Generation of digital twins of semiconductor manufacturing equipment

Abstract

Various embodiments herein relate to systems, methods, and media for generating a digital twin of a semiconductor manufacturing equipment. In some embodiments, a digital twin of a process chamber of the semiconductor manufacturing equipment is provided, comprising one or more non-transitory machine-readable media comprising logic configured to implement a first model of a first location of the process chamber and a second model of a second location of the process chamber, the first model being coupled to a second model, the first model and the second model each being a model type that is one of: 1) an AI / ML model, 2) an HFS model, and 3) a closed form solution, and the first model and the second model each representing a class of physical phenomena that is one of: 1) thermal properties, 2) plasma properties, 3) fluid dynamics, 4) structural properties, and 5) chemical reactions.

Description

Incorporation by Reference

[0001] As part of the present application, a PCT application request is filed concurrently with the present specification. Each application that is specified in this concurrently filed PCT application request and from which the present application claims benefit or priority is hereby incorporated by reference in its entirety into the present specification for all purposes. Background Art

[0002] It may be useful to provide a model of an entire process chamber used for manufacturing electronic devices such as semiconductor integrated circuits. For example, such a model can be used to evaluate manufacturing recipes, designs of process chamber components, and the like. However, since a typical process chamber involves many different physical phenomena that interact in complex ways (e.g., fluid dynamics, temperature and temperature flux, plasma behavior, chemical reactions, structural characteristics, etc.), providing a model of the entire process chamber can be difficult. Furthermore, models of different components of a reactor or process chamber may require very different temporal or spatial scales for accuracy, thus making it difficult to combine models of different components.

[0003] The background description provided herein is for the purpose of generally presenting the context of the present disclosure. To the extent described in this Background section, the work of the presently named inventors, as well as aspects of the description that may not otherwise be considered prior art as of the filing date of the present application, are not admitted, either expressly or implicitly, to be prior art against the present disclosure. Summary of the Invention

[0004] Methods, systems, and media for generating digital twins of semiconductor manufacturing equipment are disclosed herein.

[0005] According to some embodiments of the disclosed subject matter, a digital twin of a process chamber of semiconductor manufacturing equipment is provided, comprising one or more non-temporary machine-readable media, the machine-readable media comprising logic configured to implement a first model of a first location of the process chamber and a second model of a second location of the process chamber, the first model of the first location of the process chamber coupled to the second model of the second location of the process chamber, the first model of the first location of the process chamber and the second model of the second location of the process chamber being model types which are one of 1) AI / ML models, 2) HFS models, and 3) closed-form solutions, and the first model of the first location of the process chamber and the second model of the second location of the process chamber being a class of physical phenomena which are one of 1) thermal properties, 2) plasma properties, 3) hydrodynamics, 4) structural properties, and 5) chemical reactions.

[0006] In some embodiments, the first model of the first location in the process chamber is of a different model type from the second model of the second location in the process chamber.

[0007] In some embodiments, the first model of the first location in the process chamber represents a different class of physical phenomena than the second model of the second location in the process chamber.

[0008] In some embodiments, the first location is one of the following: 1) the base of the ESC, 2) the showerhead, 3) the gap between the base and the showerhead, 4) the chamber wall, and 5) the surface of the wafer fabricated by the process chamber.

[0009] In some embodiments, a first model of a first location in a process chamber, coupled to a second model of a second location in a process chamber, comprises a first model of a first location in a process chamber that provides the output used by the second model of the second location in a process chamber to the second model of the second location in a process chamber.

[0010] In some embodiments, a first model of a first location in a process chamber, coupled to a second model of a second location in a process chamber, comprises a first model of a first location in a process chamber that receives outputs used by the first model of the first location in a process chamber from the second model of the second location in a process chamber.

[0011] According to some embodiments of the disclosed subject matter, a computer program product is provided for generating a digital twin of a process chamber, the computer program product comprising a non-temporal computer-readable medium comprising computer-executable instructions for generating a digital twin by comprising the trained AI / ML model of the process chamber and the model of the process chamber, the computer program product generating a plurality of HFS values ​​for a first location of the process chamber using a high-fidelity simulation (HFS) model of the first location of the process chamber, receiving a plurality of sensor measurements corresponding to the first location of the process chamber, training an artificial intelligence / machine learning (AI / ML) model of the process chamber using the plurality of HFS values ​​and at least one of the plurality of sensor measurements, and combining the trained AI / ML model of the process chamber and the model of the process chamber, the digital twin of the process chamber comprising a computer-executable medium comprising computer-readable instructions for generating a digital twin by comprising the trained AI / ML model of the process chamber and the model of the process chamber

[0012] In some embodiments, the second model of the second location of the process chamber is one of 1) the AI / ML model, 2) the HFS model, and 3) the closed-form solution.

[0013] In some embodiments, both the HFS model and the AI / ML model for the first location of the process chamber model the same class of physical phenomena.

[0014] In some embodiments, a trained AI / ML model at a first location in the process chamber and a model at a second location in the process chamber each model a certain class of physical phenomena.

[0015] In some embodiments, the class of physical phenomena is one of thermal properties, plasma properties, fluid dynamics, structural properties, and chemical reactions.

[0016] In some embodiments, the trained AI / ML model for a first location in the process chamber and the model for a second location in the process chamber model different classes of physical phenomena.

[0017] In some embodiments, the HFS model of the first location in the process chamber generates simulation values ​​at a shorter time step than the AI / ML model of the first location in the process chamber.

[0018] In some embodiments, the first location of the process chamber is one of the following: 1) the base of the electrostatic chuck (ESC), 2) the shower head, 3) the gap between the shower head and the base, 4) the chamber wall, and 5) the surface of the wafer fabricated by the process chamber.

[0019] In some embodiments, combining a trained AI / ML model of a first location in a process chamber with a model of a second location in a process chamber involves providing multiple outputs of the trained AI / ML model of the first location in the process chamber to the model of the second location in the process chamber.

[0020] In some embodiments, providing multiple outputs of a trained AI / ML model at a first location in the process chamber to a model at a second location in the process chamber comprises waiting until multiple outputs of the trained AI / ML model at the first location in the process chamber are received, and then sending the multiple outputs to the model at the second location in the process chamber.

