleration voltage between the anode and cathode of charged-particle source 402 in Fig. 4). The quantity of BSEs and SEs may be more or fewer (or even the same) than the injected electrons of primary charged-particle beam 410.

[0056] Fig. 5 is a block diagram of an example server 500, consistent with some embodiments of the disclosure. As shown in Fig. 5, server 500 can include processor 502. When processor 502 executes instructions described herein, server 500 can become a specialized machine. Processor 502 can be any type of circuitry capable of manipulating or processing information. For example, processor 502 can include any combination of any number of a central processing unit (“CPU”), a graphics processing unit (“GPU”), a neural processing unit (“NPU”), a microcontroller unit (“MCU”), an optical processor, a programmable logic controller, a microcontroller, a microprocessor, a digital signal processor, an intellectual property (IP) core, a Programmable Logic Array (PLA), a Programmable Array Logic (PAL), a Generic Array Logic (GAL), a Complex Programmable Logic Device (CPLD), a Field-Programmable Gate Array (FPGA), a System On Chip (SoC), an Application- Specific Integrated Circuit (ASIC), or the like. In some embodiments, processor 502 can also be a set of processors grouped as a single logical component. For example, as shown in Fig. 5, processor 502 can include multiple processors, including processor 502a, processor 502b, and processor 502n.

[0057] Server 500 can also include memory 504 configured to store data (e.g., a set of instructions, computer codes, intermediate data, or the like). For example, as shown in Fig. 5, the stored data can include program instructions and data for processing. Processor 502 can access the program instructions and data for processing (e.g., via bus 510), and execute the program instructions to perform an operation or manipulation on the data for processing. Memory 504 can include a highspeed random -access storage device or a non-volatile storage device. In some embodiments, memory504 can include any combination of any number of a random-access memory (RAM), a read-only memory (ROM), an optical disc, a magnetic disk, a hard drive, a solid-state drive, a flash drive, a security digital (SD) card, a memory stick, a compact flash (CF) card, or the like. Memory 504 can also be a group of memories (not shown in Fig. 5) grouped as a single logical component.

[0058] Bus 510 can be a communication device that transfers data between components inside server 500, such as an internal bus (e.g., a CPU-memory bus), an external bus (e.g., a universal serial bus port, a peripheral component interconnect express port), or the like.

[0059] For ease of explanation without causing ambiguity, processor 502 and other data processing circuits are collectively referred to as a “data processing circuit” in this disclosure. The data processing circuit can be implemented entirely as hardware, or as a combination of software, hardware, or firmware. In addition, the data processing circuit can be a single independent module or can be combined entirely or partially into any other component of server 500.

[0060] Server 500 can further include network interface 506 to provide wired or wireless communication with a network (e.g., the Internet, an intranet, a local area network, a mobile communications network, or the like). In some embodiments, network interface 506 can include any combination of any number of a network interface controller (NIC), a radio frequency (RF) module, a transponder, a transceiver, a modem, a router, a gateway, a wired network adapter, a wireless network adapter, a Bluetooth adapter, an infrared adapter, a near-field communication (“NFC”) adapter, a cellular network chip, or the like.

[0061] In some embodiments, optionally, server 500 can further include peripheral interface 508 to provide a connection to one or more peripheral devices. As shown in Fig. 5, the peripheral device can include, but is not limited to, a cursor control device (e.g., a mouse, a touchpad, or a touchscreen), a keyboard, a display (e.g., a cathode-ray tube display, a liquid crystal display, or a light-emitting diode display), a video input device (e.g., a camera or an input interface coupled to a video archive), or the like.

[0062] Consistent with some embodiments of this disclosure, the computer-implemented method of using a diffusion model to enhance image quality for SCPM images may also include training a machine learning model using the obtained training data. In some embodiments, the machine learning model may be trained by a computer hardware system. In some embodiments, as described elsewhere in this disclosure, the training data may include previously obtained images or may include simulated images.

[0063] In some embodiments, machine learning may be employed in the generation and enhancement of inspection images, reference images, or other images associated with apparatus 300 or apparatus 400. For example, in some embodiments, a machine learning system may be operated in association with, e.g., controller 209, image processing system 350, image acquirer 360, storage 370, image processing system 490, image acquirer 492, or storage 494 of FIGs. 2-4. In someembodiments, machine learning may be employed in the method of using a diffusion model to enhance image quality for SCPM images, e.g., method 1000 of FIG. 10.

