Use of synthetic photo-realistic images for semiconductor fabrication apparatuses

Abstract

In accordance with some embodiments, methods, systems, and media for using synthetic images are provided. In some embodiments, a method involves obtaining a set of images associated with the integrated circuit fabrication chamber, wherein the set of images comprises at least a subset of synthetic images. Each synthetic image of the subset of synthetic images is a photo-realistic synthetic image. In some embodiments, at least a portion of the synthetic images represent an anomalous condition associated with the fabrication chamber, and each image in the set of images represents at least one of: 1) components of the fabrication chamber; 2) a process occurring in the fabrication chamber; or 3) wafer characteristics of a wafer being processed in the fabrication chamber. The method may further involve training a machine learning model using the set of images, wherein the trained machine learning model is usable to predict the anomalous condition.

Description

USE OF SYNTHETIC PHOTO-REALISTIC IMAGES FOR SEMICONDUCTOR FABRICATION APPARATUSESINCORPORATION BY REFERENCE

[0001] A PCT Request Form is filed concurrently with this specification as part of the present application. Each application that the present application claims benefit of or priority to as identified in the concurrently filed PCT Request Form is incorporated by reference herein in its entirety and for all purposes.BACKGROUND

[0002] Anomalous conditions in semiconductor fabrication equipment can lead to tool downtime, unusable processed wafers, and other events that are undesirable. However, it can be difficult to quickly identify anomalous conditions, whether for a wafer undergoing processing or related to one or more components of the fabrication equipment.

[0003] The background description provided herein is for the purposes of generally presenting the context of the disclosure. Work of the presently named inventors, to the extent it is described in this background section, as well as aspects of the description that may not otherwise qualify as prior art at the time of filing, are neither expressly nor implicitly admitted as prior art against the present disclosure.SUMMARY

[0004] Techniques for use of synthetic photo-realistic images for semiconductor fabrication apparatuses are provided. In some embodiments, a method of using synthetic images in connection with an integrated circuit fabrication chamber is provided. The method may comprise obtaining a set of images associated with the integrated circuit fabrication chamber, wherein the set of images comprises at least a subset of synthetic images. In some embodiments, each synthetic image of the subset of synthetic images is a photo-realistic synthetic image of an image captured by a camera disposed in or on a portion of the fabrication chamber; at least a portion of the subset of synthetic images represent at least one anomalous condition associated with the fabrication chamber; and each image in the set of images represents at least one of 1) components of the fabrication chamber; 2) a process occurring in the fabrication chamber; or 3) wafer characteristics of a wafer being processed in the fabrication chamber. In some embodiments, the method may comprise training a machine learning model using the set of images, wherein the trained machine learning model is usable to predict the at least one anomalous condition represented in the at least the portion of the subset of synthetic images during operation of the fabrication chamber.

[0005] In some examples, the camera is configured to operate in the ultraviolet, visible, and / or infrared range.

[0006] In some examples, the wafer characteristics comprise film growth during a deposition process.

[0007] In some examples, the trained machine learning model takes chamber features as an input and generates an output usable for determining a time point at which to initiate a cleaning cycle. In some examples, the chamber features comprise features associated with at least one of a focus ring, a showerhead face, or any combination thereof.

[0008] In some examples, the components of the fabrication chamber comprise at least one of a showerhead; a pedestal; one or more lift pins; an indexer; or a carousel. In some examples, the at least one anomalous condition comprises a tilt of the pedestal. In some examples, the at least one anomalous condition comprises uneven heights of the one or more lift pins.

[0009] In some examples, the at least one anomalous condition comprises a hollow cathode discharge. In some examples, an output of the trained machine learning model is usable to determine depth information associated with the hollow cathode discharge.

[0010] In some examples, the set of images represents the process occurring in the fabrication chamber, and wherein the process is a plasma-enhanced process. In some examples, the at least one anomalous condition comprises parasitic plasma within the fabrication chamber. In some examples, the at least one anomalous condition comprises plasma non-uniformity that exceeds a uniformity threshold.

[0011] In some examples, the at least one anomalous condition comprises an indication of liquid droplets on the wafer being processed.

[0012] In some examples, the at least one anomalous condition comprises an indication of a liquid phenomena occurring within a portion of the fabrication chamber, wherein the liquid phenomena includes at least one of bubbling, frothing, or change of liquid color.

[0013] In some examples, the wafer characteristics of the wafer comprise a degree to which the wafer is centered on a pedestal.

[0014] In some examples, the wafer characteristics of the wafer comprise a wafer rotational angle and / or a wafer angular velocity. In some examples, the subset of synthetic images comprise at least one of synthetic images that include different cross-sectional views of a wafer, syntheticimages that include different robot end effector shapes, or synthetic images that include different robot end effector arm shapes.

[0015] In some examples, each synthetic image is generated using computer-aided design information depicting the fabrication chamber and based on light propagation information.

[0016] In some examples, each synthetic image is generated using data from a spectral sensor.

[0017] In some examples, a rate at which the portion of the subset of the synthetic images representing the at least one anomalous condition is higher than a rate at which the at least one anomalous condition naturally occurs in the fabrication chamber.

[0018] In some examples, the trained machine learning model is configured to predict the at least one anomalous condition of the fabrication chamber when at least one component of the fabrication chamber is operating at a temperature above at least about 400 degrees Celsius.

[0019] In some examples, the trained machine learning model is configured to predict the at least one anomalous condition of the fabrication chamber when at least one component of the fabrication chamber is operating at a temperature above at least about 100 degrees Celsius.

[0020] In some examples, the trained machine learning model is configured to segment real images based on the set of images used to train the machine learning model, and wherein the segmented real images comprise an indication of at least one of: a wafer, a pedestal, a lift pin, a robotic arm, or a showerhead.

[0021] According to some embodiments, a method of using synthetic images in connection with an integrated circuit fabrication chamber is provided. The method may comprise: receiving image data from one or more cameras disposed in or on a portion of the fabrication chamber; providing the image data to a trained machine learning model to generate an output, wherein the trained machine learning model was trained using a set of image comprising at least a subset of synthetic images, and wherein each synthetic image of the subset of synthetic images is a photorealistic synthetic image of an image captured by the one or more cameras; and utilizing the output to predict at least one anomalous condition associated with the fabrication chamber.

[0022] In some examples, the method may comprise taking at least one action based on the prediction of the at least one anomalous condition. In some examples, the at least one action comprises causing an indication of the at least one anomalous condition to be presented on a console associated with the fabrication chamber. In some examples, the at least one actioncomprises causing at least one control action to be implemented to compensate for the at least one anomalous condition. In some examples, the at least one control action comprises causing at least one modification to at least one process parameter to be implemented. In some examples, the at least one process parameter comprises at least one of: an RF power setting, or a gas flow setting, a number of processing cycles, or a processing time. In some examples, the at least one modification to at least one process parameter is implemented by a second fabrication chamber that is configured to perform a downstream process on a wafer associated with the at least one anomalous condition.

[0023] According to some embodiments, a method comprises generating a training set, the training set comprising at least a subset of synthetic images associated with a fabrication chamber, each synthetic image being a photo-realistic image of an image captured by at least one camera disposed in or on a portion of the fabrication chamber; and utilizing the training set to train a machine learning model, an output of the machine learning model being usable to perform at least one of: 1) generation of one or more enhanced images depicting at least a portion of the fabrication chamber; or 2) determination of one or more modified camera settings and / or lighting settings for use in future operation of the at least one camera disposed in or on the portion of the fabrication chamber.

[0024] In some examples, the subset of synthetic images comprises images at different focuses. In some examples, the images at different focuses are utilized to generate the one or more enhanced images, the one or more enhanced images comprising a deblurred version of a photorealistic image.

[0025] In some examples, the training set comprises data from one or more sensors other than camera sensors. In some examples, the data from the one or more sensors other than the camera sensors are used to generate the subset of the synthetic images.

[0026] In some examples, the one or more enhanced images are compiled to generate a video depicting a state of at least a portion of the fabrication chamber.

[0027] In some examples, the one or more enhanced images comprise a three-dimensional reconstruction of the fabrication chamber.

[0028] In some examples, the one or more modified camera settings and / or lighting settings comprise one or more camera resolutions.

[0029] In some examples, the one or more modified camera settings and / or lighting settings comprise orientations of one or more lighting devices within the fabrication chamber.

[0030] In some examples, the one or more modified camera settings and / or lighting settings comprise orientations of one or more baffles configured to block light into a lens of one or more cameras.

[0031] According to some embodiments, a calibration disc for calibrating wafer rotation is provided. The calibration disc may comprise: a top surface; a bottom surface; and a side edge that joins the top surface and the bottom surface, wherein the side edge comprises a plurality of pattern markings, and wherein a snapshot of a subset of the pattern markings is usable to determine an angular rotation or angular velocity of the calibration disc.

[0032] In some examples, the plurality of pattern markings comprise at least one of dots, rectangles, squares, or numbers.

[0033] In some examples, pattern markings of the plurality of pattern markings are of different sizes and / or of different separation distances from one another. In some examples, the sizes and / or the separation distances of the subset of the pattern markings specifies the angular rotation or the angular velocity of the calibration disc.BRIEF DESCRIPTION OF THE DRAWINGS

[0034] Figure 1A is a schematic diagram of an example apparatus in accordance with some embodiments.

[0035] Figure IB is a schematic diagram of an example multi-station tool in accordance with some embodiments.

[0036] Figure 2A is a flowchart of an example process for training a model using synthetic images in accordance with some embodiments.

[0037] Figure 2B is a flowchart of an example process for utilizing a trained model trained using synthetic images in accordance with some embodiments.

[0038] Figure 3A is a flowchart of an example process for training a model using synthetic images, where the model is usable to predict an anomalous condition in accordance with some embodiments.

[0039] Figure 3B is a flowchart of an example process for utilizing a trained model to predict an anomalous condition in accordance with some embodiments.

[0040] Figure 4 shows examples of synthetic images that depict plasma anomalies in accordance with some embodiments.

[0041] Figures 5A and 5B illustrates synthetic data usable to train a model to detect hollow cathode discharges in accordance with some embodiments.

[0042] Figure 6 shows examples of synthetic images that depict film growth in accordance with some embodiments.

[0043] Figures 7A and 7B depict synthetic images that may be used to train a model to detect a wafer’s angle or angular velocity within a fabrication chamber in accordance with some embodiments.

[0044] Figures 7C, 7D, and 7E illustrate images that may be used to detect a wafer’s angle or angular velocity within a fabrication chamber in accordance with some embodiments.

[0045] Figures 7F and 7G depict example wafer arm angles and shapes that may be used to train a model to detect a wafer’s angle or angular velocity within a fabrication chamber in accordance with some embodiments.

[0046] Figure 8 shows an example synthetic image of various chamber components in accordance with some embodiments.

[0047] Figures 9A and 9B illustrate a centered wafer and an offset wafer, respectively, in accordance with some embodiments.