[0021] In some embodiments, combining a trained AI / ML model of a first location in a process chamber with a model of a second location in a process chamber involves providing multiple outputs of the model of the second location in a process chamber to the trained AI / ML model of the first location in a process chamber.

[0022] In some embodiments, the computer program product includes a trained AI / ML model of a first location in the process chamber in a digital twin, and then further comprises computer-executable instructions for verifying the performance of the trained AI / ML model of the first location in the process chamber.

[0023] In some embodiments, verifying the performance of a trained AI / ML model involves generating simulated data using a digital twin that includes the trained AI / ML model at a first location of the process chamber and the model at a second location of the process chamber, and comparing the simulated data with experimental data collected using multiple sensors associated with the physical process chamber.

[0024] In some embodiments, the model of the second location in the process chamber is an HFS model, and further comprises computer-executable instructions for replacing the HFS model of the second location in the process chamber with a trained AI / ML model of the second location in a digital twin.

[0025] According to some embodiments of the disclosed subject matter, there is provided a computer program product for using a digital twin of a process chamber, the computer program product comprising a non-transitory computer-readable medium provided with computer-executable instructions for: identifying a plurality of inputs to the digital twin of the process chamber, wherein the digital twin comprises a first model of a first location of the process chamber and a second model of a second location of the process chamber, the first model of the first location of the process chamber and the second model of the second location of the process chamber are coupled, and the plurality of inputs represent operating conditions of the process chamber; providing the plurality of inputs to the digital twin; and using the digital twin to generate predicted wafer characteristics of a simulated wafer.

[0026] In some embodiments, the first model of the first location of the process chamber includes specifications of components of the process chamber, and the computer-executable instructions further comprise computer-executable instructions for verifying the specifications of the components based on the predicted wafer characteristics.

[0027] In some embodiments, the plurality of inputs include parameters of a recipe implemented by the process chamber, and the computer-executable instructions further comprise computer-executable instructions for verifying at least one parameter of the recipe based on the predicted wafer characteristics.

[0028] In some embodiments, the predicted wafer characteristics comprise an indicator of defects in the simulated wafer.

[0029] In some embodiments, the computer program product further comprises computer-executable instructions for identifying a recommendation to modify at least one of the operating conditions based on the predicted wafer characteristics.

[0030] In some embodiments, the recommendation is identified in response to a determination that the predicted wafer characteristics indicate defects in the simulated wafer.

[0031] In some embodiments, the recommendation is identified in response to a determination that at least one of the first and second models has generated a value indicating abnormal operating conditions for the process chamber. [Brief explanation of the drawing]

[0032] [Figure 1] Figure 1 is a schematic diagram of a digital twin of a process chamber according to several embodiments of the disclosed subject.

[0033] [Figure 2] Figure 2 is a block diagram for training artificial intelligence / machine learning (AI / ML) models according to several embodiments of the disclosed subject matter.

[0034] [Figure 3] Figure 3 is a block diagram of a combined digital twin model according to several embodiments of the disclosed subject.

[0035] [Figure 4A] Figure 4A is a diagram illustrating the operation of a processor for generating a digital twin and the operation of a processor for using a digital twin, respectively, according to several embodiments of the disclosed subject matter. [Figure 4B] Figure 4B is a diagram illustrating the operation of a processor for generating a digital twin and the operation of a processor for using a digital twin, respectively, according to several embodiments of the disclosed subject matter.

[0036] [Figure 5] Figure 5 is a diagram illustrating an exemplary computer system that may be used to implement a particular embodiment described herein. [Modes for carrying out the invention]

[0037] Technical terms The following terms are used throughout this specification:

[0038] The terms “semiconductor wafer,” “wafer,” “substrate,” “wafer substrate,” and “partially fabricated integrated circuit” may be used interchangeably. Those skilled in the art will understand that the term “partially fabricated integrated circuit” can refer to a semiconductor wafer at any of the many stages of integrated circuit fabrication. Wafers or substrates used in the semiconductor device industry typically have a diameter of 200 mm, 300 mm, or 450 mm. In addition to semiconductor wafers, other workpieces from which the disclosed embodiments can be utilized include a variety of articles such as printed circuit boards, magnetic recording media, magnetic recording sensors, mirrors, optical elements, display devices, or components such as backplanes and micromechanical devices for pixelated display devices. Workpieces may be of various shapes, sizes, and materials.

[0039] As used herein, “semiconductor device fabrication operation” refers to an operation performed during the fabrication of a semiconductor device. Typically, the entire fabrication process comprises multiple semiconductor device fabrication operations, each performed with its own unique semiconductor fabrication tools, such as plasma reactors, electroplating cells, chemical mechanical planarization tools, and wet etching tools. Categories of semiconductor device fabrication operations include subtractive processes such as etching and planarization processes, and material addition processes such as deposition processes (e.g., physical vapor deposition, chemical vapor deposition, atomic layer deposition, electrochemical deposition, electroless deposition). In the context of etching processes, a substrate etching process includes a process of etching a mask layer, or more generally, a process of etching any layer of material previously deposited on and / or present on the substrate surface. Such etching processes can etch stacks of layers on the substrate.

[0040] "Manufacturing equipment" refers to equipment on which a manufacturing process is carried out. Manufacturing equipment often has a process chamber in which a workpiece is located during processing. Typically, during use, manufacturing equipment performs one or more semiconductor device fabrication operations. Examples of manufacturing equipment for semiconductor device fabrication include deposition reactors such as electroplating cells, physical vapor deposition reactors, chemical vapor deposition reactors, and atomic layer deposition reactors, as well as subtractive process reactors such as dry etching reactors (e.g., chemical and / or physical etching reactors), wet etching reactors, and ashers.

[0041] As used herein, "artificial intelligence / machine learning (AI / ML) models" are trained computational algorithms that have been trained to construct computational models of relationships between data points. Trained AI / ML models are capable of generating outputs based on learned relationships without being explicitly programmed to generate outputs using explicitly defined relationships.

[0042] Examples of AI / ML models include autoencoder networks (e.g., Long-Term Short-Term Memory (LSTM) autoencoders, convolutional autoencoders, deep autoencoders, variational autoencoders, and / or any other suitable type of autoencoder network), neural networks (e.g., convolutional neural networks, deep convolutional networks, recurrent neural networks, and / or any other suitable type of neural network), clustering algorithms (e.g., nearest neighbor clustering, K-means clustering, and / or any other suitable type of clustering algorithm), and random forest models including deep random forests, restricted Boltzmann machines, deep belief networks (DBNs), recurrent tensor networks, regression, and gradient-boosted trees.