[0064] A generative model can be generally defined as a model that is probabilistic in nature. In other words, a “generative” model is not one that performs forward simulation or rule-based approaches and, as such, it may not be necessary to model the physics of the processes involved in generating an actual image or output (for which a simulated image or output is being generated). Instead, the generative model can be learned (in that its parameters can be learned) based on a suitable training set of data. Such generative models may have a number of advantages for the embodiments described herein. In addition, the generative model may be configured to have a deep learning architecture in that the generative model may include multiple layers, which may perform a number of algorithms or transformations. The number of layers included in the generative model may depend on the particular use case. For practical purposes, a suitable range of layers is from two layers to a few tens of layers.

[0065] Deep learning is a type of machine learning. Machine learning can be generally defined as a type of artificial intelligence (Al) that provides computers with the ability to learn without being explicitly programmed. Machine learning focuses on the development of computer programs that can teach themselves to grow and change when exposed to new data. Machine learning explores the study and construction of algorithms that can learn from and make predictions on data — such algorithms overcome following strictly static program instructions by making data driven predictions or decisions, through building a model from sample inputs.

[0066] The machine learning described herein may be further performed as described in “Introduction to Statistical Machine Learning,” by Sugiyama, Morgan Kaufmann, 2016, 534 pages; “Discriminative, Generative, and Imitative Learning,” by Jebara, MIT Thesis, 2002, 212 pages; and “Principles of Data Mining (Adaptive Computation and Machine Learning)” by Hand et al., MIT Press, 2001, 578 pages; which are incorporated by reference as if fully set forth herein. The embodiments described herein may be further configured as described in these references.

[0067] In some embodiments, a machine learning system may comprise a neural network. For example, a model may be a deep neural network with a set of weights that model the world according to the data that it has been fed to train it. Neural networks can be generally defined as a computational approach that is based on a relatively large collection of neural units loosely modeling the way a biological brain solves problems with relatively large clusters of biological neurons connected by axons. Each neural unit is connected with many others, and links can be enforcing or inhibitory in their effect on the activation state of connected neural units. These systems are self-learning and trained rather than explicitly programmed and excel in areas where the solution or feature detection is difficult to express in a traditional computer program.

[0068] Neural networks typically consist of multiple layers, and the signal path traverses from front to back. The goal of the neural network is to solve problems in the same way that the human brainwould, although several neural networks are much more abstract. Modem neural network projects typically work with a few thousand to a few million neural units and millions of connections. The neural network may have any suitable architecture or configuration known in the art.

[0069] In a further example, a model may comprise a convolutional neural network. For example, the embodiments described herein can take advantage of learning concepts such as a convolutional neural network to solve the normally intractable representation conversion problem (e.g., rendering). The model may have any convolutional neural network configuration or architecture known in the art.

[0070] A neural network, as used herein, may refer to a computing model for analyzing underlying relationships in a set of input data by way of mimicking human brains. Similar to a biological neural network, the neural network may include a set of connected units or nodes (referred to as “neurons”), structured as different layers, where each connection (also referred to as an “edge”) may obtain and send a signal between neurons of neighboring layers in a way similar to a synapse in a biological brain. The signal may be any type of data (e.g., a real number). Each neuron may obtain one or more signals as an input and output another signal by applying a non-linear function to the inputted signals. Neurons and edges may typically be weighted by corresponding weights to represent the knowledge the neural network has acquired. During a training process (similar to a learning process of a biological brain), the weights may be adjusted (e.g., by increasing or decreasing their values) to change the strengths of the signals between the neurons to improve the performance accuracy of the neural network. Neurons may apply a thresholding function (referred to as an “activation function”) to its output values of the non-linear function such that a signal is outputted only when an aggregated value (e.g., a weighted sum) of the output values of the non-linear function exceeds a threshold determined by the thresholding function. Different layers of neurons may transform their input signals in different manners (e.g., by applying different non-linear functions or activation functions). The output of the last layer (referred to as an “output layer”) may output the analysis result of the neural network, such as, for example, a categorization of the set of input data (e.g., as in image recognition cases), a numerical result, or any type of output data for obtaining an analytical result from the input data.

[0071] Training of the neural network, as used herein, may refer to a process of improving the accuracy of the output of the neural network. Typically, the training may be categorized into three types: supervised training, unsupervised training, and reinforcement training. In the supervised training, a set of target output data (also referred to as “labels” or “ground truth”) may be generated based on a set of input data using a method other than the neural network. The neural network may then be fed with the set of input data to generate a set of output data that is typically different from the target output data. Based on the difference between the output data and the target output data, the weights of the neural network may be adjusted in accordance with a rule. If such adjustments are successful, the neural network may generate another set of output data more similar to the target output data in a next iteration using the same input data. If such adjustments are not successful, theweights of the neural network may be adjusted again. After a sufficient number of iterations, the training process may be terminated in accordance with one or more predetermined criteria (e.g., the difference between the final output data and the target output data is below a predetermined threshold, or the number of iterations reaches a predetermined threshold). The trained neural network may be applied to analyze other input data.