[0048] Figure 10 is an annotated synthetic image that may be used to determine a pedestal tilt in accordance with some embodiments.

[0049] Figure 11 A is a flowchart of an example process for training a model, using synthetic images, to generate one or more enhanced images and / or to determine modified camera or light settings in accordance with some embodiments.

[0050] Figure 1 IB is a flowchart of an example process for using a trained model to generate one or more enhanced images and / or to determined modified camera and / or light settings in accordance with some embodiments.

[0051] Figure 12 illustrates example synthetic images that may be used to train a model to generate deblurred enhanced images in accordance with some embodiments.

[0052] Figure 13 is a block diagram ofa system for utilizing images to generate, using a model, a reconstructed video of a fabrication apparatus, in accordance with some embodiments.

[0053] Figure 14 depicts example techniques and systems for utilizing real and synthetic images to generate a three-dimensional reconstruction of a fabrication apparatus in accordance with some embodiments.

[0054] Figure 15 illustrates example graphs that depict the effects of camera settings on light intensities in accordance with some embodiments.

[0055] Figure 16 presents an example computer system that may be employed to implement certain embodiments described herein.

[0056] Figure 17 depicts an example synthetic image of droplets on a wafer in accordance with some embodiments.

[0057] Figures 18A and 18B depict example synthetic images illustrating liquids in an electrofill system in accordance with some embodiments.DETAILED DESCRIPTION

[0058] Disclosed herein are methods, systems, and media for using synthetic photo-realistic images of fabrication apparatuses. Disclosed herein are techniques for training a machine learning model using one or more synthetic images. A synthetic image may depict an anomalous condition associated with a fabrication apparatus and / or a wafer undergoing fabrication, and the model may be trained to detect anomalous conditions based on the training set. By utilizing a training set comprising synthetic images, the model may be trained with relatively more training samples depicting anomalous conditions than may be obtained if only real images were used to train the model. Therefore, a model trained using synthetic images depicting anomalous conditions may be more accurate and / or more robust. Moreover, in some cases, certain anomalous conditions are hard to replicate in real time or in near real-time. By synthetically generating images representing anomalous conditions, a model can be trained to identify the anomalous conditions when encountered in real-time. Furthermore, responsive to a model detecting an anomalous condition, a real image may be captured which can then be used to update or re-train the model. Additionally or alternatively, in some embodiments, synthetically generated images may be used to overcomethe manual resources typically needed to label a training set of real images. Conventionally, a training set of real images may need to be labeled (e.g., with ground truth classifications) prior to training a model with the training set. However, synthetic images may be auto-labeled by an algorithm (e.g., an algorithm that generates the synthetic images, using computer-aided design information depicting a fabrication apparatus, etc.), which may drastically reduce the manual effort needed to generate a training set.

[0059] Disclosed herein are techniques for training a machine learning model to generate enhanced images. For example, the enhanced images may include a deblurred version of one or more real images provided to the model as an input. As another example, the enhanced images may include a video that depicts, e.g., a current state of a fabrication apparatus, where one or more frames of the video comprise synthetic images. As yet another example, the enhanced images may include a three-dimensional reconstruction of a fabrication apparatus that stitches together two or more two-dimensional real or synthetic images.

[0060] Disclosed herein are techniques for using synthetic images to train a machine learning model to determine modified settings for one or more physical camera devices and / or lighting devices. By determining and implementing modified settings for one or more physical devices, aspects of the apparatus, such as a lighting intensity, may be controlled over time regardless of overall system drift over time.

[0061] Figure 1A shows a fabrication tool denoted as substrate processing apparatus 100. Apparatus 100 may be configured for depositing films on or over a semiconductor substrate utilizing any number of processes. For example, apparatus 100 may be adapted for performing, in particular, PECVD, ALD, or ALE. Apparatus 100 includes a camera or camera sensor 117 on a chamber wall. Camera sensor 117 is configured to capture image data from the interior of apparatus 100.

[0062] Processing apparatus 100 of Figure 1A may employ a single process station 102 of a process chamber with a single substrate holder 108 (e.g., a pedestal) in an interior volume, which may be maintained under vacuum by a vacuum pump 118. A showerhead 106 and a gas delivery system 101, which are fluidically coupled to the process chamber, may permit the delivery of film precursors, for example, as well as carrier and / or purge and / or process gases, secondary reactants, etc.

[0063] In Figure 1 A, gas delivery system 101 includes a mixing vessel 104 for blending and / or conditioning process gases for delivery to showerhead 106. One or more mixing vessel inletvalves 120 may control introduction of process gases to mixing vessel 104. Particular reactants may be stored in liquid form prior to vaporization and subsequent delivery to process station 102 of a process chamber. The implementation of Figure 1A includes a vaporization point 103 for vaporizing liquid reactant to be supplied to mixing vessel 104. In some implementations, vaporization point 103 may include a heated liquid injection module. In some other implementations, vaporization point 103 may include a heated vaporizer. In yet other implementations, vaporization point 103 may be eliminated from the process station. In some implementations, a liquid flow controller upstream of vaporization point 103 may be provided for controlling a mass flow of liquid for vaporization and delivery to process station 102.

[0064] Showerhead 106 may operate to distribute process gases and / or reactants (e.g., film precursors) toward substrate 112 at the process station, the flow of which may be controlled by one or more valves upstream from the showerhead (e.g., valves 120, 120A, 105). In the implementation depicted in Figure 1A, substrate 112 is depicted as located beneath showerhead 106, and is shown resting on a pedestal 108. Showerhead 106 may include any suitable shape and may include any suitable number and arrangement of ports for distributing process gases to substrate 112. In some implementations involving two or more stations, gas delivery system 101 includes valves or other flow control structures upstream from the showerhead, which can independently control the flow of process gases and / or reactants to each station so as to permit gas flow to one station while prohibiting gas flow to a second station. Furthermore, gas delivery system 101 may be configured to independently control process gases and / or reactants delivered to each station in a multi-station apparatus such that the gas composition provided to different stations is different; e.g., the partial pressure of a gas component may vary between stations at the same time.

[0065] In the implementation of Figure 1A, gas volume 107 is depicted as being located beneath showerhead 106. In some implementations, pedestal 108 may be raised or lowered to expose substrate 112 to gas volume 107 and / or to vary the size of gas volume 107. The separation between pedestal 108 and showerhead 106 is sometimes referred to as a “gap.” Optionally, pedestal 108 may be lowered and / or raised during portions of the deposition process to modulate process pressure, reactant concentration, etc., within gas volume 107. Showerhead 106 and pedestal 108 are depicted as being electrically coupled to RF signal generator 114 and matching network 116 for coupling power to a plasma generator. Thus, showerhead 106 may function as an electrode for coupling radio frequency power into process station 102. RF signal generator 114 and matching network 116 may be operated at any suitable RF power level, which may operate to form plasma having a desired composition of radical species, ions, and electrons. In addition, RFsignal generator 114 may provide RF power having more than one frequency component, such as a low-frequency component (e.g., less than about 2 MHz) as well as a high frequency component (e.g., greater than about 2 MHz). In some implementations, plasma ignition and maintenance conditions are controlled with appropriate hardware and / or appropriate machine-readable instructions in a system controller which may provide control instructions via a sequence of input / output control instructions.

[0066] For simplicity, processing apparatus 100 is depicted in Figure 1A as a standalone station (102) of a process chamber for maintaining a low-pressure environment. However, some fabrication tools employ a plurality of process stations such as shown in Figure IB, which schematically depicts an implementation of a multi-station fabrication tool 105. Fabrication tool 150 employs a process chamber 165 that includes multiple fabrication process stations, each of which may be used to perform processing operations on a substrate held in a wafer holder, such as pedestal 108 of Figure 1A, at a particular process station. In the implementation of Figure IB, the process chamber 165 is shown as having four process stations 151, 152, 153, and 154. However, in certain other implementations, multi-station processing apparatuses may have more or fewer process stations depending on the implementation and, for instance, the desired level of parallel wafer processing, size / space constraints, cost constraints, etc. Figure IB additionally shows substrate handler robot 175, which may operate under the control of system controller 190, configured to move substrates from a wafer cassette (not shown in Figure IB) from loading port 180 and into multi-station process chamber 165, and onto one of process stations 151, 152, 153, and 154.

[0067] As depicted, process station 153 has an associated camera or camera sensor 121 located and configured to obtain images from within process station 153 and, in some embodiments, from within process chamber 154. Process station 151 has two associated cameras or camera sensors 123 and 125. Camera sensor 123 is located and configured to obtain images from within process station 151 and, in some embodiments, from within process chamber 152. Camera sensor 125 is located and configured to obtain images from within process station 151 and, in some embodiments, from within process chamber 153. Process station 152 has an associated camera or camera sensor 127 located and configured to obtain images from within process station 152 and, in some embodiments, from within process chamber 154. Any one or more of camera sensors 121, 123, 125, and 127 may be optically coupled to the interior of process chamber 165 via view port, window, or other optical access aperture disposed in or on the chamber such as in the chamber wall. It should be understood that Figures 1A, IB, and other system / apparatus figures herein merely present examples showing the number and location of camera sensors. Any number ofcamera sensors may be employed for a fabrication tool, and these sensors may be located at any of various positions in or around the tool and its chamber wall. For example, one or more cameras may be used at any given station. Further, a given camera sensor may be positioned and oriented to “look” at other stations other than the one it is most directly associated with (within line of sight). A chamber wall may include a substantially vertically aligned peripheral portion and / or a top portion and / or a bottom portion. A window or optical aperture may be placed in any one or more such portions. A camera sensor associated with any window or aperture may have a substantially vertical line of sight (e.g., it is substantially parallel to a vertical portion of a chamber wall), a substantially horizontal line of sight (e.g., it is substantially parallel to top and / or bottom portion of a chamber wall), or an oblique line of sight.

[0068] It should be noted that, as used herein, a camera sensor may capture light in various wavelength ranges, such as in the ultraviolet range, in the visible range, in the infrared range, or the like. In some implementations, a camera sensor may be associated with various spectral filters that are tuned for various wavelengths of interest. In some implementations, image data captured by a camera sensor may be digitally filtered to obtain data relevant to particular wavelengths. In some implementations, different channels may be collected separately by a camera sensor. For example, in some embodiments, a camera sensor may collect data for red, green, and blue channels separately such that the individual channels may be separately processed.

[0069] Fabrication tool 150 includes a system controller 190 configured to control process conditions and hardware states of process tool 150. It may interact with one or more sensors, gas flow subsystems, temperature subsystems, and / or plasma subsystems — collectively represented as block 191 — to control process gas flow, thermal conditions, and plasma conditions as appropriate for controlling a fabrication process. System controller 190 and subsystems 191 may act to implement a recipe or other process conditions in the stations of process chamber 165.