[0043] It should be noted that some AI / ML models are characterized as “deep learning” models. Unless otherwise specified, references to AI / ML in this specification include embodiments of deep learning. Deep learning models can be implemented in various forms, such as neural networks (e.g., convolutional neural networks). Generally, but not always, deep learning models contain multiple layers. Each such layer contains multiple processing nodes, and the layers process sequentially, with nodes in layers closer to the model input layer processing before nodes in layers closer to the model output layer. In various embodiments, some layers feed to the next layer, and so on.

[0044] In various embodiments, deep learning models can have considerable depth. In some embodiments, the model has three or more (or four or more, five or more, or six or more) layers of processing nodes that receive values ​​from previous layers (or as direct inputs) and output values ​​to subsequent layers (or final outputs). Internal nodes are often "hidden" in the sense that their input and output values ​​are not visible outside the model. In various embodiments, the operation of hidden nodes is not monitored or recorded during operation.

[0045] The nodes and connections of a deep learning model can be trained and retrained without having to redesign their number, arrangement, etc.

[0046] As shown, in various embodiments, many deep learning models have other structures and forms, although node layers can collectively form a neural network. In some cases, deep learning models do not have a layered structure, in which case the above characterization of "deep" as having many layers is not appropriate.

[0047] As used herein, “physical phenomena” refers to observable properties or conditions within a particular class. Examples of classes of physical phenomena may include plasma properties, thermal properties, mechanical or structural properties, chemical properties, and / or hydrodynamic properties.

[0048] A "high-fidelity simulation (HFS) model" refers to values ​​generated using a model or simulation that incorporates various physics-based equations. In an HFS model, the governing equations are derived from first principles of given physical phenomena, such as the conservation of mass and energy in a flow field, or the equilibrium of forces in a stress field. The equations can be solved simultaneously in their original partial differential equation (PDE) form by various numerical methods. HFS models can be used, for example, to predict the physical reality of test conditions by considering variables corresponding to those conditions. Because HFS models may be associated with numerous variables (all of which cannot be measured), HFS models are typically calibrated against a given test condition and then used to predict other test conditions. It should be noted that if an HFS model is properly calibrated, there should be little difference between the output of the HFS model and the actual test data.

[0049] The HFS model can use any suitable technique to model specific components or locations within a process chamber with respect to a particular class of physical phenomena, using explicitly defined physical laws or equations. For example, the HFS model can simulate the thermal properties of a specific component (e.g., within the base of an electrostatic chuck (ESC)) and / or a specific location within the process chamber (e.g., the gap between the showerhead and the base). As another example, the HFS model can simulate the structural properties of a specific component (e.g., the base of the ESC, one or more screws attaching the base to the base, etc.) and / or a specific location within the process chamber (e.g., a specific wall, etc.). In some embodiments, the HFS model can use numerical modeling techniques that generate simulations of physical phenomena over a series of time steps and / or spatial steps. Examples of techniques that can be used include finite element modeling, finite difference modeling, and finite volume modeling.

[0050] As used herein, "closed-form solution" refers to an equation, function, or set of equations or functions that describes a particular physical phenomenon. For example, closed-form solutions can be used to calculate flow through a pipe. Another example is the use of closed-form solutions to calculate flow on a flat plate.

[0051] As used herein, the “digital twin” of a process chamber or other type of digital equipment refers to a model of the entire process chamber. In some embodiments, the digital twin may consist of multiple models of different types, each representing a different class of physical phenomena and / or different locations within the process chamber. For example, the digital twin may include a structural model of the showerhead, a thermal model of the showerhead, a chemical model of the gap between the showerhead and the base, a computational fluid dynamics (CFD) model of the gap between the showerhead and the base, and so on. In some embodiments, each model constituting the digital twin may be one of 1) a closed-form solution, 2) an AI / ML model, and 3) an HFS model. In other words, the digital twin may consist of any combination of a closed-form solution, an AI / ML model, and / or an HFS model.

[0052] In some embodiments, the digital twin may include different models coupled to form the digital twin (e.g., models of different locations in the process chamber, models of different classes of physical phenomena, and / or different types of models). For example, the output of a first model (e.g., an HFS thermal model of the gap between the showerhead and the base) can be used as input to a second model (e.g., an AI / ML structural model of the process chamber wall).

[0053] A digital twin of a manufacturing device can be configured to output any of several types of information about the manufacturing device. Such information may include information about devices on substrates processed using the manufacturing device or devices partially fabricated using the device, information about one or more components of the manufacturing device (e.g., plasma generator, process gas inlet, substrate support, etc.), and / or information about process conditions encountered at one or more locations within the manufacturing device.

[0054] As used herein, “predicted wafer characteristics” can be the output of a digital twin of a process chamber or other manufacturing equipment. In particular, the predicted wafer characteristics may be any suitable characteristics of a simulated wafer fabricated using the digital twin under the operating conditions used as input to the digital twin.

[0055] In some embodiments, the predicted wafer properties may include “defects” in the simulated wafer. As used herein, “defect” is a deviation from the proper functioning of a process, layer, or product. A process defect is a deviation from the expected process that may cause a malfunction in the fabricated device or product. An example of a process defect is scum, which is residue left on the wafer after delamination from the photoresist. Another example is an undesirable bridge, which may cause a short circuit between elements in the device. Particle defects can be classified by properties such as composition, shape (or form), size, and location on the wafer. Defects on semiconductor substrates can often be caused by one or more sources in the substrate processing chamber. Process chamber components such as showerheads, chamber walls, seals, and windows may scatter material in the form of particles, which can result in wafer defects. In addition, some fabrication processes, such as etching processes, may leave redeposits or residues on the substrate, thereby causing defects. Furthermore, defects may be caused by the movement of material on the substrate, such as reflow of material during a thermal process, or by the unintended deposition of particles on the bottom or sides of the wafer that later move and redeposit on the top of the wafer.

[0056] In some embodiments, the predicted wafer properties may include indicators of “features” on the substrate. As used herein, “features” are non-planar structures on the substrate surface, typically surfaces modified during semiconductor device fabrication operations. Examples of features include trenches, vias, pads, pillars, and domes. Features may be created by photoresist development, mask demarcation, lithographic etching, lithographic deposition, epitaxial growth, damascene deposition, and the like. In some embodiments, the predicted wafer properties may include the aspect ratio of the features, the width dimensions of the features, and the like.