[0072] In the unsupervised training, the neural network is trained without any external gauge (e.g., labels) to identify patterns in the input data rather than generating labels for them. Typically, the neural network may analyze shared attributes (e.g., similarities and differences) and relationships among the elements of the input data in accordance with one or more predetermined rules or algorithms (e.g., principal component analysis, clustering, anomaly detection, or latent variable identification). The trained neural network may extrapolate the identified relationships to other input data.

[0073] In the reinforcement learning, the neural network is trained without any external gauge (e.g., labels) in a trial -and-error manner to maximize benefits in decision making. The input data sets of the neural network may be different in the reinforcement training. For example, a reward value or a penalty value may be determined for the output of the neural network in accordance with one or more rules during training, and the weights of the neural network may be adjusted to maximize the reward values (or to minimize the penalty values). The trained neural network may apply its learned decisionmaking knowledge to other input data.

[0074] During the training of a neural network, a loss function (or referred to as a “cost function”) may be used to evaluate the output data. The loss function, as used herein, may map output data of a machine learning model (e.g., the neural network) onto a real number (referred to as a “loss” or a “cost”) that intuitively represents a loss or an error (e.g., representing a difference between the output data and target output data) associated with the output data. The training of the neural network may seek to maximize or minimize the loss function (e.g., by pushing the loss towards a local maximum or a local minimum in a loss curve). For example, one or more parameters of the neural network may be adjusted or updated purporting to maximize or minimize the loss function. After adjusting or updating the one or more parameters, the neural network may obtain new input data in a next iteration of its training. When the loss function is maximized or minimized, the training of the neural network may be terminated.

[0075] By way of example, Fig. 6 is a schematic diagram illustrating an example neural network 600, consistent with some embodiments of the present disclosure. As depicted in Fig. 6, neural network 600 may include an input layer 620 that receives inputs, including input 610-1, . . ., input 610-m (m being an integer). For example, an input of neural network 600 may include any structure or unstructured data (e.g., an image). In some embodiments, neural network 600 may obtain a plurality of inputs simultaneously. For example, in Fig. 6, neural network 600 may obtain m inputs simultaneously. In some embodiments, input layer 620 may obtain m inputs in succession such thatinput layer 620 receives input 610-1 in a first cycle (e.g., in a first inference) and pushes data from input 610-1 to a hidden layer (e.g., hidden layer 630-1), then receives a second input in a second cycle (e.g., in a second inference) and pushes data from input the second input to the hidden layer, and so on. Input layer 620 may obtain any number of inputs in the simultaneous manner, the successive manner, or any manner of grouping the inputs.

[0076] Input layer 620 may include one or more nodes, including node 620-1, node 620-2, . . ., node 620-a (a being an integer). A node (also referred to as a “machine perceptron” or a “neuron”) may model the functioning of a biological neuron. Each node may apply an activation function to received inputs (e.g., one or more of input 610-1, . . ., input 610-m). An activation function may include a Heaviside step function, a Gaussian function, a multiquadratic function, an inverse multiquadratic function, a sigmoidal function, a rectified linear unit (ReLU) function (e.g., a ReLU6 function or a Leaky ReLU function), a hyperbolic tangent (“tanh”) function, or any non-linear function. The output of the activation function may be weighted by a weight associated with the node. A weight may include a positive value between 0 and 1, or any numerical value that may scale outputs of some nodes in a layer more or less than outputs of other nodes in the same layer.

[0077] As further depicted in Fig. 6, neural network 600 includes multiple hidden layers, including hidden layer 630-1, . . ., hidden layer 630-n (n being an integer). When neural network 600 includes more than one hidden layer, it may be referred to as a “deep neural network” (DNN). Each hidden layer may include one or more nodes. For example, in Fig. 6, hidden layer 630-1 includes node 630-1 - 1, node 630-1-2, node 630-1-3, . . ., node 630-1-b (b being an integer), and hidden layer 630-n includes node 630-n-l, node 630-n-2, node 630-n-3, . . ., node 630-n-c (c being an integer). Similar to nodes of input layer 620, nodes of the hidden layers may apply the same or different activation functions to outputs from connected nodes of a previous layer, and weight the outputs from the activation functions by weights associated with the nodes.