[0070] In multi-station fabrication tools, an RF signal generator may be coupled to an RF signal distribution unit, which is configured to divide the power of the input signal into, for example, four output signals. Output signals from an RF signal distribution unit may possess similar levels of RF voltage and RF current, which may be conveyed to individual stations of a multi-station fabrication tool.

[0071] In some implementations, images captured from a particular camera sensor that is oriented predominantly to capture images from a first station of a multi-station apparatus may additionally capture images associated with other stations of the multi-station apparatus.

[0072] In some embodiments, a machine learning model may be trained using a training set that includes one or more synthetic images. Each synthetic image may be representative of a process performed in a fabrication apparatus, one or more components of the fabrication apparatus, a wafer undergoing processing in the fabrication apparatus, or the like. In some embodiments, each synthetic image may be a photo-realistic image. For example, each synthetic image may be photo-realistic compared to an image that may be captured using one or more physical cameras disposed in or on the fabrication apparatus (e.g., as shown in and described above in connection with Figures 1A and IB). Each synthetic image may be generated using, e.g., a generative- adversarial network (GAN) or another type of generative machine learning model. Note that, as described below in connection with Figures 3 A, 3B, 4, 5A, 5B, 6, 7A, 7B, 7C, 7D, 7E, 7F, 7G, 8, 9 A, 9B, and 10, the synthetic images may represent an anomalous condition associated with a wafer undergoing processing in the fabrication apparatus, and / or an anomalous condition associated with one or more components of the fabrication apparatus. Note that, in some embodiments, the synthetic images may also include images representing non-anomalous conditions, which may be used to not create class imbalances in the training set. Moreover, such non-anomalous synthetic images may be used to explore advantages of different types of cameras, such as a higher resolution camera before investing in such a device. An example of an anomalous condition associated with the wafer may include a wafer offset (e.g., not centered on a pedestal or chuck) due to, e.g., misaligned lift pins, pedestal tilt, pressure differences for a non-clamped system, etc. Examples of anomalous conditions associated with one or more components of the fabrication apparatus include instances of parasitic plasma, instances of hollow cathode discharges (HCDs), instances of misaligned apparatus components, or the like. By generating synthetic images that represent anomalous conditions, more training examples depicting anomalous conditions may be used to train the model relative to real images captured by one or more cameras, due to the anomalous conditions not occurring sufficiently frequently to capture a suitable number of real images. In some embodiments, the synthetic images may represent example locations of a wafer, e.g., on a rotational indexer or other robot arm within the fabrication apparatus. For example, a wafer may be moved from station to station via a robot arm, or rotate within a station using a rotational indexer configured to rotate about a rotational axes within the station (sometimes referred to herein as “a spindex.”) Synthetic images may be used to determine an angular rotation or angular velocity of the wafer when being rotated within a station or moved from one station to another. Determination of the angular rotation or angular velocity may be used to determine anomalies with a robot arm or spindex. For example, in an instance in which a spindex is to rotate a wafer by 90 degrees and in which a trained model determines that the spindex has rotated the wafer 89 degrees based on the determined angular rotation or angular velocity of the wafer, ananomaly with the spindex (e.g., one or more gears of the spindex, etc.) may be identified, and a controller may then issue instructions that compensate for the anomalies of the spindex.

[0073] In some embodiments, the machine learning model may additionally or alternatively be trained using data indicative of a design of the fabrication apparatus and / or information indicative of one or more processes to be performed using the fabrication apparatus. For example, data indicative of a design of the fabrication apparatus may include one or more computer-aided design (CAD) diagrams of the fabrication apparatus. In some embodiments, the design information may be utilized to auto-label training data used to train the machine learning model and / or a downstream machine learning model. As another example, information indicative of one or more processes to be performed using the fabrication apparatus may include information indicative of gas flow and / or plasma physics that are utilized when performing one or more processes in the fabrication apparatus. As yet another example, information indicating camera properties and / or lens properties of one or more cameras associated with the fabrication apparatus may be utilized to train the machine learning model.

[0074] The machine learning model may have any suitable architecture. For example, the machine learning model may be a generative model, such as a generative-adversarial network model (GAN). As another example, the machine learning model may be a deep learning network, such as a convolutional neural network (CNN), a recurrent neural network (RNN), autoencoders, or the like. The machine learning model may be trained using the information indicating design of the fabrication process, information associated with cameras of the fabrication apparatus, and / or training images of the training set (which may include one or more synthetic images).

[0075] Figure 2A is a flowchart of an example process 200 for training a machine learning model in accordance with some embodiments. Blocks of process 200 may be implemented by one or more processors of a computing device, such as a server device. In some embodiments, blocks of process 200 may be executed in an order other than that shown in Figure 2A. In some embodiments, two or more blocks of process 200 may be executed substantially in parallel. In some embodiments, one or more blocks of process 200 may be omitted.

[0076] At 202, process 200 can receive data indicative of a design of a fabrication apparatus and / or information indicative of processes performed in the fabrication apparatus. The data indicative of the design of the fabrication apparatus may include CAD data, or the like. The information indicative of processes performed in the fabrication apparatus may include information indicative of gas flows and / or plasma within the fabrication apparatus duringperformance of the processes. The information may additionally or alternatively include information indicative of camera properties of one or more cameras disposed in or on the fabrication apparatus and used to capture images of components of the fabrication apparatus and / or wafer undergoing processing.

[0077] At 204, process 200 can generate a set of training images, the set of training images including a subset of synthetic images. The set of training images may include real and / or synthetic images of a wafer undergoing processing, one or more components of the fabrication apparatus, etc. Note that the content of the images, whether real or synthetic, may be dependent on the purpose of the model being trained (e.g., at block 206). For example, a model configured to predict or detect parasitic plasma may be trained with real and synthetic images a portion of the fabrication apparatus such as a region between a pedestal and a showerhead, or the like. As another example, a model configured to predict or detect a wafer being off-center may be trained using real and synthetic images of wafer during processing and / or real and synthetic images of components of the fabrication apparatus that affect wafer centering, such as lift pins of the pedestal.

[0078] At 206, process 200 may train a machine learning model using the set of training images, the data indicative of the design of the fabrication apparatus, and / or the information indicative of processes performed in the fabrication apparatus. The machine learning model may have any suitable architecture, such as that of a GAN, a deep learning network (e.g., a CNN, an RNN, etc.), or the like. In some embodiments, conventional computer vision techniques, e.g., that perform object recognition, edge detection, etc. may be utilized.

[0079] In some embodiments, a trained machine learning model may take, as input, video and / or image data associated with a fabrication apparatus. The video and / or image data may be real images and / or real video comprised of real images. The images and / or video may be captured from one or more cameras disposed in or on the fabrication apparatus. In some embodiments, the trained machine learning model may be configured to generate a prediction based on the input video and / or image data. For example, the prediction may be a prediction of an anomalous state associated with a wafer undergoing processing, an anomalous state of one or more components of the fabrication apparatus, or the like. As another example, the prediction may be a current angular rotation or angular velocity associated with a wafer being rotated by a spindex and / or a robot arm of the apparatus. Examples of these use cases are shown in and described below in connection with FIGS. 3A, 3B, 54, 5A, 5B, 7, 7A, 7B, 7C, 7D, 7E, 7F, 7G, 8, 9A, 9B, and 10. As another example, the output of the machine learning model may be an enhanced image and / or a predictionof modified camera settings for one or more cameras of the apparatus that may yield improved images and / or video. An enhanced image may include a deblurred image, a three-dimensional reconstruction of the apparatus, a real-time video of the apparatus, or the like. Examples of these use cases are shown in and described below in connection with FIGS. 11 A, 11B, 12, 13, 14, and 15.

[0080] Figure 2B is a flowchart of an example process 250 for utilizing a trained machine learning model to generate outputs associated with a fabrication apparatus in accordance with some embodiments. Blocks of process 250 may be implemented by one or more processors of a computing device, such as a server device, or an edge compute node. For example, in some embodiments, at inference time, real-time video may be processed on an edge compute node proximate to a fabrication tool such that decisions and / or actions may be taken in near real-time based on the video. In other embodiments, video data may be transmitted to a remote server device (e.g., in instances in which real-time or near real-time processes is not needed). In some embodiments, blocks of process 250 may be executed in an order other than that shown in Figure 2B. In some embodiments, two or more blocks of process 250 may be executed substantially in parallel. In some embodiments, one or more blocks of process 250 may be omitted.

[0081] At 252, process 250 can obtain video and / or image data associated with a fabrication apparatus. The video and / or image data may be real video and / or image data obtained using one or more cameras disposed in or on the fabrication apparatus (e.g., as shown in and described above in connection with Figures 1 A and IB).

[0082] At 254, process 250 can utilize a machine learning model trained using a set of training images that includes a subset of synthetic images to generate a prediction associated with the video and / or image data. In other words, the trained machine learning model may take the real video and / or image data as an input, and generate a prediction corresponding based on the input. The synthetic images used to train the model may be photo-realistic synthetic images corresponding to the real and / or video data used as input. For example, the synthetic images may be synthetic images of the same portion of the fabrication apparatus, or the like. The prediction may be an anomalous state of a wafer being processed, an anomalous state of one or more components of the fabrication apparatus, a current location or angular rotation or velocity of wafer in the apparatus, an enhanced image based on the input image or video, or modified camera settings to be applied to a camera (e.g., a camera that captured the video and / or image data obtained at block 252).

[0083] At 256, process 250 can take at least one action based on an inference generated by the trained machine learning model. For example, in some embodiments, the at least one action may include recording data in a data file and / or recording errors in a data file or a log file. As another example, in some embodiments, the at least one action may include displaying errors, e.g., in a console associated with the tool. Note that the data may include data relating to a wafer undergoing processing, data associated with one or more components of the tool, etc. An error may include an indication of an anomalous condition associated with a wafer undergoing processing and / or a component of the tool. As yet another example, in some embodiments, the at least one action may include shutting down the fabrication apparatus. In such instances, a user may be required to restart the tool and / or re-start the fabrication process. As still another example, in some embodiments, the at least one action may include causing a real-time or near real-time control action to be performed, e.g., utilizing one or more controllers of the fabrication apparatus. The control action may include changing process parameters, e.g., to overcome or compensate for an error detecting based on the inference. Control actions may include, e.g., changing RF power settings, changing gas flow, changing a number of deposition cycles, etc. Note that, in some embodiments, the control action may not only compensate for an error, but may additionally compensate for the time required to implement the control action. In some embodiments, a control action may be determined and implemented by a subsequent fabrication apparatus used to process the wafer to compensate for anomalies or errors caused by the current fabrication apparatus.