[0057] In some embodiments, the predicted wafer properties may include the geometric properties of the substrate. In some embodiments, the geometric properties may include a set of points in space representing the location of a feature or group of features, which may include etched features, deposited features, planarized features, etc. Examples of geometric properties include critical dimensions, pitch, depth, aspect ratio, and sidewall angles of a feature or group of features.

[0058] In some embodiments, the predicted wafer properties may include the optical or chemical properties of the substrate or specific features or layers on the substrate. Examples of optical or geometric properties of the substrate, features, or layers include the extinction coefficient, refractive index, chemical composition, and atomic composition.

[0059] As used herein, “combining” or “combined” refers to using the output of one model as input to another model, or vice versa. “Combining” may also refer to running two or more models in parallel and combining their outputs, or using them together when characterizing manufacturing equipment produced using processes carried out within manufacturing equipment such as process chambers or structures. Collectively, in some embodiments, combined models work together to implement a digital twin.

[0060] In some embodiments, an HFS model can be coupled to an AI / ML model, and vice versa. In some embodiments, the two models can be "continuously coupled" or "fully coupled." "Continuously coupled" refers to one-way communication from the first model to the second model. For example, the output of the first model can be used as an input to the second model. "Fully coupled" refers to two-way communication between the first and second models. For example, the output of the first model can be used as an input to the second model, and the output of the second model can be used as an input to the first model.

[0061] "Predictive maintenance" refers to monitoring and predicting the health of a manufacturing machine or its components based on the characteristics of the manufacturing machine and / or the components of the manufacturing machine. In some embodiments, the manufacturing machine may include a chamber system or subsystem, such as an ESC, showerhead, plasma source, radio frequency (RF) generator, and / or any other suitable type of manufacturing system or subsystem. In some embodiments, the components of the manufacturing machine may include individual components of a system and / or subsystem, such as a pedestal, edge ring of an ESC, specific valves (e.g., of a gas box supplying gas to a showerhead), and / or any other suitable components.

[0062] In some embodiments, digital twins can be used for predictive maintenance. For example, in some embodiments, a digital twin can be used to simulate a deployed and currently operating process chamber. Continuing this example, a digital twin can be used to simulate a deployed process chamber under various degradation conditions, such as the typical degradation rates of various components due to wear, or a deployed process chamber operating with a specific failed component. Further continuing this example, in some embodiments, the digital twin can then be used to generate any appropriate predictive maintenance metrics, such as the remaining useful life (RUL) of a particular component or the mean time to failure (MTTF) of a particular system or subsystem.

[0063] In some embodiments, the digital twin can identify any appropriate prescriptive maintenance recommendations, which may include, additionally or alternatively, one or more recommendations to extend the service life of components in the deployed process chamber. For example, in some embodiments, the digital twin can be used to identify recipe parameter changes that are likely to extend the service life of a particular component, such as an ESC base. As another example, in some embodiments, the digital twin can be used to identify components that can be replaced to extend the life of different components that are likely to fail.

[0064] overview Digital twins for process chambers or other manufacturing equipment are described herein. A digital twin may include multiple models that can be coupled together to form a digital twin. For example, the multiple models may include models of different systems, subsystems, components, and / or locations of a process chamber, such as the base of an electrostatic chuck (ESC), a showerhead, the gap between the showerhead and the base, the chamber walls, the surface of a simulated wafer fabricated by the process chamber, and specific pipes. In addition, each model may represent a particular class of physical phenomena such as thermal properties, plasma properties, structural properties, chemicals and chemical reactions, and / or fluid dynamics.

[0065] In some embodiments, the digital twin comprises multiple models, each characterized by at least the following characteristics: a) the type of computing tool, b) the class of physical phenomena to be predicted, and c) the location within the process chamber or other manufacturing equipment represented by the digital twin.

[0066] The type of computational tool describes the computational components and / or logical structure of operations that make up the model. In various embodiments, the model included in a digital twin may be one of the following types of computational tools: 1) AI / ML, 2) HFS, and 3) closed-form modeling. For example, HFS models can be used for locations and certain classes of physical phenomena where conditions change over relatively short time scales or small spatial scales. Specific examples where HFS models can be used include plasma models of the gap between a pedestal and a showerhead, and chemical models of wafer surfaces.

[0067] As another example, AI / ML models can be used for locations and certain classes of physical phenomena where conditions are relatively stable over time and / or spatial location. Specific examples of where AI / ML models can be used include thermal models of chamber walls. Yet another example is that for locations and certain classes of physical phenomena that may be too complex to simulate using HFS models, AI / ML models can be trained using training data generated by physical sensor measurements and / or calculations representing the physical conditions within manufacturing equipment in operation. A specific example is a thermal model simulating temperature flux or heat transfer within a ceramic pedestal containing an embedded mesh.

[0068] As yet another example, for the location and physical phenomena of certain equipment, the characteristics of the phenomena can be accurately and reliably represented by simple equations or other closed-form solutions. In such cases, closed-form solutions can be used instead of HFS or AI / ML representations of the physical phenomena, thereby saving computational resources for other more computationally intensive models required to predict more complex physical phenomena. Specific examples of physical phenomena for which closed-form solutions can be used include flow through pipes, flow on flat plates, and bending of plates.

[0069] By combining different model types for different classes of locations and physical phenomena within a digital twin, the digital twin can accurately simulate a process chamber while balancing the use of computational resources. For example, the digital twin can conserve computational resources by using a computationally intensive HFS model for the location and / or dynamically changing physical phenomena of the process chamber, while using a less intensive AI / ML model and / or closed-form solution, thereby modeling locations and / or physical phenomena that can be accurately modeled by the AI / ML model or closed-form solution.

[0070] A digital twin of a process chamber can be used for any appropriate application. For example, a digital twin can be used to evaluate or verify the design of a system, subsystem, or component, such as a new potential design. Another example is using a digital twin to evaluate or verify a recipe or process performed by a process chamber, such as a change in the process gas used or a change in setpoints. Yet another example is using a digital twin to perform predictive maintenance by simulating a currently deployed process chamber. In a specific example, by simulating a currently deployed process chamber, a digital twin can be used to identify potential future failures of the deployed process chamber. Furthermore, a digital twin can be used to identify recommendations that can mitigate potential future failures.