[0078] As further depicted in Fig. 6, neural network 600 may include an output layer 640 that finalizes outputs, including output 650-1, output 650-2, . . ., output 650-d (d being an integer). Output layer 640 may include one or more nodes, including node 640-1, node 640-2, . . ., node 640-d. Similar to nodes of input layer 620 and of the hidden layers, nodes of output layer 640 may apply activation functions to outputs from connected nodes of a previous layer and weight the outputs from the activation functions by weights associated with the nodes.

[0079] Although nodes of each hidden layer of neural network 600 are depicted in Fig. 6 to be connected to each node of its previous layer and next layer (referred to as “fully connected”), the layers of neural network 600 may use any connection scheme. For example, one or more layers (e.g., input layer 620, hidden layer 630-1, . . ., hidden layer 630-n, or output layer 640) of neural network 600 may be connected using a convolutional scheme, a sparsely connected scheme, or any connection scheme that uses fewer connections between one layer and a previous layer than the fully connected scheme as depicted in Fig. 6.

[0080] Moreover, although the inputs and outputs of the layers of neural network 600 are depicted as propagating in a forward direction (e.g., being fed from input layer 620 to output layer 640, referred to as a “feedforward network”) in Fig. 6, neural network 600 may additionally or alternatively use backpropagation (e.g., feeding data from output layer 640 towards input layer 620) for other purposes. In some embodiments, neural network can be a diffusion model.

[0081] Consistent with embodiments of the present disclosure, a diffusion model may be trained with higher quality images. For the diffusion model training, lower quality images are not needed and therefore, alignment between the lower quality images and the higher quality images is not required, thereby reducing processing times for the large number of images that are produced during the wafer inspection process.

[0082] To train the diffusion model, a different level of noise is added to a higher quality image at each training sample. For each training iteration, a batch of data including multiple training samples is collected. Each training sample in the batch is a high quality image combined with a random level of noise. The output of the neural network for the sample is the random level of noise added to the sample. The diffusion model is built such that during inference, the diffusion model can start with a noisy input image and can determine the amount of noise that was introduced in the noisy input image. The diffusion model can then remove the noise from the noisy input image to obtain a higher quality image as the end result. By using a diffusion model, if there is a defect on a wafer (as shown in the higher quality image), the defect will not get “filtered out” during model training, as may be done in prior ML-related methods.

[0083] Fig. 7 is an example of a process 700 for training a diffusion model using a SCPM image, consistent with embodiments of the present disclosure. In some embodiments, the process 700 may be performed by a server, such as the server 500 shown in Fig. 5 and using a neural network, such as the neural network 600 shown in Fig. 6. Training a diffusion model includes a forward (noising) process and a backward (denoising) process. The forward process starts with a higher quality start image 702. Noise is generated and is added to the start image 702 to obtain a first noised image 704. The forward process is repeated for several steps; in implementation, more steps than that shown in Fig. 7 may be used. For example, as shown in Fig. 7, noise is generated and is added to the first noised image 704 to obtain a second noised image 706. Then noise is generated and is added to the second noised image 706 to obtain a third noised image 708. Each successive image in the forward process is of a lower quality (e.g., has more noise) than the preceding image.

[0084] The backward (denoising) process may be performed by the diffusion model. The backward process starts with a noisy image (e.g., third noised image 708) and predicts the noise to be removed from the image to obtain the start image 702. At each step of the backward process, a part of the predicted noise is subtracted from the input image (e.g., third noised image 708) to obtain the prior image in the process (e.g., from third noised image 708, the model subtracts noise to first attempt to obtain second noised image 706). At each step of the backward process, the part of the predictednoised that is subtracted from the input image gets larger, such that it is possible to determine the start image 702 from the input image (e.g., third noised image 708).

[0085] Fig. 8 is an example of a denoising process 800 using a trained diffusion model, consistent with some embodiments of the present disclosure. The denoising process 800 begins with a noisy image 802 (step 0). At each step of the denoising process 800, a part of the noise in the noisy image 802 is iteratively removed, resulting in a series of subsequently clearer (e.g., less noisy) images 804- 824 (steps 1-11). While 11 denoising steps are shown in Fig. 8, it is to be understood that the denoising process 800 may take more or fewer steps than the 11 steps shown. In some embodiments, the number of denoising steps used in the denoising process 800 may be based on minimizing a value of a loss function to a predetermined level or by performing a predetermined number of steps.