[0084] In some embodiments, a machine learning model may be trained with a set of training images that includes a subset of synthetic images. Each synthetic image may represent an anomalous condition associated with the fabrication chamber. For example, the anomalous condition may represent an anomalous wafer state (e.g., the wafer being mispositioned or misaligned), an anomalous condition associated with one or more components of the apparatus, anomalous plasma characteristics within the apparatus (e.g., parasitic plasma being present), or the like. Each synthetic image may be a photo-realistic image corresponding to an image that may be taken with a camera disposed in or on the fabrication apparatus. In some embodiments, anomalies represented in the set of synthetic images may occur at relatively low frequencies, such that obtaining real images depicting such anomalies may not be possible. Accordingly, the synthetic images may be used to train the machine learning model to detect such anomalous conditions in situations where it is not practicable to obtain real images. The machine learning model may then be trained to detect one or more anomalous conditions. Examples of such anomalous conditions are shown in and described below in connection with FIGS. 4-10.

[0085] Figure 3A is a flowchart of an example process 300 for training a machine learning model to predict or detect at least one anomalous condition using a training set that comprises at least one synthetic image in accordance with some embodiments. Blocks of process 300 may be implemented by one or more processors of a computing device, such as a server device. In some embodiments, blocks of process 300 may be executed in an order other than that shown in Figure 3 A. In some embodiments, two or more blocks of process 300 may be executed substantially in parallel. In some embodiments, one or more blocks of process 300 may be omitted.

[0086] Process 300 can begin at 302 by obtaining a set of images associated with a fabrication apparatus, where the set of images comprises at least a subset of synthetic images. The synthetic images may represent at least one anomalous condition associated with the fabrication apparatus. Examples of anomalous conditions include a wafer being misaligned or mispositioned on a pedestal or chuck, a misalignment of one or more chamber components, an occurrence of parasitic plasma, an occurrence of HCDs, or the like.

[0087] At 304, process 300 can train the machine learning model using the set of images, where the trained machine learning model is usable to predict the at least one anomalous condition represented in the subset of synthetic images. The machine learning model have any suitable architecture, such as a deep learning network (e.g., a CNN, an RNN, etc.). The machine learning model may be trained using various techniques, such as b ackpropagation, etc. In some embodiments, the training set may be annotated using design information associated with the apparatus, e.g., a CAD design of the apparatus.

[0088] Figure 3B is a flowchart of an example process 350 for utilizing a trained machine learning model to predict or detect at least one anomalous condition in accordance with some embodiments. Blocks of process 350 may be implemented by one or more processors of a computing device, such as a server device and / or edge compute node. In some embodiments, blocks of process 350 may be executed in an order other than that shown in Figure 3B. In some embodiments, two or more blocks of process 350 may be executed substantially in parallel. In some embodiments, one or more blocks of process 350 may be omitted.

[0089] Process 350 may begin at 352 by receiving data from one or more cameras disposed in or on a portion of a fabrication apparatus. The data may be image data or video data. The data may include content depicting a wafer being processed, one or more components of the fabrication apparatus, or the like.

[0090] At 354, process 350 may provide the image data to a trained machine learning model to generate an output, where the trained machine learning model was trained using a set of images comprising at least a subset of synthetic images, and where each synthetic image is a photorealistic image of an image captured by the one or more cameras. An example of a training process is shown in and described above in connection with Figure 3A.

[0091] At 356, process 350 may utilize the output of the machine learning model to predict at least one anomalous condition of the fabrication apparatus or a wafer undergoing processing in the apparatus. Example use cases corresponding to an output of the machine learning model are shown in and described below in connection with Figures 4-10.

[0092] At 358, process 350 may take at least one action based on the at least one anomalous condition. Similar to what is described above in connection with block 256 of Figure 2B, the at least one action may include recording the at least one anomalous condition in a log file, displaying an indication of the at least one anomalous condition (e.g., on a console), causing a shut down of the fabrication apparatus, and / or causing at least one compensatory control action to be performed. The at least one compensatory control action may include changes to process parameters and / or changes to components of the tool. The at least one compensatory control action may be implemented on the current fabrication apparatus, or may be implemented on a downstream fabrication apparatus to compensate for the at least one anomalous condition in a downstream process performed on the wafer.

[0093] In some embodiments, synthetic images may be photo-realistic images that depict various characteristics of plasma during plasma-based fabrication operations. For example, the synthetic images may depict examples of non-anomalous plasma conditions, examples of parasitic plasma (e.g., plasma located in regions of a station other than the desired regions of the station, such as outside of a region between a showerhead and a pedestal or chuck on which a wafer is positioned), examples of hollow cathode discharges (HCDs), or the like. In some examples, anomalous plasma conditions may be represented in the set of synthetic images at a rate or frequency higher than a rate or frequency at which the anomalous plasma conditions occur in reality. The set of synthetic images may be generated using one or more machine learning models, such as a GAN, autoencoders, or other generative model. In some embodiments, synthetic images may be generated based on behavioral or mathematical parameters (e.g., relating to functions to numerically simulate a process) or from computer simulations that simulate the physics and / or chemistry of a given process. The set of synthetic images may be generated using real images of anomalous and non-anomalous plasma conditions, equations that correspond to plasma generationor flow, etc. The set of synthetic images may be included in a training set, which may include the set of synthetic images and optionally one or more real images, to train a machine learning model. The machine learning model may be trained to generate, as an output, a prediction or a detection of an anomalous plasma condition given an input of real image or video data, e.g., of portions of a station or chamber in operation. Note that, in some embodiments, the training set may be annotated manually and / or using algorithmic labeling. For example, algorithmic labeling may utilize various computer vision techniques to detect, e.g., plasma non-uniformities, HCDs, parasitic plasma locations, or the like.

[0094] Figure 4 depicts various examples of synthetic images of plasma. For example, images 404 and 406 depict non-anomalous plasma having varying intensities. Note that images 404 and 406 may represent different process conditions. Images 402, 408, 410, 412, 414, 416, 418, 420, and 422 depict examples of parasitic plasma, in which plasma is located at undesired regions of the station (e.g., above a showerhead, below the pedestal, beside the showerhead or pedestal, etc.). In other words, parasitic plasma is generally classified as plasma that does not contribute to the main plasma utilized in the process. Changes in a shape of parasitic plasma may indicate a failure or an imminent failure. Note that the synthetic images depicted in Figure 4 are merely examples. Moreover, in some embodiments, a synthetic image may include two or more examples of anomalous plasma conditions, such as non-uniform plasma, HCDs, parasitic plasma, or the like.

[0095] In some embodiments, synthetic images may be photo-realistic images depicting ingap HCDs, which are generally instances of anomalous plasma within the gap between the showerhead and pedestal. Note that, in some cases, HCDs may occur outside of the gap between the showerhead and pedestal. In some embodiments, synthetic images of HCDs (and / or other types of anomalous plasma) may be generated using design information of a specific tool (e.g., a CAD design of a specific apparatus or tool), gas flow and / or plasma physics information, information relating to camera and / or lens properties of one or more cameras that would be used to image plasma conditions, or any combination thereof. The synthetic images may be generated using a GAN, algorithms that simulate lighting interactions, physics-based or behavior based simulations or other numeric simulations, or other typs of models. By utilizing multiple sources of information, in some embodiments, the synthetic images may be used to train a machine learning model that is configured to detect instances of HCDs, as well as spatial locations of HCDs. The spatial location may include (x, y) coordinates, which may be determined using computer vision techniques applied to real images. In some embodiments, a trained machine learning model may predict a z-depth of an HCD based on the training data used to train the machine learning model, where the training data may include synthetic images of HCDs at different z-depths. Insome embodiments, the trained model may be configured to take as input multiple images captured concurrently from different cameras, each positioned at a different orientation with respect to the apparatus. Use of multiple images may allow for reduced uncertainty, e.g., in predicting a z-depth of a given HCD.

[0096] Figures 5 A and 5B illustrate use of synthetic images for detecting HCDs and predicting spatial location (including z-depth) of detected HCDs. As illustrated in Figure 5A, one or more synthetic images of HCDs (e.g., image 502) may be generated, with each HCD at different spatial locations specified by (x, y) coordinates. As illustrated in image 504, each HCD of the synthetic image may also have a different z-depth. The model may then be trained using the synthetic image and the z-depth information. Image 506 illustrates a three-dimensional reconstruction of a portion of a fabrication apparatus that includes the HCDs of the synthetic image 502. A machine learning model may be trained using the synthetic images (and optionally, one or more real images) of HCDs. Note that a synthetic image may be constructed using CAD design information associated with the tool, lighting information associated with the tool and / or optics properties of the camera, or the like. For example, an HCD nearer to a location of a camera may appear larger than an HCD farther from the camera in a given synthetic image.

[0097] Turning to Figure 5B, images depicting use of a trained model for HCD detection at inference time are shown in accordance with some embodiments. As illustrated, image 552 indicates a real image that may be captured from a camera associated with a fabrication apparatus. HCD instances are marked in image 552, which may be detected via the trained model, via computer vision techniques, etc. Image 554 illustrates use of image 552 by a trained machine learning model to predict z-depths associated with each HCD. Image 556 illustrates a three- dimensional reconstruction of a portion of the apparatus with the HCDs positioned at their detected spatial locations.

[0098] In some embodiments, synthetic images may be used to train a machine learning model configured to detect film growth on various surfaces within a station or chamber, and / or to predict an amount of film growth on various surfaces within the station or chamber. Output of the trained machine learning model may be used to determine a time point at which to initiate a cleaning cycle of the station or chamber, e.g., based on the predicted amount of film growth exceeding a cleaning threshold. In other words, chamber accumulation may be predicted by correlating a thickness profile on a given component (e.g., an edge ring) to the predicted chamber accumulation. In some embodiments, if the correlation between thickness profile and chamber accumulation drifts (e.g., due to surface degradation), a model may be able to learn the system drift over time, e.g., due toupdated training samples provided to the model. Note that film may grow on surfaces of a station or chamber at different rates (e.g., different growth rates for different surfaces). Moreover, some surfaces may be easier to visualize with camera data than others. Accordingly, synthetic images may be generated for a variety of surfaces, which may allow the machine learning model to detect film growth and / or quantify film growth more accurately, even for difficult to visualize regions of a station or chamber. In some embodiments, synthetic images depicting film growth may be generated using design information regarding a station or chamber, stochastic models of film growth for a given process and / or region of the apparatus, one or more real images depicting film growth, or the like. The trained machine learning model may be one or more types of models (e.g., a CNN, or other deep learning network), and may additionally or utilize various computer vision techniques to identify and quantify regions of film growth.

[0099] Figure 6 depicts synthetic images 602, 604, 606, and 608. Each of images 602, 604, 606, and 608 depicts differing amounts of film growth. Note that images 602, 604, and 606 depict film growth when backlit by plasma, whereas image 608 depicts the film with the light on and no plasma. It may not be possible to capture a real image corresponding to image 608, because it may be dangerous to open a station or chamber with such accumulation due to toxicity. However, images such as image 608 may be provide useful information to a process engineer, and / or may be used to train additional downstream models.