[0071] Digital twin of semiconductor manufacturing equipment Referring to Figure 1, an exemplary schematic diagram of a digital twin 100 of a process chamber according to several embodiments of the disclosed subject matter is shown.

[0072] The digital twin 100 can be a model of a substantial part of the process chamber or the entire process chamber. That is, in some embodiments, the digital twin 100 can include models of different systems or subsystems of the process chamber that encompass different classes of physical phenomena occurring within the process chamber.

[0073] In some embodiments, the digital twin 100 can take an input 102 and generate predicted substrate characteristics 104 as an output. In some embodiments, the digital twin 100 can generate information about one or more components of a process chamber and / or information about one or more process conditions that occur while processing a substrate within the process chamber.

[0074] In some embodiments, input 102 may include any appropriate parameter values ​​corresponding to the control of the process chamber modeled by the digital twin 100. For example, as shown in Figure 1, input 102 may include various process chamber settings, such as chamber pressure, coolant flow rate (or other heat flux control), gas type, gas species mixture composition, chemical properties, RF power, heater power, transformer-coupled plasma (TCP) settings, bias, transformer-coupled capacitive tuning (TCCT) circuit settings, and / or any other appropriate inputs. Note that the inputs shown in Figure 1 are illustrative only. In some embodiments, any of the inputs shown in Figure 1 may be omitted. Additionally or alternatively, in some embodiments, any other parameters not shown in Figure 1 may be included in input 102, such as inputs related to the conditions of the pre-treated substrate (e.g., stack, structure, and / or variability of the substrate), inputs related to a series of process steps (e.g., a series of process steps prior to the current step), inputs related to the system state, inputs related to hardware or software configuration, etc.

[0075] The predicted wafer characteristics 104 can be any suitable predicted characteristics of the simulated wafer fabricated by the process chamber represented by the digital twin 100. In some embodiments, the predicted characteristics may include information about the features of the simulated wafer, such as etched features, deposited features, and planarized features. For example, the feature information may include geometric information indicating the aspect ratio, width, height, etc. of the feature. In more specific examples, the geometric information may include etching depth, sidewall angle, etc. In some embodiments, the predicted characteristics may include information about defects in the simulated wafer, such as the location and type of the defect.

[0076] In some embodiments, the predicted wafer characteristics 104 can be used for any appropriate purpose such as 1) design verification, 2) process verification, and / or 3) predictive maintenance.

[0077] Design verification may involve verifying a specific system, subsystem, or component of a process chamber. For example, design verification may include verifying the structure of a system (e.g., showerhead, ESC, plasma source, RF generator, etc.), a subsystem (e.g., base of an ESC), or a component (e.g., edge ring of an ESC, valve of a gas box, etc.). In some embodiments, design verification can be used to verify potential new structures, such as a newly designed base of an ESC, potential modifications to a showerhead (e.g., including more holes, including holes of different sizes or patterns), and / or any other suitable potential new structures. For example, in some embodiments, design verification can be used to evaluate potential new structures for a system, subsystem, or component of a process chamber by evaluating the expected wafer properties 104 produced using the potential new structures.

[0078] Process verification can be a verification of the process or recipe used by the process chamber. For example, process verification may include verification of changes to the process or recipe currently used by the process chamber. In some embodiments, exemplary process or recipe changes may include changes in the applied temperature (e.g., the temperature applied to the base), changes in the process gas used (e.g., changes in the gas composition or gas mixing ratio), and changes in the plasma pulse profile. In some embodiments, process verification can be used to evaluate potential modifications to the process or recipe by evaluating the expected wafer characteristics 104 produced using the modified process or recipe.

[0079] Predictive maintenance may include identifying process chamber systems, subsystems, or components that are likely to fail, and / or periods during which such systems, subsystems, or components are likely to fail (e.g., periods during which the ESC base is likely to break or break within a certain timeframe, periods during which the showerhead is likely to fail within a certain timeframe, periods during which a particular valve is likely to fail within a certain timeframe). Additionally or alternatively, in some embodiments, predictive maintenance may include identifying possible causes of failure for specific systems, subsystems, or components of the process chamber. For example, a possible cause of failure for the ESC base may be identified as an abnormal temperature gradient or abnormal temperature flux. In some embodiments, possible failures and / or possible causes of failure may be identified based on values ​​in individual models of the digital twin 100. For example, a value in the thermal model of the ESC base may be identified as being outside the normal operating range. Continuing this example, the base may be identified as likely to fail within a certain time range due to abnormal values ​​in the thermal model of the base.

[0080] In some embodiments, predictive maintenance may include a prediction that a fabricated wafer is likely to contain defects due to, for example, a potential failure of a particular system, subsystem, or component of the process chamber. For example, the prediction may indicate that a fabricated wafer contains defects due to excess particles due to cracking, wear, or other failure of a showerhead or other component inside the process chamber. In some embodiments, such a prediction may be generated based on predicted wafer characteristics 104 that indicate defects under specific operating conditions of the digital twin 100.

[0081] In some embodiments, predictive maintenance may further include recommendations to mitigate the likelihood of failure of a system, subsystem, or component, and / or to prevent defects in fabricated wafers. For example, if the predicted wafer characteristics 104 indicate a potential for defects in the wafer, it may be possible to identify recommendations to modify the parameters of the recipe used to fabricate the wafer. More specifically, recommendations such as changing the operating temperature or gas flow rate may be identified and presented.

[0082] As shown in Figure 1, the digital twin 100 can include multiple models of different systems or subsystems of the process chamber represented by the digital twin 100. For example, the digital twin 100 can include models of the ESC base, showerhead, chamber wall, gap between the showerhead and base, and / or any other suitable system or subsystem. In some embodiments, each model may be one of 1) an AI / ML model, 2) an HFS model, and 3) a closed-form solution.

[0083] In some embodiments, each model included in the digital twin 100 may represent a particular class of physical phenomena. Exemplary classes of physical phenomena may include: 1) plasma properties within an operating device; 2) hydrodynamic properties within an operating device; 3) thermal properties of device components; 4) structural properties of device components; and 5) chemical properties or chemical reactions and / or unreacted chemical species within an operating device. Some or all of the models may represent only one or more regions within the manufacturing device. Such models may have geometric boundaries defining limited regions within the manufacturing device. In some cases, such boundaries are represented within the model by boundary conditions.