[0086] As used in the present disclosure, the term “raw image” refers to a SCPM image directly from the scanner. In some embodiments, to obtain a higher quality image, a number of raw images may be averaged together. For example, a number of raw images taken from a fixed location may be stacked on top of each other and an average of each pixel of each image in the stack may be obtained. It is noted that other image averaging techniques may be used and are contemplated to be within the scope of the present disclosure. In some embodiments, the raw images may be cleaned before averaging. In some embodiments, the number of raw images to use for averaging may depend on the quality of the raw images or how many gauges can be extracted from the averaged images. It is assumed that all dies are created from the same mask, so any noise appearing in the images should result from manufacturing variations (e.g., the noise is “white noise,” such as random noise or Gaussian noise), and not a problem with the mask. Averaging a number of raw images together may lead to a clearer raw image. Different embodiments may use a different number of images to average (for example, five, ten, or 20 images).

[0087] Similarly, denoised SCPM images may be averaged to obtain a higher quality image. The process used to average the denoised SCPM images may be the same as the process used for averaging raw images.

[0088] From the averaged image (whether a raw image or a denoised image), features shown in the image may be measured. For example, one edge of a feature to an opposite edge of the feature may be measured (also referred to as a “gauge”). As an example, if the feature to be measured is a rectangle, the gauge may be the top edge of the rectangle to the bottom edge of the rectangle, or the left edge of the rectangle to the right edge of the rectangle. Using clear images for the measurements is important because the image does not get sent to the next step in the process; the measurements obtained from the image are what is sent to be used later in the processing, so it is preferable to be able to have clear images to determine accurate measurements.

[0089] Fig. 9 is an example of measurements that may be taken from an averaged raw image 900 and an averaged denoised SCPM image 902, consistent with some embodiments of the present disclosure. The averaged raw image 900 is based on averaging 20 raw images together. The averaged denoisedimage 902 is based on averaging 20 denoised images together. It is noted that the number of images used to obtain the averaged images 900, 902 is exemplary and that a different number of images (e.g., five images, ten images, or another number of images) may be used. In some embodiments, a number of gauges present in the averaged image may be used to determine how many images to average. For example, if five images are averaged, there may be a sufficient number of gauges, but the statistics based on the averaged image may not be sufficient, e.g., a large standard deviation for some gauges. As another example, if ten images are averaged, there may be a sufficient number of gauges and the statistics based on the averaged image may be good, e.g., the standard deviation may be below a threshold. To compare the quality of the averaged ten images, an averaged image based on 20 images may be determined and the statistics of the “average 20” image may be compared with the statistics of the “average 10” image. If the gauge quality between the “average 10” image is similar to the gauge quality of the “average 20” image (e.g., there is an insignificant difference between the standard deviation of the “average 10” image and the “average 20” image), then the “average 10” image may be used.

[0090] As shown in Fig. 9, the averaged raw image 900 includes several complete features indicated as rectangles 910a-910e. The averaged denoised image 902 includes complete features indicated as rectangles 920a and 920b. It is noted that the features that may be measured in the images may be any type of feature that may be manufactured on an IC. The features shown in Fig. 9 are used for ease of illustration and explanation.

[0091] Depending on the pattern being examined (e.g., rectangle features as shown in Fig. 9), a number of different measurements and measurement statistics may be obtained from the averaged image. For example, a measurement count may be taken, which is a number of times that the same feature can be measured in a single image. For example, in image 900, there are five rectangles (e.g., features 910a-910e) that may be measured. As another example, in image 902, there are two rectangles (e.g., features 920a and 920b) that may be measured.

[0092] As another example, the gauge of each feature may be measured. For example, with feature 910a, the gauge may be measured from a left edge of feature 910a to a right edge of feature 910a. The gauge of feature 910a may also be measured from a top edge of feature 910a to a bottom edge of feature 910a. From the gauge measurements, the critical dimension (CD) and edge placement error (EPE) of the associated pattern may be determined. Because there are multiple pattems / features shown in a single image, additional statistics may be collected, such as maximum CD, minimum CD, the standard deviation of the measured CDs, and the standard deviation of the EPEs.

[0093] When taking measurements, some measurements may need to be discarded. For example, with the rectangle features shown in Fig. 9, if only a left edge of a rectangle can be detected and measured but the corresponding right edge of the same rectangle cannot be detected (e.g., as shown by feature 922 in image 902), then the left edge measurement for feature 922 needs to be discarded. As another example with a rectangle feature, if the left edge of a first rectangle can be detected (e.g., theleft edge of rectangle 910c) but the corresponding right edge of the rectangle (e.g., the right edge of rectangle 910c) cannot be located but a left edge of a second rectangle (e.g., the left edge of rectangle 912) is detected, the left edge of the second rectangle 912 may be incorrectly interpreted as the right edge of the first rectangle 914c. In this example, the gauge of the rectangle 910c will be much larger than the gauge of the other measured rectangles in the image 900 and any measurements based on this inaccurate determination should be discarded.