[0100] In some embodiments, synthetic images may be used to train a machine learning model configured to detect a rotation angle, an angular velocity, and / or a position of a wafer in a station or chamber, e.g., a wafer being rotated (e.g., by a spindex), moved from a first station to a second station by a robot arm, and / or moved from a vacuum transfer module to a process module by a robot arm. In some embodiments, the synthetic images may depict a top-down view of the wafer and / or a side-view of the wafer to correspond to real images that may be captured by a camera disposed in a top-down viewport and / or a side-view viewport. In some embodiments, the machine learning model may be trained (e.g., using the synthetic images and / or one or more real images) to identify the location of a wafer, e.g., by performing semantic segmentation on the image. After identification of the wafer by the model, location information associated with the wafer (e.g., rotational angle as a function of time, angular velocity, position as a function of time, etc.) may be determined using regression analysis performed on, e.g., two or more successive images.

[0101] Figures 7A and 7B depict synthetic images of top-down views of a spindex configured to rotate a wafer, e.g., 90 degrees and a rotational indexer configured to rotate a wafer from one station to another, respectively. As described above, as used herein, a spindex is a robot armconfigured to rotate about its own rotational axes, e.g., to rotate a wafer within a given station. Note that a spindex and / or a rotational indexer may generally be configured to rotate or move a wafer by a given, predetermined angle, e.g., 90 degrees. However, due to system drift, such as process chemistry that affects the gears of a spindex or rotational indexer, the spindex and / or rotational indexer may fail to rotate or move the wafer by the predetermined angle. For example, rather than 90 degrees, a wafer may be rotated 89 degrees. A model may be trained, using synthetic images, to detect the angular rotation or angular velocity of a wafer to detect instances in which the spindex or rotational indexer fails to rotate or move the wafer by the predetermined angle. Responsive to detecting an anomalous rotation, a control action may be performed to compensate for the failure to rotate or move the wafer by the predetermined angle. For example, responsive to determining that a spindex has rotated the wafer within a station by 89 degrees rather than the desired 90 degrees, control instructions may be provided to the spindex to perform extra rotation in order to achieve the desired 90 degrees.

[0102] The example synthetic images shown in Figures 7A and 7B reflect a camera with a top- down view of a given station. Acquiring real images from such a camera may require installation of additional viewports in a tool, whereas such a tool may already be equipped with cameras in side viewports. In such cases, it may be desirable to train a model to detect angular rotation and / or angular velocity using information obtained from a side view camera rather than installing and configuring a top-down viewport camera. In some embodiments, the side view information may include images of a side view of a calibration disc (e.g., a calibration wafer). Figures 7C, 7D, and 7E depict side views of example calibration discs. Synthetic images may be generated that depict patterns on the side of a given calibration disc at, e.g., different angular rotations or different angular velocities. A model may then be trained using the synthetic images to predict an angular rotation or velocity associated with a given image. At inference time, the angular rotation or velocity may be determined based on real side view images of the calibration disc. Figure 7C illustrates an example side view of a calibration disc that utilizes different sized dots. Figure 7D illustrates an example side view of a calibration disc that utilizes rectangles having different widths and spacings. Figures 7E illustrates an example side view of a calibration disc that utilizes a different pattern of numbers. Note that the sizings and spaces between pattern elements are an example, and, in some embodiments, a calibration disc may utilize pattern elements having the same size and different inter-element spacings, different sizes and the same inter-element spacing, or any combination thereof.

[0103] In some embodiments, a calibration disc may be impractical to detect wafer angular rotation or velocity. For example, in stations that are at high temperature, material choices maybe limited for a calibration disc. In some embodiments, wafer angular rotation or velocity may be determined based on a shape of a robot arm or robot arm end effector. Figures 7F and 7G depict different wafer arm shapes and angles that may be used by a model to determine wafer position. Note that, the wafer arm shape and angle may be determined using a side view camera. Figure 7F depicts three different example rotations of a robot arm end effector (e.g., for a spindex). In some embodiments, a model may be trained using synthetic images of the end effector at different angles, each corresponding to a different wafer rotation. At inference time, an image of the end effector may be captured, and the trained model may output the wafer angular rotation or velocity. Figure 7G depicts different angular rotations of a robot arm, e.g., configured to move a wafer from one station to another. In some embodiments, a model may be trained using synthetic images of the robot arm at different angular rotations, each corresponding to a different wafer rotation from one station to another. At inference time, an image of the robot arm may be captured, and the trained model may output the wafer position with respect to a given station or a wafer rotation with respect to a given station.

[0104] In some embodiments, synthetic images may be used to train models to detect various components of a station or chamber, such as a pedestal, one or more lift pins, a showerhead, robot arms, a wafer, etc. Detection of components that may move, e.g., during processing, may be especially critical, as these components may become out of alignment due to system drift over time. Additionally, detection of components such as a showerhead may be critical for identifying where plasma is relative to a wafer, or the like. In some embodiments, the synthetic images may be used to train a model configured to perform semantic segmentation, e.g., to identify particular components within the images. Such models may be a CNN or other deep learning network, a GAN, or the like. At inference time, after segmentation (e.g., by a model trained using one or more synthetic images), one or more downstream models may be used to determine a size of a component, coordinates of a component, a distance between two components, or the like. Downstream models may additionally be configured to detect imminent failure of one or more components.

[0105] Figure 8 depicts an example synthetic image that includes multiple station components in accordance with some embodiments. For example, the synthetic image shown in Figure 8 includes a pedestal, lift pins, a showerhead, etc. Such a synthetic image may be included in a training set used to train a machine learning model configured to perform semantic segmentation, e.g., to identify various station components in the image.

[0106] In some instances, misalignment or miscalibration of a component may cause a wafer to be offset or mispositioned. For example, uneven lift pin heights or a misaligned edge ring may cause a wafer to be offset relative to a central position. In some embodiments, synthetic images may be used to train a machine learning model which may in turn be used, at inference time, to detect wafer misalignment. For example, the trained machine learning model may be used to segment lift pins, and computer vision techniques may be applied to the segmented images to determine heights of each lift pin. A set of lift pins with uneven heights may be used to predict an offset wafer. As another example, synthetic images of wafers (offset and not offset) may be used to train a machine learning model configured to take, as an input, a real image of a wafer, and detect, as an output, whether and / or to what degree the wafer is offset. Note that, in some instances, a wafer may enter a station centered but due to pressure fluctuations (which may be due to a process recipe), the wafer may oscillate (e.g., when the wafer is not clamped), leading to an offset wafer during a portion of a process.

[0107] Figure 9A illustrates a centered (i.e., non offset) wafer 902 in accordance with some embodiments. Figure 9B illustrates an offset wafer 904 in accordance with some embodiments. Note that wafer 904 is not centered relative to pedestal 906, as indicated by the different gaps, A and A’ to the edge of pedestal 906.

[0108] As described above, in some implementations, a misaligned or tilted pedestal may cause non-uniformities in film thickness on the wafer when the wafer undergoes processing. In some embodiments, synthetic images may be used to train a machine learning model that performs semantic segmentation on an input image to identify portions of the pedestal. From the segmented portions of the pedestal, computer vision techniques may be used to identify heights of a two or more points along the pedestal. Based on the identified heights, an angle of tilt of the pedestal may be determined. Figure 10 depicts measurements of a set of heights of a pedestal to determine an angle of pedestal tilt of about 0.581 degrees. In particular, datum line 1002 represents chamber surfaces taken from pixels of a side view of a synthetic camera image. Pedestal line 1004 represents the location of the pedestal at a series of pixels of the side view of the synthetic camera image. The tilt of the pedestal may be determined based on an angle across the series of pixels between pedestal line 1004 and datum line 1002. A machine learning model trained on synthetic images of different pedestal tilts may then be configured to, at inference time, take a real side view image and generate an output pedestal tilt (which may be represented as an angle, as shown in Figure 10, or in a deviation from a specification). Note that, in some embodiments, by training a machine learning model using synthetic images, the machine learning model may be able to, at inference time, compensate for distortions (e.g., lens distortion and / or perspective distortion) inreal images. Additionally, it should be noted that by training a model on synthetic images, a pedestal tilt may then be determined when a station is at a relatively high temperature by using real images captured from a side viewport during operation of the station at the high temperature.

[0109] In some embodiments, synthetic images may be used to train a machine learning model to detect an anomalous condition of droplets on a wafer. For example, for particular processes, such as a cleaning process or an electrofill process, it may be a requirement for the wafer, after undergoing the process, to not have residual droplets on the wafer. However, in some instances, a wafer may still have remaining droplets, which may be droplets of water or any other liquid. It may be desirable to detect the liquid droplets on the wafer prior to the wafer undergoing further processing (e.g., downstream processing). In some embodiments, images captured of the wafer may be provided to a trained machine learning model, where the trained machine learning model generates, as an output, an indication of whether or not there are residual droplets on the wafer, where the droplets are located, etc. Note that detection of wafer droplets is a more challenging task for patterned wafers, as the patterns may change from application to application and the droplet interaction may not be unique. Training a machine learning model to detect wafer droplets with real image data would require an impractical amount of data collection, including manual addition of droplets and manual labeling of such droplets prior to training the machine learning model. Moreover, such manual additional of droplets may be destructive. Using synthetic image data to train such a machine learning model may allow for synthetic data representing several thousand droplet cases to be generated in a short time, thereby allowing for more robust machine learning model training. Figure 17 depicts an example synthetic image of liquid droplets on a wafer. Such synthetic images may be used to train a machine learning model configured to take real images of a wafer undergoing processing (or after undergoing a particular step of a wet chemistry process) to detect droplets on the real wafer. In some embodiments, the trained machine learning model may be configured to determine or predict a shape and / or a volume of a given droplet. In some embodiments, the shape and / or the volume of a given droplet, and / or an aggregate volume of multiple detected droplets, may be used to determine whether a drying process is necessary, and / or to determine parameters of such a drying process.

[0110] In some embodiments, synthetic images may be used to detect an anomalous condition associated with a manufacturing apparatus used for performing wet chemistry operations, such as an electrofill process. Figures 18A and 18B illustrate two such example synthetic images. Note that although Figures 18A and 18B illustrate synthetic images from a top-view perspective, in some embodiments, synthetic images may be generated that correspond to a viewpoint of physical cameras with respect to the apparatus. As illustrated in Figure 18 A, the synthetic image includesmore bubbles on the right side of the apparatus 1804 relative to the left side 1802. Accordingly, the synthetic image illustrates an example anomalous condition that leads to uneven bubbling across different sides of the apparatus. In contrast, Figure 18B is a synthetic image of a portion of the apparatus under normal operating conditions, as illustrated by the lack of bubbles in the image of Figure 18B. The synthetic images shown in Figures 18A and 18B may be used to train a machine learning model that takes, as input, a real image of a manufacturing apparatus and generates, as an output, an indication of whether the apparatus is operating normally or anomalously. Additionally, in some implementations, a trained machine learning model may be trained to automatically detect a current liquid level based on an image provided as input. It should be understood that although the examples shown in and described above in connection with Figures 18A and 18B relate to bubbles and liquid level, in some embodiments, other liquid phenomena may be represented in synthetic images, and a machine learning model may be trained to detect and / or characterize these other liquid phenomena. Examples of such liquid phenomena include liquid color change, frothing, etc.[OHl] It should be noted that, to generate synthetic images depicting liquids, whether droplets on a wafer surface, bubbles in an apparatus, or the like, the synthetic images may be generated using known material properties of the liquid to be depicted in the synthetic images. For example, material properties of water may be different than those of other liquids, such as isopropyl alcohol, which may lead to different droplet shapes, etc. Additionally, in some embodiments, material properties of solid objects with which the liquid may interact may be used to generate the synthetic images. For example, a combination of material properties of water and the specific metallic surface of a wafer may be used to generate a synthetic image of water droplets on the wafer, where the material properties are utilized to generate synthetic droplets having a certain shape, spreading on the wafer surface, surface energy, contact angle, etc.