[0084] In some embodiments, plasma properties may include plasma characteristics such as plasma temperature, potential, density, composition (e.g., ion-to-electron ratio), and / or plasma properties. In some embodiments, plasma properties may be determined as a function of time based, for example, on the plasma pulse profile. In some embodiments, plasma properties may be particularly relevant to the location of the reactor, such as between the showerhead and base of the ESC, within a parasitic outer gap region, and / or any other suitable location.

[0085] In some embodiments, the hydrodynamic properties may include the flow from a gas inlet, such as a showerhead, and / or the fluid flow around a component (e.g., a showerhead, base, chamber wall, etc.).

[0086] In some embodiments, thermal properties may include the thermal properties of any gas, solid, and / or plasma within the reactor. For example, thermal properties may include thermal transfer or heat transfer within components such as a substrate base or process gas showerhead, or within open areas of the equipment such as a gap between the showerhead and the base.

[0087] In some embodiments, structural characteristics may include mechanical stresses, forces, and pressures at the boundaries between different components of the reactor and / or between components. For example, structural characteristics may include pressure on a particular valve, stress on the wafer during fabrication, and forces on the chamber wall.

[0088] In some embodiments, chemical properties may include reaction kinetics on the substrate surface and / or on equipment components such as chamber walls, showerheads, or substrate bases. In some embodiments, chemical properties may include equilibrium or non-equilibrium concentrations of one or more chemical species on the substrate surface and / or on equipment components such as chamber walls, showerheads, or substrate bases. In some embodiments, chemical properties may include mass transfer properties of one or more chemical species between substrate surfaces and / or on equipment components such as chamber walls, showerheads, or substrate bases.

[0089] As shown in Figure 1, the digital twin 100 can include individual models of different locations in the process chamber, each of which is a specific model type and represents a specific class of physical phenomena. For example, the digital twin 100 may include an HFS thermal model 106 of the pedestal, an HFS plasma model 108 of the gap between the showerhead and the pedestal, an AI / ML thermal model 110 of the chamber wall, a closed-form CFD model 112 of the flow through the pipe, an AI / ML structural model 114 of the chamber wall, an AI / ML plasma model 116 of the region adjacent to the pedestal, an AI / ML CFD model 118 of the region adjacent to the pedestal, an HFS CFD model 120 of the gap between the showerhead and the pedestal, an HFS chemical model 122 of the gap between the showerhead and the pedestal, an HFS structural model 124 of the chamber wall, an AI / ML plasma model 126 of the region adjacent to the showerhead, an HFS thermal model 128 of the chamber wall, and / or an AI / ML CFD model 130 of the chamber wall. It should be noted that the model assignments shown in the digital twin 100, which have a specific model type, represent a specific class of physical phenomena, and are used in a specific location or system of the process chamber, are merely illustrative. In some embodiments, the model types and classes of physical phenomena can be any appropriate type and combination.

[0090] In some embodiments, the digital twin 100 can use AI / ML models to simulate specific locations, systems, or subsystems of a process chamber with respect to a particular class of physical phenomena that are relatively stable over time. For example, AI / ML models can be used when the values ​​of parameters associated with a class of physical phenomena do not change substantially within a relatively short period (e.g., within nanoseconds, milliseconds, etc.). As a more specific example, AI / ML models can be used to simulate structural effects, plasma conditions, thermal effects, and / or fluid dynamics of the chamber wall in any of the various regions outside the gap between the process chamber base and the showerhead.

[0091] Additionally or alternatively, in some embodiments, the digital twin 100 can use an AI / ML model to simulate specific locations or systems within the process chamber with respect to a particular class of physical phenomena for which the HFS model cannot correctly produce results. For example, it is possible to use the AI / ML model to simulate the thermal properties of a ceramic pedestal, including an inner mesh. In some such embodiments, the AI / ML model can be trained using physical sensor measurements, where sensors are placed at various physical locations within the process chamber.

[0092] In some embodiments, the digital twin 100 can use an HFS model to simulate a specific location or system within a process chamber with respect to a particular class of physical phenomena that change over a short time scale. For example, as shown in Figure 1, the HFS model can be used to simulate the plasma, thermal, and / or chemical properties within the gap between the showerhead and the base.

[0093] In some embodiments, the digital twin 100 can use closed-form solutions, including one or more closed-form physical equations, in situations where such closed-form solutions are known. Examples of situations in which closed-form solutions can be used include flow on a flat plate, flow through a pipe, bending of a plate, plasma conductivity, electron temperature, and sheath thickness under specific operating conditions.

[0094] In some embodiments, the type of model included in the digital twin 100 used for a particular location and a particular class of physical phenomena can be switched. For example, if an HFS model is used for a particular location or system in a process chamber (e.g., a base, chamber wall, etc.), a trained AI / ML model of that particular location or system in the process chamber representing the same class of physical phenomena can replace the HFS model. In this way, the models included in the digital twin 100 can be modularized. This may be appropriate, for example, when a new model is developed or when one or more components of a manufacturing machine are changed while other components remain unchanged.

[0095] Referring to Figure 2, schematic diagrams are shown for training AI / ML models according to several embodiments of the disclosed subject matter.

[0096] As shown, the AI / ML model 230 can be trained using virtual sensor data generated by the HFS model 210 and / or physical sensor data measured using the physical chamber 220.

[0097] It should be noted that in some embodiments, the AI / ML model 230 and the HFS model 210 can correspond to the same class of physical phenomena (e.g., thermal, plasma, chemical, CFD, and / or structural) and represent the same location or system in the process chamber.

[0098] If available, training data from sensors in a functioning physical system is used. However, often the available physical training data is insufficient to successfully train a reliable AI / ML model. In some embodiments, this challenge is addressed by generating virtual sensor data in any suitable manner using the HFS model 210 or other models. For example, in some embodiments, the HFS model 210 can use input 102 to simulate a physical phenomenon such as heat flux or heat flow in a series of time and / or spatial steps and generate simulated time-series data in a group of virtual sensors. As a more specific example, if the HFS model 210 is a thermal model, the HFS model 210 can generate a simulated time series of simulated thermocouple measurements from a group of virtual thermocouples at different locations. As another more specific example, if the HFS model 210 is a structural model, the HFS model 210 can generate a simulated time series of measurements such as pressure, force, etc., from a group of virtual sensors simulated at different locations. As another, more specific example, if HFS model 210 is a chemical model, it can generate a simulated time series of chemical reaction states based on molecular dynamics. It should be noted that in some embodiments, the virtual training data is generated for locations of manufacturing equipment that cannot be accessed (at least in practice) using physical sensors. For example, it is often impractical to collect physical data in the gap between the pedestal and the showerhead during a plasma reaction.