[0094] In some embodiments, because the averaged denoised image is generally a clearer image than the averaged raw image, the measurement count may be higher for the averaged denoised image. Similarly, because the measurement count increases, the standard deviation among the measurements may be lower, leading to more accurate measurements and a higher confidence in the CD value or the EPE value obtained from the measurements. The collected statistics may also help determine the quality of the image clean-up performed by the trained diffusion model.

[0095] Because taking a large number of SCPM images is a time-consuming process (e.g., as would be used to generate an averaged raw image), it may be beneficial to have a higher quality image (e.g., a denoised image or an averaged denoised image) such that more measurements may be taken from a single image and the measurements may have a higher accuracy.

[0096] In some embodiments, it may be desirable to reduce the number of images to be averaged to obtain the averaged image yet still result in a high-quality averaged image. As noted, taking SCPM images is a time-consuming process and if fewer images can be taken to obtain a high-quality averaged image, then a wafer may be processed faster. In some embodiments, the determination of the number of images to be averaged may be based on the standard deviation of the CD or the EPE obtained from the image. For example, a threshold may be established for the standard deviation and if the standard deviation is larger than the threshold, then a higher number of images may need to be used to generate the averaged image to attempt to reduce the standard deviation. In some embodiments, if the standard deviation is larger than the threshold, the gauges may be discarded, thus fewer gauges may be collected in the averaged image. For example, the number of images may increase in steps, such as one image, five images, ten images, or 20 images. It is noted that other numbers of images may be used to generate the averaged image. In some embodiments, the determination of the number of images used to generate the averaged image may be automated. For example, if the threshold for the standard deviation is used (as described above), then if the standard deviation is greater than the threshold when using X images to determine the averaged image, then the next-highest number of images may be used to determine the averaged image and the standard deviation is compared to the threshold. The process may be repeated until the standard deviation is less than the threshold. For example, the process may start with five images used to determine the averaged image, then proceed to ten images, 20 images, and subsequently higher numbers of images, e.g., in increments of five or ten images for each subsequent iteration.

[0097] Fig. 10 is a flowchart of an example method 1000 for using a diffusion model with SCPM images, consistent with embodiments of the present disclosure. In some embodiments, the method 1000 may be performed by image processing system 350 of Fig. 3, by image processing system 490 of Fig. 4, or by the server 500 of Fig. 5, any of which may also use the neural network 600 of Fig. 6.

[0098] At step 1002, a trained diffusion neural network model may be applied to denoise SCPM images. For example, the SCPM images may be received from SCPM system 200 and the denoising process 800 as shown in Fig. 8 may be applied to denoise the SCPM images.

[0099] At step 1004, an averaged image may be determined as an average of the denoised SCPM images. For example, a number of images taken from a fixed location may be stacked on top of each other and an average of each pixel of each image in the stack may be obtained.

[0100] At step 1006, measurements may be determined from the averaged image, such as a measurement count, a gauge, an EPE, a maximum CD, a minimum CD, a standard deviation of the CD, and a standard deviation of the EPE. For example, the measurements may be determined from the averaged image as described in connection with Fig. 9.

[0101] A non-transitory computer readable medium may be provided that stores instructions for a processor of a controller (e.g., controller 209 of Fig. 2, controller 209 of Fig. 3, controller 496 of Fig. 4, or processor 502 of Fig. 5) to carry out, among other things, image inspection, image acquisition, stage positioning, beam focusing, electric field adjustment, beam bending, condenser lens adjusting, activating charged particle source, beam deflecting, and operations 700 and 800 and method 1000. Common forms of non-transitory media include, for example, a floppy disk, a flexible disk, hard disk, solid state drive, magnetic tape, or any other magnetic data storage medium, a Compact Disc Read Only Memory (CD-ROM), any other optical data storage medium, any physical medium with patterns of holes, a Random Access Memory (RAM), a Programmable Read Only Memory (PROM), and Erasable Programmable Read Only Memory (EPROM), a FLASH-EPROM or any other flash memory, Non-Volatile Random Access Memory (NVRAM), a cache, a register, any other memory chip or cartridge, and networked versions of the same.