[0112] In some embodiments, synthetic images can be used to train a machine learning model, the output of which may be used to generate one or more enhanced images. In some embodiments, an enhanced image may be a deblurred version of an input image provided to a model. In some embodiments, the one or more enhanced images may comprise a synthetic video of a portion of a fabrication apparatus. In some embodiments, the one or more enhanced images may include a three-dimensional reconstruction of at least a portion of a fabrication apparatus based on one or more two-dimensional images provided to a model. In some embodiments, the one or more enhanced images may include a super resolution image. For example, a machine learning model may be trained to take, as an input, a real image having a first resolution, and generate, as anoutput, a synthetic image having a higher resolution. For example, a 1920x1608 image may be used to generate a synthetic 4K image.

[0113] Figure 11 A is a flowchart of an example process 1100 for training a machine learning model using one or more synthetic images, wherein an output of the machine learning model is used to generate one or more enhanced images. The one or more enhanced images may be a deblurred version of an input image, a super resolution image, a three-dimensional reconstruction based on one or more two-dimensional images, a synthetic video (e.g., depicting a near real-time representation of a current state or predicted future state of a fabrication apparatus at a time point specified by a user), or the like. Blocks of process 1100 may be implemented on a server device or other computing device. In some embodiments, blocks of process 1100 may be executed in an order other than what is shown in Figure 11 A. In some implementations, two or more blocks of process 1100 may be executed substantially in parallel. In some implementations, one or more blocks of process 1100 may be omitted.

[0114] Process 1100 can begin at 1102 by generating a training set comprising a set of images including at least a subset of synthetic images associated with a fabrication apparatus. Each synthetic image may be a photo-realistic image of an image captured by a camera disposed in or on a portion of the fabrication apparatus. In some embodiments, the training set may include pairs of synthetic images, such as a blurry synthetic image and a focused synthetic image such that the machine learning model may be trained to generate the focused synthetic image using the blurry synthetic image as an input. In some embodiments, each training sample may include multiple synthetic images (e.g., three, four, ten, etc.). For example, such a training set may be used to train the machine learning model to generate a three-dimensional reconstruction based on the multiple images in the training sample, a video using the multiple images in the training sample, or the like. In some embodiments, the training set may include data from other sensors, such as spectral data, temperature data, etc.

[0115] At 1104, process 1100 can utilize the training set to train a machine learning model. The output of the machine learning model may be usable to perform at least one of: 1) generation of one or more enhanced images depicting at least a portion of the fabrication apparatus; or 2) determination of one or more modified camera settings for use in future operation of the camera disposed in or on the portion of the fabrication apparatus. Example use cases of enhanced images are shown in and described below in connection with Figures 12 and 14. Examples use cases of determining modified camera settings are shown in and described below in connection with Figure 15.

[0116] Figure 1 IB is a flowchart of an example process 1150 for utilizing a trained machine learning model that was trained using one or more synthetic images, wherein an output of the machine learning model is used to generate one or more enhanced images. The one or more enhanced images may be a deblurred version of an input image, a three-dimensional reconstruction based on one or more two-dimensional images, a synthetic video (e.g., depicting a near real-time representation of a current state of a fabrication apparatus), or the like. Blocks of process 1150 may be implemented on a server device or other computing device. In some embodiments, blocks of process 1150 may be executed in an order other than what is shown in Figure 1 IB. In some implementations, two or more blocks of process 1150 may be executed substantially in parallel. In some implementations, one or more blocks of process 1150 may be omitted.

[0117] Process 1150 can begin at 1152 by obtaining one or more images from one or more cameras disposed in or on a portion of a fabrication apparatus. The one or more images may be real images captured from any suitable location and / or orientation with respect to the fabrication apparatus. The one or more images may be images of the same region of the fabrication apparatus (e.g., captured from different angles) or may images of different regions and / or components.

[0118] At 1154, process 1150 can provide the one or more images to a trained machine learning model to generate an output. As described in connection with Figure 11 A, the machine learning model may have been trained using a training set comprising a set of images including at least a subset of synthetic images associated with a fabrication apparatus. Each synthetic image may be a photo-realistic image of an image captured by a camera disposed in or on a portion of the fabrication apparatus. Techniques for training such a model are shown in and described above in connection with Figure 11 A.

[0119] At 1156, process 1150 can utilize the output to perform at least one of: 1) generation of one or more enhanced images depicting at least a portion of the fabrication apparatus; or 2) determination of one or more modified camera settings for use in future operation of the one or more cameras disposed in or on the fabrication apparatus. For example, as shown in and described below in connection with Figure 12, the one or more enhanced images may comprise deblurred versions of the one or more images provided as input. As another example, as shown in and described below in connection with Figure 13, the one or more enhanced images may comprise synthetic images used as frames of a generated video. As another example, as shown in and described below in connection with Figure 14, the one or more enhanced images may comprise a three-dimensional reconstruction of a portion of the fabrication apparatus based on the one or more images provided as input. As yet another example, the one or more enhanced images may includea video depicting a present state of the fabrication apparatus. For example, the video may include synthetic and / or real images of a wafer undergoing processing, synthetic and / or real images of a interior portions of a station or chamber, or the like. As still another example, as shown in and described below in connection with Figure 15, one or more modified camera settings may be determined. The modified camera settings may include changes to white balancing factors, exposure time, camera gain, or the like. In some embodiments, changes to lighting settings may be determined, such as changes to lighting intensities or orientations, or the like. The modified camera and / or lighting settings may compensate for system drift in components of the apparatus. Note that, in some embodiments, the output of the machine learning model may additionally be based on data from other non-camera sensors, such as spectral sensors, temperature sensors, voltage or current sensors, or the like. Additionally or alternatively, in some embodiments, the output of the machine learning model may be used in conjunction with data from non-camera sensors to generate the enhanced image(s) and / or to determine the modified camera / lighting settings.

[0120] In some embodiments, synthetic images may be used to train a model to generate an enhanced image that is a deblurred image. Such a model may be utilized to generate an enhanced image that depicts multiple components at different depths and / or distances from a camera. In such cases, it may be difficult to acquire a real image clearly depicting these multiple components, because a real camera may not be able to adequately focus on these multiple components at different depths and / or distances from the camera. In some embodiments, the model may be trained to take, as input, a real image that depicts one or more components, and generate, as an output, an enhanced image that sharpens a visual presentation of at least a subset of the one or more components. Note that, in some implementations, the enhanced image may be used by one or more downstream models that e.g., determine a tilt of components depicted in the enhanced image, heights of components depicted in the enhanced image, etc.

[0121] Turning to Figure 12, a real image 1202 obtained using a camera device with a fixed focus is shown. Note that some components of the image are blurry relative to other components, due to the fixed focus of the camera device. Images 1204, 1206, 1208, and 1210 are synthetic images that reproduce the content of real image 1202 but with differing simulated focal planes. Enhanced image 1212 is a synthetic image that may be generated by a trained machine learning model to generate, based on any of images 1202, 1204, 1206, 1208, and / or 1210, an enhanced image. Note that components the representation of components such as lift pins 2 and 3 are sharpened in enhanced image 1212. Note that each of images 1204, 1206, 1208, and 1210 are sharpened images created by the model, and an algorithm may be used to select the image that, ifused, would estimate values associated with various components depicted in the image with the highest accuracy (e.g., position, angle, etc.). In the example shown in Figure 12, image 1212 corresponds to the selected image.

[0122] In some embodiments, synthetic images may be used to generate on-demand video, depicting, e.g., a state of a fabrication apparatus at a given time (e.g., during performance of a fabrication apparatus, during a cleaning cycle, etc.). The on-demand video may be generated by utilizing a trained machine learning model (e.g., a GAN, autoencoder, encoding / decoding network, a deep learning network, etc.) to generate frames of the video. The frames of the video may be based on a set of keyframes, which may be used to generate the synthetic frames of the video. For example, in some embodiments, a video may be generated based on less than about 50 keyframes, less than about 10 keyframes, less than about 5 keyframes, or the like. Use of a relatively small number of keyframes may allow the file size of the video to be decreased by a range of about 10 - 50 times (thereby reducing an amount of storage needed to save the video). Additionally, videos at different resolutions may be generated on-demand, which may reduce a required network bandwidth. Moreover, in some embodiments, by utilizing synthetic images to generate an on-demand video, processing may primarily be done on a graphics processing unit (GPU) rather than on a CPU.

[0123] Figure 13 is a block diagram of a system for generating a reconstructed video in accordance with some embodiments. As illustrates, a model 1302, such as a GAN or a deep learning network may be trained to generate a reconstructed video 1304. The reconstructed video 1304 may be generated responsive to a user request. The user request may specify the resolution, a portion of a fabrication apparatus the video is to depict, or the like. Model 1302 may take, as input, key frames 1306, data 1308, tool information 1310, process information 1312, and / or historic data 1314. Key frames 1306 may be real image data obtained by one or more camera devices. There may be within a range of 5-50 keyframes, within a range of about 5-20 keyframes, or the like. Data 1308 may include data extracted from different use cases of the fabrication apparatus, and may include, e.g., data from various sensors of the fabrication apparatus (e.g., temperature sensors, spectral sensors, voltage / current sensors, analysis of data from the camera, or the like). The data may be obtained from a database. Tool information 1310 may include design information associated with the fabrication apparatus. For example, tool information 1310 may include CAD design information associated with the apparatus. Process information 1312 may include information relating to a process currently being performed (e.g., at the time the video is requested) by the apparatus. The information may include gas flow information, temperature information, plasma characteristic information, chemistry species information, or the like. Processinformation 1312 may include information indicating a current recipe step being performed and / or parameters associated with the current recipe step. Historic data 1314 may be obtained from a database. In some embodiments, historic data 1314 may include process conditions from prior runs, results of analysis (e.g., from one or more cameras and / or one or more other sensors) from prior runs, camera settings, tool configuration, etc.