[0099] The physical chamber 220 can generate physical sensor data in any suitable manner. For example, in some embodiments, physical sensors that measure any suitable type of physical phenomenon (e.g., temperature, pressure, force, power, etc.) can be located at any suitable physical location within the process chamber. In some embodiments, the physical sensor data may be any suitable time-series data measured at any suitable frequency or time step.

[0100] Although input 102 is shown as an input to both the HFS model 210 and the physical chamber 220, it should be noted that in some embodiments, the input used by the HFS model 210 may differ from the input used by the physical chamber 220, which may be a generally controllable parameter of the physical chamber. For example, in some embodiments, the HFS model 210 may obtain a subset of input 102 as inputs relating to a class of physical phenomena represented by the HFS model 210 and / or relating to a location or process chamber system modeled by the HFS model 210.

[0101] The AI / ML model 230 can be trained in any suitable manner. For example, in some embodiments, it is possible to construct a training set that includes training samples generated using virtual sensor data from the HFS model 210 and / or physical sensor data generated using the physical chamber 220. As a more specific example, for a particular training sample, the input value can correspond to the value of input 102 and the target output can correspond to the sensor data, whether virtual (i.e., if the training sample is based on the HFS model 210) or physical (i.e., if the training sample is based on physical sensor data from the physical chamber 220).

[0102] It should be noted that in some embodiments, the AI / ML model 230 may be trained using only virtual sensor data from the HFS model 210, or only physical sensor data generated using the physical chamber 220. For example, if the AI / ML model 230 represents a class of locations or systems and physical phenomena that are too complex to have an associated HFS model (e.g., the thermal properties of a ceramic pedestal with an embedded mesh), the AI / ML model 230 can be trained using only physical sensor data. Conversely, if the AI / ML model 230 represents a class of locations or systems and physical phenomena where physical sensor data cannot be measured (e.g., because a physical sensor cannot be placed at the location in the process chamber), the AI / ML model 230 can be trained using only virtual sensor data.

[0103] It should be noted that in some embodiments, some of the physical sensor data may be saved to validate the AI / ML model 230 after it has been trained using the virtual sensor data and / or the rest of the physical sensor data.

[0104] In addition, it should be noted that in some embodiments, the HFS model 210 can generate virtual sensor data on relatively short timescales (e.g., picoseconds, nanoseconds, etc.) and / or relatively small spatial scales (e.g., angstroms, nanometers, etc.). In some embodiments, data on short timescales and / or small spatial scales can be used to train the AI / ML model 230, which, once trained, can generate predictive outputs on longer timescales (e.g., milliseconds, seconds, minutes, hours, etc.) and / or larger spatial scales (millimeters, centimeters, etc.) corresponding to the target timescale or spatial scale in the full operation of the process chamber. For example, an HFS chemistry model can generate simulated chemical reaction kinetics on a picosecond timescale (e.g., to simulate the reaction chemistry of a wafer). Continuing this example, an HFS chemistry model can be used to train a corresponding AI / ML chemistry model that generates outputs on relatively long timescales (e.g., seconds) which can be used as input to other models in the digital twin.

[0105] Referring to Figure 3, a block diagram of a coupled model for generating a digital twin of a process chamber is shown, according to several embodiments of the disclosed subject matter.

[0106] As described above, the digital twin 100 can include multiple models, each associated with a process chamber location or system, representing a specific class of physical phenomena and of a specific type (i.e., HFS, AI / ML, or closed-form solution).

[0107] For example, as shown in Figure 3, the digital twin 100 may include an HFS plasma model 302, an HFS CFD model 304, an HFS thermal model 306, an HFS structural model 308, an AI / ML model 310, an AI / ML CFD model 312, an AI / ML thermal model 314, and / or an AI / ML structural model 316.

[0108] It should be noted that in some embodiments, various models that can be incorporated into the digital twin 100 are not shown in Figure 3. For example, in some embodiments, it is possible to include one or more closed-form solutions not shown in Figure 3 in the digital twin 100. Another example is the inclusion of additional HFS and AI / ML models beyond those shown in Figure 3. In a more specific example, the HFS plasma model 302 may correspond to a specific system or location in the process chamber, such as the gap between the base and the showerhead. Continuing this specific example further, in some embodiments, a second HFS plasma model 302 corresponding to a different system or location in the process chamber (e.g., inside a pipe) may be included in the digital twin 100.

[0109] In addition, it should be noted that in some embodiments, any of models 302 to 316 may be omitted. For example, if AI / ML plasma model 310 adequately represents the plasma characteristics for a particular location or system in the process chamber, then HFS plasma model 302 for the same location or system in the process chamber may be omitted. As another example, if AI / ML plasma model 310 does not adequately represent the plasma characteristics for a particular location or system in the process chamber, then AI / ML plasma model 310 may be omitted.

[0110] As shown in Figure 3, the output from one model can be used as an input to another model. For example, the AI / ML model 330 can receive the output generated by the HFS plasma model 302 as input. As another example, the AI / ML CFD model 312 can receive the output generated by the HFS CFD model 304 and / or the AI / ML plasma model 310 as input. Note that in some embodiments, an AI / ML model can obtain the output generated by another AI / ML model, an HFS model, and / or a closed-form solution as input. Similarly, an HFS model can obtain the output generated by another HFS model, an AI / ML model, and / or a closed-form solution as input.

[0111] It should be noted that the model connections shown in Figure 3 are for illustrative purposes only. In some embodiments, a model can take as input the outputs generated by any suitable number of models (e.g., 1, 2, 5, 10, and / or any other suitable number). In addition, it should be noted that two models can be connected sequentially or completely. For example, when two models are connected sequentially, the second model can be configured to take as input the outputs generated by the first model. Continuing this example, when two models are connected sequentially, the first model does not take as input the outputs generated by the second model. Conversely, when two models are completely connected, the first model can produce the outputs obtained as input by the second model, and in addition, it can take as input the outputs generated by the second model.