[0102] The embodiments may further be described using the following clauses:1.A non-transitory computer readable medium that stores a set of instructions that is executable by at least one processor of a computing device to cause the computing device to perform operations for determining measurements of a feature on a wafer from a denoised image, the operations comprising: applying a trained diffusion neural network model to denoise scanning charged-particle microscope (SCPM) images, wherein the trained diffusion neural network model has been trained to denoise a received SCPM image of the wafer; determining an averaged image as an average of the denoised SCPM images; and determining measurements of the feature based on the averaged image.2. The non-transitory computer readable medium of clause 1, wherein determining measurements includes measuring a gauge of the feature in the denoised SCPM images.3. The non -transitory computer readable medium of clauses 1 or 2, wherein determining measurements includes measuring a critical dimension of the feature in the denoised SCPM images.4. The non-transitory computer readable medium of any one of clauses 1-3, wherein determining measurements includes measuring an edge placement error of the feature in the denoised SCPM images.5. The non-transitory computer readable medium of any one of clauses 1-4, wherein determining measurements includes determining a measurement count, wherein the measurement count is a number of measurements that can be taken of the feature in the denoised SCPM images.6. The non-transitory computer readable medium of clause 1, wherein determining measurements includes determining measurement statistics for the feature in the denoised SCPM images.7. The non-transitory computer readable medium of clause 6, wherein the measurement statistics include one or more of: a critical dimension of the feature; an edge placement error of the feature; a measurement count, wherein the measurement count is a number of measurements that can be taken of the feature; and a standard deviation of the critical dimension or the edge placement error.8. The non-transitory computer readable medium of clause 7, wherein determining the averaged image includes: comparing the standard deviation to a threshold; on a condition that the standard deviation is greater than the threshold, increasing a number of SCPM images used to determine the averaged image; and repeating the comparing and increasing steps until the standard deviation is less than the threshold.9. The non-transitory computer readable medium of any one of clauses 1-8, wherein a number of SCPM images used to determine the averaged image is at least five images, ten images, or twenty images.10. An apparatus for determining measurements of a feature on a wafer from a denoised image, comprising: a memory storing a set of instructions; and at least one processor configured to execute the set of instructions to cause the apparatus to perform operations comprising: applying a trained diffusion neural network model to denoise scanning charged-particle microscope (SCPM) images, wherein the trained diffusion neural network model has been trained to denoise a received SCPM image of the wafer; determining an averaged image as an average of the denoised SCPM images; and determining measurements of the feature based on the averaged image, including determining measurement statistics for the feature in the denoised SCPM images.11. The apparatus of clause 10, wherein the measurement statistics include one or more of: a critical dimension of the feature; an edge placement error of the feature; and a standard deviation of the critical dimension or the edge placement error.12. The apparatus of clause 11, wherein determining the averaged image includes: comparing the standard deviation to a threshold; on a condition that the standard deviation is greater than the threshold, increasing a number of SCPM images used to determine the averaged image; and repeating the comparing and increasing steps until the standard deviation is less than the threshold.13. The apparatus of any one of clauses 10-12, wherein determining measurements further includes measuring a gauge of the feature in the denoised SCPM images.14. The apparatus of any one of clauses 10-13, wherein determining measurements further includes determining a measurement count, wherein the measurement count is a number of measurements that can be taken of the feature in the denoised SCPM images.15. The apparatus of any one of clauses 10-14, wherein a number of SCPM images used to determine the averaged image is at least five images, ten images, or twenty images.16. A method for determining measurements of a feature on a wafer from a denoised image, comprising: applying a trained diffusion neural network model to denoise scanning charged-particle microscope (SCPM) images, wherein the trained diffusion neural network model has been trained to denoise a received SCPM image of the wafer; determining an averaged image as an average of the denoised SCPM images; and determining measurements of the feature based on the averaged image, including determining measurement statistics for the feature in the denoised SCPM images.17. The method of clause 16, wherein the measurement statistics include one or more of: a critical dimension of the feature; an edge placement error of the feature; and a standard deviation of the critical dimension or the edge placement error.18. The method of claim 17, wherein determining the averaged image includes: comparing the standard deviation to a threshold; on a condition that the standard deviation is greater than the threshold, increasing a number of SCPM images used to determine the averaged image; and repeating the comparing and increasing steps until the standard deviation is less than the threshold.19. The method of any one of clauses 16-18, wherein determining measurements further includes measuring a gauge of the feature in the denoised SCPM images.20. The method of any one of clauses 16-19, wherein determining measurements further includes determining a measurement count, wherein the measurement count is a number of measurements that can be taken of the feature in the denoised SCPM images.21. The method of any one of clauses 16-20, wherein a number of SCPM images used to determine the averaged image is at least five images, ten images, or twenty images.