[0124] In some embodiments, a model may be trained to generate a synthetic image that is a three-dimensional reconstruction of a fabrication apparatus or a portion of the apparatus. The model may generate the three-dimensional reconstruction using one or more real images (e.g., obtained from one or more cameras disposed in or on various locations of the apparatus) and / or one or more synthetic images (e.g., synthetic images depicting portions of the apparatus from different viewpoints). For example, the model may be trained to stitch together a set of two- dimensional images (which may be real and / or synthetic) provided as input to the model to generate the three-dimensional reconstruction as an output. A three-dimensional reconstruction may be used for any number of purposes, such as allowing a digital immersive experience using a virtual reality or augmented reality headset, allowing depiction of hard to image anomalies or components (e.g., a parasitic plasma that is outside a line of sight of a single camera, but that may be reconstructed based on multiple images, or the like), etc.

[0125] Turning to Figure 14, a top view of a fabrication apparatus 1402 is shown with multiple cameras disposed around apparatus 1402, such as camera 1404. These cameras may generate a set of images, such as image 1406, where each may depict a view of apparatus 1402 from a different orientation or viewpoint. A trained model may take the images as an input, and generate, as an output, a three-dimensional reconstruction 1408. Note that although Figure 14 shows the set of images taken as an input by the model being real images, in some embodiments, the model may take any combination of real images and / or synthetic images as an input. For example, the model may take one or more real images as an input, and additionally one or more synthetic images, which may be generated based on the one or more real images to, e.g., depict a portion of the apparatus at which a camera is not disposed.

[0126] In some embodiments, synthetic images may be used by a machine learning model to identify modified camera settings and / or modified lighting settings. For example, synthetic images may be generated using different camera settings (e.g., different focuses, different lens properties, etc.), and / or different lighting settings (e.g., lighting devices producing light of different intensities and / or positioned in different locations). The trained model may generate, as an output, modifications to one or more physical components, such as one or more physical cameradevices and / or one or more physical lighting devices. By utilizing a machine learning model to determined modified settings of physical components, the model may be able to compensate for drift in system components over time, or the like.

[0127] Figure 15 illustrates the effects of modifying camera settings over time, e.g., responsive to output from a trained machine learning model. As illustrated in plot 1502, when using fixed camera settings, there may be an intensity change over time due to drift in the overall system. In particular, due to film deposits, the intensity may be reduced over time. As illustrated in plot 1504, by modifying the camera settings over time, the intensity may be controlled over time, resulting in a constant intensity regardless of system drift. For example, various camera settings such as white balancing, exposure time, camera gain, etc. may be modified to maintain a constant intensity regardless of film deposits over time (or other system drifts).

[0128] Note that, a model trained on synthetic data representing images associated with different camera settings, may be configured to generate, as an output, the camera settings used to capture a given real image. For example, the model may predict a shutter speed, a white balance setting, a camera gain, or the like. Based on the predicted camera settings associated with the real image analyzed at inference time, an algorithm may then be able to suggest camera setting modifications to, e.g., improve real images captured by the camera.CONTEXT FOR DISCLOSED COMPUTATIONAL EMBODIMENTS

[0129] Certain embodiments disclosed herein relate to computational systems for generation and use of synthetic images.

[0130] Many types of computing systems having any of various computer architectures may be employed as the disclosed systems for implementing algorithms as described herein. For example, the systems may include software components executing on one or more general purpose processors or specially designed processors such as Application Specific Integrated Circuits (ASICs) or programmable logic devices (e.g., Field Programmable Gate Arrays (FPGAs)). Further, the systems may be implemented on a single device or distributed across multiple devices. The functions of the computational elements may be merged into one another or further split into multiple sub-modules.

[0131] In some embodiments, code executed during generation or execution of a technique described herein on an appropriately programmed system can be embodied in the form of software elements which can be stored in a nonvolatile storage medium (such as optical disk, flash storagedevice, mobile hard disk, etc.), including a number of instructions for making a computer device (such as personal computers, servers, network equipment, etc.).

[0132] At one level a software element is implemented as a set of commands prepared by the programmer / developer. However, the module software that can be executed by the computer hardware is executable code committed to memory using “machine codes” selected from the specific machine language instruction set, or “native instructions,” designed into the hardware processor. The machine language instruction set, or native instruction set, is known to, and essentially built into, the hardware processor(s). This is the “language” by which the system and application software communicates with the hardware processors. Each native instruction is a discrete code that is recognized by the processing architecture and that can specify particular registers for arithmetic, addressing, or control functions; particular memory locations or offsets; and particular addressing modes used to interpret operands. More complex operations are built up by combining these simple native instructions, which are executed sequentially, or as otherwise directed by control flow instructions.

[0133] The inter-relationship between the executable software instructions and the hardware processor is structural. In other words, the instructions per se are a series of symbols or numeric values. They do not intrinsically convey any information. It is the processor, which by design was preconfigured to interpret the symbols / numeric values, which imparts meaning to the instructions.

[0134] The methods and techniques used herein may be configured to execute on a single machine at a single location, on multiple machines at a single location, or on multiple machines at multiple locations. When multiple machines are employed, the individual machines may be tailored for their particular tasks. For example, operations requiring large blocks of code and / or significant processing capacity may be implemented on large and / or stationary machines.

[0135] In addition, certain embodiments relate to tangible and / or non-transitory computer readable media or computer program products that include program instructions and / or data (including data structures) for performing various computer-implemented operations. Examples of computer-readable media include, but are not limited to, semiconductor memory devices, phase-change devices, magnetic media such as disk drives, magnetic tape, optical media such as CDs, magneto-optical media, and hardware devices that are specially configured to store and perform program instructions, such as read-only memory devices (ROM) and random access memory (RAM). The computer readable media may be directly controlled by an end user or the media may be indirectly controlled by the end user. Examples of directly controlled media includethe media located at a user facility and / or media that are not shared with other entities. Examples of indirectly controlled media include media that is indirectly accessible to the user via an external network and / or via a service providing shared resources such as the “cloud.” Examples of program instructions include both machine code, such as produced by a compiler, and files containing higher level code that may be executed by the computer using an interpreter.

[0136] In various embodiments, the data or information employed in the disclosed methods and apparatus is provided in an electronic format. Such data or information may include various coefficients to be used in calculations, and the like. As used herein, data or other information provided in electronic format is available for storage on a machine and transmission between machines. Conventionally, data in electronic format is provided digitally and may be stored as bits and / or bytes in various data structures, lists, databases, etc. The data may be embodied electronically, optically, etc.

[0137] System software typically interfaces with computer hardware and associated memory. In some embodiments, the system software includes operating system software and / or firmware, as well as any middleware and drivers installed in the system. The system software provides basic non-task-specific functions of the computer. In contrast, the modules and other application software are used to accomplish specific tasks. Each native instruction for a module is stored in a memory device and is represented by a numeric value.

[0138] Figure 16 is a block diagram of an example of the computing device 1600 suitable for use in implementing some embodiments of the present disclosure. For example, device 1600 may be suitable for implementing some or all functions of image generation logic, image analysis logic, machine learning training and / or inference logic, etc. disclosed herein.

[0139] Computing device 1600 may include a bus 1602 that directly or indirectly couples the following devices: memory 1604, one or more central processing units (CPUs) 1606, one or more graphics processing units (GPUs) 1608, a communication interface 1610, input / output (VO) ports 1612, input / output components 1614, a power supply 1616, and one or more presentation components 1618 (e.g., display(s)). In addition to CPU 1606 and GPU 1608, computing device 1600 may include additional logic devices that are not shown in Figure 16, such as but not limited to an image signal processor (ISP), a digital signal processor (DSP), an ASIC, an FPGA, or the like.

[0140] Although the various blocks of Figure 16 are shown as connected via the bus 1602 with lines, this is not intended to be limiting and is for clarity only. For example, in some embodiments,a presentation component 1618, such as a display device, may be considered an I / O component 1614 (e.g., if the display is a touch screen). As another example, CPUs 1606 and / or GPUs 1608 may include memory (e.g., the memory 1604 may be representative of a storage device in addition to the memory of the GPUs 1608, the CPUs 1606, and / or other components). In other words, the computing device of Figure 16 is merely illustrative. Distinction is not made between such categories as “workstation,” “server,” “laptop,” “desktop,” “tablet,” “client device,” “mobile device,” “hand-held device,” “electronic control unit (ECU),” “virtual reality system,” and / or other device or system types, as all are contemplated within the scope of the computing device of Figure 16.

[0141] Bus 1602 may represent one or more busses, such as an address bus, a data bus, a control bus, or a combination thereof. The bus 1602 may include one or more bus types, such as an industry standard architecture (ISA) bus, an extended industry standard architecture (EISA) bus, a video electronics standards association (VESA) bus, a peripheral component interconnect (PCI) bus, a peripheral component interconnect express (PCIe) bus, and / or another type of bus.

[0142] Memory 1604 may include any of a variety of computer-readable media. The computer-readable media may be any available media that can be accessed by the computing device 1600. The computer-readable media may include both volatile and nonvolatile media, and removable and non-removable media. By way of example, and not limitation, the computer- readable media may comprise computer- storage media and / or communication media.

[0143] The computer- storage media may include both volatile and nonvolatile media and / or removable and non-removable media implemented in any method or technology for storage of information such as computer-readable instructions, data structures, program modules, and / or other data types. For example, memory 1604 may store computer-readable instructions (e.g., that represent a program(s) and / or a program element(s), such as an operating system. Computerstorage media may include, but is not limited to, RAM, ROM, EEPROM, flash memory or other memory technology, CD-ROM, digital versatile disks (DVD) or other optical disk storage, magnetic cassettes, magnetic tape, magnetic disk storage or other magnetic storage devices, or any other medium which can be used to store the desired information and which can be accessed by computing device 1600. As used herein, computer storage media does not comprise signals per se.

[0144] The communication media may embody computer-readable instructions, data structures, program modules, and / or other data types in a modulated data signal such as a carrier wave or other transport mechanism and includes any information delivery media. The term“modulated data signal” may refer to a signal that has one or more of its characteristics set or changed in such a manner as to encode information in the signal. By way of example, and not limitation, the communication media may include wired media such as a wired network or direct- wired connection, and wireless media such as acoustic, RF, infrared and other wireless media. Combinations of any of the above should also be included within the scope of computer-readable media.

[0145] CPU(s) 1606 may be configured to execute the computer-readable instructions to control one or more components of the computing device 1600 to perform one or more of the methods and / or processes described herein. CPU(s) 1606 may each include one or more cores (e.g., one, two, four, eight, twenty-eight, seventy-two, etc.) that are capable of handling a multitude of software threads simultaneously. CPU(s) 1606 may include any type of processor and may include different types of processors depending on the type of computing device 1600 implemented (e.g., processors with fewer cores for mobile devices and processors with more cores for servers). For example, depending on the type of computing device 1600, the processor may be an ARM processor implemented using Reduced Instruction Set Computing (RISC) or an x86 processor implemented using Complex Instruction Set Computing (CISC). Computing device 1600 may include one or more CPUs 1606 in addition to one or more microprocessors or supplementary co-processors, such as math co-processors.

[0146] GPU(s) 1608 may be used by computing device 1600 to render graphics (e.g., 3D graphics). GPU(s) 1608 may include many (e.g., tens, hundreds, or thousands) of cores that are capable of handling many software threads simultaneously. GPU(s) 1608 may generate pixel data for output images in response to rendering commands (e.g., rendering commands from CPU(s) 1606 received via a host interface). GPU(s) 1608 may include graphics memory, such as display memory, for storing pixel data. The display memory may be included as part of memory 1604. GPU(s) 1608 may include two or more GPUs operating in parallel (e.g., via a link). When combined, each GPU 1608 can generate pixel data for different portions of an output image or for different output images (e.g., a first GPU for a first image and a second GPU for a second image). Each GPU can include its own memory or can share memory with other GPUs.

[0147] In examples where the computing device 1600 does not include the GPU(s) 1608, the CPU(s) 1606 may be used to render graphics.

[0148] Communication interface 1610 may include one or more receivers, transmitters, and / or transceivers that enable computing device 1600 to communicate with other computing devices viaan electronic communication network, included wired and / or wireless communications. Communication interface 1610 may include components and functionality to enable communication over any of a number of different networks, such as wireless networks (e.g., WiFi, Z-Wave, Bluetooth, Bluetooth LE, ZigBee, etc.), wired networks (e.g., communicating over Ethernet), low-power wide-area networks (e.g., LoRaWAN, SigFox, etc.), and / or the internet.

[0149] I / O ports 1612 may enable the computing device 1600 to be logically coupled to other devices including VO components 1614, presentation component(s) 1618, and / or other components, some of which may be built in to (e.g., integrated in) computing device 1600. Illustrative I / O components 1614 include a microphone, mouse, keyboard, joystick, track pad, satellite dish, scanner, printer, wireless device, etc. I / O components 1614 may provide a natural user interface (NUI) that processes air gestures, voice, or other physiological inputs generated by a user. In some instances, inputs may be transmitted to an appropriate network element for further processing. An NUI may implement any combination of speech recognition, stylus recognition, facial recognition, biometric recognition, gesture recognition both on screen and adjacent to the screen, air gestures, head and eye tracking, and touch recognition (as described in more detail below) associated with a display of computing device 1600. Computing device 1600 may be include depth cameras, such as stereoscopic camera systems, infrared camera systems, RGB camera systems, touchscreen technology, and combinations of these, for gesture detection and recognition. Additionally, computing device 1600 may include accelerometers or gyroscopes (e.g., as part of an inertia measurement unit (IMU)) that enable detection of motion. In some examples, the output of the accelerometers or gyroscopes may be used by computing device 1600 to render immersive augmented reality or virtual reality.

[0150] Power supply 1616 may include a hard-wired power supply, a battery power supply, or a combination thereof. Power supply 1616 may provide power to computing device 1600 to enable the components of computing device 1600 to operate.

[0151] Presentation component(s) 1618 may include a display (e.g., a monitor, a touch screen, a television screen, a heads-up-display (HUD), other display types, or a combination thereof), speakers, and / or other presentation components. Presentation component(s) 1618 may receive data from other components (e.g., GPU(s) 1608, CPU(s) 1606, etc.), and output the data (e.g., as an image, video, sound, etc.).

[0152] The disclosure may be described in the general context of computer code or machine- useable instructions, including computer-executable instructions such as program modules, beingexecuted by a computer or other machine, such as a personal data assistant or other handheld device. Generally, program modules including routines, programs, objects, components, data structures, etc., refer to code that perform particular tasks or implement particular abstract data types. The disclosure may be practiced in a variety of system configurations, including hand-held devices, consumer electronics, general-purpose computers, more specialty computing devices, etc. The disclosure may also be practiced in distributed computing environments where tasks are performed by remote-processing devices that are linked through a communications network.CONCLUSION

[0153] In the description, numerous specific details were set forth in order to provide a thorough understanding of the presented embodiments. The disclosed embodiments may be practiced without some or all of these specific details. In other instances, well-known process operations were not described in detail to not unnecessarily obscure the disclosed embodiments. While the disclosed embodiments were described in conjunction with the specific embodiments, it will be understood that the specific embodiments are not intended to limit the disclosed embodiments.

[0154] Unless otherwise indicated, the method operations and device features disclosed herein involves techniques and apparatus commonly used in metrology, semiconductor device fabrication technology, software design and programming, and statistics, which are within the skill of the art.

[0155] Unless defined otherwise herein, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art. Various scientific dictionaries that include the terms included herein are well known and available to those in the art. Although any methods and materials similar or equivalent to those described herein find use in the practice or testing of the embodiments disclosed herein, some methods and materials are described.

[0156] Numeric ranges are inclusive of the numbers defining the range. It is intended that every maximum numerical limitation given throughout this specification includes every lower numerical limitation, as if such lower numerical limitations were expressly written herein. Every minimum numerical limitation given throughout this specification will include every higher numerical limitation, as if such higher numerical limitations were expressly written herein. Every numerical range given throughout this specification will include every narrower numerical range that falls within such broader numerical range, as if such narrower numerical ranges were all expressly written herein.

[0157] The headings provided herein are not intended to limit the disclosure.

[0158] As used herein, the singular terms “a,” “an,” and “the” include the plural reference unless the context clearly indicates otherwise. The term “or” as used herein, refers to a nonexclusive or, unless otherwise indicated.

[0159] Various computational elements including processors, memory, instructions, routines, models, or other components may be described or claimed as “configured to” perform a task or tasks. In such contexts, the phrase “configured to” is used to connote structure by indicating that the component includes structure (e.g., stored instructions, circuitry, etc.) that performs the task or tasks during operation. As such, the unit / circuit / component can be said to be configured to perform the task even when the specified component is not necessarily currently operational (e.g., is not on).

[0160] The components used with the “configured to” language may refer to hardware — for example, circuits, memory storing program instructions executable to implement the operation, etc. Additionally, “configured to” can refer to generic structure (e.g., generic circuitry) that is manipulated by software and / or firmware (e.g., an FPGA or a general-purpose processor executing software) to operate in manner that is capable of performing the recited task(s). Additionally, “configured to” can refer to one or more memories or memory elements storing computer executable instructions for performing the recited task(s). Such memory elements may include memory on a computer chip having processing logic. In some contexts, “configured to” may also include adapting a manufacturing process (e.g., a semiconductor fabrication facility) to fabricate devices (e.g., integrated circuits) that are adapted to implement or perform one or more tasks.

Claims

CLAIMS1. A method of using synthetic images in connection with an integrated circuit fabrication chamber, the method comprising: obtaining a set of images associated with the integrated circuit fabrication chamber, wherein the set of images comprises at least a subset of synthetic images, wherein: each synthetic image of the subset of synthetic images is a photo-realistic synthetic image of an image captured by a camera disposed in or on a portion of the fabrication chamber, at least a portion of the subset of synthetic images represent at least one anomalous condition associated with the fabrication chamber, and each image in the set of images represents at least one of: 1) components of the fabrication chamber; 2) a process occurring in the fabrication chamber; or 3) wafer characteristics of a wafer being processed in the fabrication chamber; and training a machine learning model using the set of images, wherein the trained machine learning model is usable to predict the at least one anomalous condition represented in the at least the portion of the subset of synthetic images during operation of the fabrication chamber.

2. The method of claim 1, wherein the camera is configured to operate in the ultraviolet, visible, and / or infrared range.

3. The method of claim 1, wherein the wafer characteristics comprise film growth during a deposition process.

4. The method of claim 1, wherein the trained machine learning model takes chamber features as an input and generates an output usable for determining a time point at which to initiate a cleaning cycle.

5. The method of claim 4, wherein the chamber features comprise features associated with at least one of: a focus ring, a showerhead face, or any combination thereof.

6. The method of claim 5, wherein the at least one anomalous condition comprises a tilt of the pedestal and / or uneven heights of the one or more lift pins7. The method of claim 1, wherein the at least one anomalous condition comprises a hollow cathode discharge.

8. The method of claim 7, wherein an output of the trained machine learning model is usable to determine depth information associated with the hollow cathode discharge.

9. The method of claim 1, wherein the set of images represents the process occurring in the fabrication chamber, and wherein the process is a plasma-enhanced process.

10. The method of claim 9, wherein the at least one anomalous condition comprises parasitic plasma within the fabrication chamber and / or plasma non-uniformity that exceeds a uniformity threshold.

11. The method of claim 1, wherein the at least one anomalous condition comprises an indication of liquid droplets on the wafer being processed.

12. The method of claim 1, wherein the at least one anomalous condition comprises an indication of a liquid phenomena occurring within a portion of the fabrication chamber, wherein the liquid phenomena includes at least one of: bubbling, frothing, or change of liquid color.

13. The method of claim 1, wherein the wafer characteristics of the wafer comprise a degree to which the wafer is centered on a pedestal, a wafer rotational angle, and / or a wafer angular velocity.

14. The method of claim 1, wherein the trained machine learning model is configured to predict the at least one anomalous condition of the fabrication chamber when at least one component of the fabrication chamber is operating at a temperature above at least about 100 degrees Celsius.

15. The method of claim 1, wherein the trained machine learning model is configured to segment real images based on the set of images used to train the machine learning model, and wherein the segmented real images comprise an indication of at least one of: a wafer, a pedestal, a lift pin, a robotic arm, or a showerhead.

16. A method of using synthetic images in connection with an integrated circuit fabrication chamber, the method comprising: receiving image data from one or more cameras disposed in or on a portion of the fabrication chamber;providing the image data to a trained machine learning model to generate an output, wherein the trained machine learning model was trained using a set of image comprising at least a subset of synthetic images, and wherein each synthetic image of the subset of synthetic images is a photo-realistic synthetic image of an image captured by the one or more cameras; and utilizing the output to predict at least one anomalous condition associated with the fabrication chamber.

17. The method of claim 16, further comprising taking at least one action based on the prediction of the at least one anomalous condition.

18. The method of claim 16, wherein the at least one action comprises causing an indication of the at least one anomalous condition to be presented on a console associated with the fabrication chamber.

19. A method, comprising: generating a training set, the training set comprising at least a subset of synthetic images associated with a fabrication chamber, each synthetic image being a photorealistic image of an image captured by at least one camera disposed in or on a portion of the fabrication chamber; and utilizing the training set to train a machine learning model, an output of the machine learning model being usable to perform at least one of: 1) generation of one or more enhanced images depicting at least a portion of the fabrication chamber; or 2) determination of one or more modified camera settings and / or lighting settings for use in future operation of the at least one camera disposed in or on the portion of the fabrication chamber.

20. A calibration disc for calibrating wafer rotation, the calibration disc comprising: a top surface; a bottom surface; and a side edge that joins the top surface and the bottom surface, wherein the side edge comprises a plurality of pattern markings, and wherein a snapshot of a subset of the pattern markings is usable to determine an angular rotation or angular velocity of the calibration disc.

Images