[0103] Block diagrams in the figures may illustrate the architecture, functionality, and operation of possible implementations of systems, methods, and computer hardware or software products according to various exemplary embodiments of the present disclosure. In some embodiments, a non- transitory computer-readable medium is provided and can include instructions to perform the functions described in connection with any one or more of Figs. 7-10. In this regard, each block in a schematic diagram may represent certain arithmetical or logical operation processing that may be implemented using hardware such as an electronic circuit. Blocks may also represent a module, segment, or portion of code that comprises one or more executable instructions for implementing the specified logical functions. It should be understood that in some alternative implementations, functions indicated in a block may occur out of the order noted in the figures. For example, two blocks shown in succession may be executed or implemented substantially concurrently, or two blocks may sometimes be executed in reverse order, depending upon the functionality involved. Some blocks may also be omitted. It should also be understood that each block of the block diagrams, and combination of the blocks, may be implemented by special purpose hardware -based systems that perform the specified functions or acts, or by combinations of special purpose hardware and computer instructions.

[0104] It will be appreciated that the embodiments of the present disclosure are not limited to the exact construction that has been described above and illustrated in the accompanying drawings, and that various modifications and changes may be made without departing from the scope thereof. The present disclosure has been described in connection with various embodiments, and other embodiments will be apparent to those skilled in the art from consideration of the specification and practice of the technology disclosed herein. It is intended that the specification and examples be considered as exemplary only, with a true scope of the invention being indicated by the following claims.

Claims

CLAIMS1. A non-transitory computer readable medium that stores a set of instructions that is executable by at least one processor of a computing device to cause the computing device to perform operations for determining measurements of a feature on a wafer from a denoised image, the operations comprising: applying a trained diffusion neural network model to denoise scanning charged-particle microscope (SCPM) images, wherein the trained diffusion neural network model has been trained to denoise a received SCPM image of the wafer; determining an averaged image as an average of the denoised SCPM images; and determining measurements of the feature based on the averaged image.

2. The non-transitory computer readable medium of claim 1, wherein determining measurements includes measuring a gauge of the feature in the denoised SCPM images.

3. The non-transitory computer readable medium of claim 1, wherein determining measurements includes measuring a critical dimension of the feature in the denoised SCPM images.

4. The non-transitory computer readable medium of claim 1, wherein determining measurements includes measuring an edge placement error of the feature in the denoised SCPM images.

5. The non-transitory computer readable medium of claim 1, wherein determining measurements includes determining a measurement count, wherein the measurement count is a number of measurements that can be taken of the feature in the denoised SCPM images.

6. The non-transitory computer readable medium of claim 1, wherein determining measurements includes determining measurement statistics for the feature in the denoised SCPM images.

7. The non-transitory computer readable medium of claim 6, wherein the measurement statistics include one or more of: a critical dimension of the feature; an edge placement error of the feature; a measurement count, wherein the measurement count is a number of measurements that can be taken of the feature; and a standard deviation of the critical dimension or the edge placement error.

8. The non-transitory computer readable medium of claim 7, wherein determining the averaged image includes: comparing the standard deviation to a threshold; on a condition that the standard deviation is greater than the threshold, increasing a number of SCPM images used to determine the averaged image; and repeating the comparing and increasing steps until the standard deviation is less than the threshold.

9. The non-transitory computer readable medium of claim 1, wherein a number of SCPM images used to determine the averaged image is at least five images, ten images, or twenty images.

10. An apparatus for determining measurements of a feature on a wafer from a denoised image, comprising: a memory storing a set of instructions; and at least one processor configured to execute the set of instructions to cause the apparatus to perform operations comprising: applying a trained diffusion neural network model to denoise scanning charged-particle microscope (SCPM) images, wherein the trained diffusion neural network model has been trained to denoise a received SCPM image of the wafer; determining an averaged image as an average of the denoised SCPM images; and determining measurements of the feature based on the averaged image, including determining measurement statistics for the feature in the denoised SCPM images.

11. The apparatus of claim 10, wherein the measurement statistics include one or more of: a critical dimension of the feature; an edge placement error of the feature; and a standard deviation of the critical dimension or the edge placement error.

12. The apparatus of claim 11, wherein determining the averaged image includes: comparing the standard deviation to a threshold; on a condition that the standard deviation is greater than the threshold, increasing a number of SCPM images used to determine the averaged image; and repeating the comparing and increasing steps until the standard deviation is less than the threshold.

13. The apparatus of claim 10, wherein determining measurements further includes measuring a gauge of the feature in the denoised SCPM images.

14. The apparatus of claim 10, wherein determining measurements further includes determining a measurement count, wherein the measurement count is a number of measurements that can be taken of the feature in the denoised SCPM images.

15. The apparatus of claim 10, wherein a number of SCPM images used to determine the averaged image is at least five images, ten images, or twenty images.

Images