[0112] In addition, it should be noted that in some embodiments, an AI / ML model can be used to match the output of a first HFS model to the expected input of a second HFS model. For example, the HFS plasma model 302 may generate a set of outputs, a subset of which is required by the HFS CFD model 304. Continuing this example, some of the outputs of the HFS plasma model 302 may not be in the format required by the HFS CFD model 304.

[0113] In some embodiments, the coupling of one model to another can be carried out using logic such as that represented by coupling block 318. In some embodiments, coupling block 318 can implement any suitable function that enables the first model to provide the second model with outputs that are used as inputs by the second model. For example, in some embodiments, coupling block 318 can determine (e.g., based on user-specified instructions and / or in any other suitable manner) that the second model should receive the outputs generated by the first model as input. Continuing this example, coupling block 318 can wait until the first model generates a specified output, such as a temperature value at a particular time step and spatial location, or the composition of a gas species at a particular time step and spatial location. Continuing this example further, in response to receiving the specified output, coupling block 318 can send the output to the second model. Specifically, coupling block 318 can call any suitable function associated with the second model, using the output of the first model as an input parameter for the function call.

[0114] It should be noted that any model included in the digital twin can be run in parallel or in series with other models in the digital twin. The coupling block 318 can be configured to send model results between models so that the models can operate in parallel and / or in series.

[0115] It should be noted that in some embodiments, the architecture of the digital twin 100, which specifies the type of model used to represent each location or system in the process chamber with respect to each class of physical phenomena, can be designed or specified in any suitable manner. For example, in some embodiments, the architecture can be specified using a user interface that allows the user to select specific models to include in the digital twin 100 (e.g., an HFS plasma model of the gap between the showerhead and the base, an AI / ML model of the thermal properties of the base, etc.). In some such embodiments, the coupling of different models to be included in the digital twin 100 can be further specified via the user interface. For example, a particular pair of models can be shown as fully coupled or sequentially coupled. In addition, in some embodiments, a particular output to be awaited by the coupling block 318 can be specified via such a user interface.

[0116] In some embodiments, the verification block 320 can verify the performance of the AI / ML model when coupled with other models in the digital twin 100. For example, in some embodiments, the verification block 320 can verify the performance of the AI / ML thermal model 314 when it receives input from other models, such as the AI / ML CFD model 312, the HFS CFD model 304, the HFS thermal model 306, and / or any other suitable model.

[0117] In some embodiments, validation can be performed using experimental results 322. In some embodiments, the validation block 320 can be configured to calibrate one or more AI / ML models to match experimental results 322. For example, in some embodiments, design of experiments (DOE) techniques can be used to find the combination of variables that best match experimental results 322. In some embodiments, an optimization algorithm can be used to identify the combination of variables that best match experimental results 322. In some embodiments, the identified combination of variables can be further validated in relation to hardware test conditions, for example, to ensure that the identified variables and / or the values ​​of the variables are physically possible under specific hardware test conditions.

[0118] In some cases, a digital twin of a manufacturing machine is generated by identifying physical phenomena and the locations within the machine where those phenomena are represented. This operation may, in some sense, involve mapping physical phenomena to specific locations within the machine. It should be noted that not all physical phenomena need to be represented to all locations within the machine. For example, plasma conditions or fluid dynamics do not need to be modeled to chamber walls and / or locations within the chamber that have little effect on the process performed on the substrate. After the physical phenomena and locations have been selected / mapped, a model type is selected for each combination of location and phenomenon. As shown, this operation may involve selecting the least computationally intensive model type available to represent the process at a level of fidelity sufficient for the digital twin to accurately produce its output. The selected phenomenon / location / model type combinations generate individual models. In some cases, this operation may involve selecting and parameterizing HFS and / or closed-form functions that accurately predict the relevant physical conditions. In some cases, this operation may involve acquiring training data and then training and validating the AI / ML model. Finally, it is possible to develop appropriate coupling logic to enable all models to run collaboratively and collectively generate digital outputs. In some embodiments, all models, including those that do not necessarily require training, operate collaboratively as part of the training process. For example, the entire output of the digital twin can be used to calculate the error in the current version of the model being trained.

[0119] Referring to Figure 4A, an example of a process for generating a digital twin of a process chamber according to several embodiments of the disclosed subject matter is shown. In 402, HFS model results can be generated for a first location and a first class of physical phenomena in the process chamber. As described above, the first location in the process chamber may be any suitable location, system, or subsystem of the process chamber, such as the ESC base, showerhead, gap between the showerhead and base, or wafer being fabricated. The first class of physical phenomena may be any suitable class of physical phenomena, such as thermal properties, chemical properties, CFD properties, structural properties, and / or plasma properties.

[0120] HFS model results can be generated in any suitable manner. For example, in some embodiments, HFS model results may include time-series data showing simulated values ​​at a series of time steps. As another example, in some embodiments, HFS model results may include simulated values ​​at different simulated spatial locations, such as the spatial locations corresponding to virtual sensors.

[0121] In 404, physical sensor data corresponding to a first location in the process chamber can be received. As described above in relation to Figure 2, the physical sensor data may include any appropriate measurements, such as temperature measurements, force measurements, pressure measurements, gas flow rate measurements, emission measurements, spectral measurements, and / or any other appropriate measurements. The physical sensor data can be collected from physical sensors located at any appropriate physical location in the process chamber.

[0122] In 406, an AI / ML model can be trained to represent a first location and a first class of physical phenomena in the process chamber. As described above in relation to Figure 2, the AI / ML model can be trained using HFS model results and / or physical sensor data. For example, in some embodiments, a training set can be created that includes HFS model results and / or physical sensor data, and the AI / ML model can be trained using this training set.

[0123] In 408, the output of the AI / ML model representing the first location and first class of physical phenomena in the process chamber can be used as input to the second model and / or second class of physical phenomena in the second location of the process chamber. Note that in some embodiments, the second model and / or second class of physical phenomena in the second location of the process chamber may be an AI / ML model, an HFS model, or a closed-form solution.

[0124] In some embodiments, the output of the AI / ML model representing the first location of the process chamber and the first physical phenomenon can be provided to the second model in any suitable manner. For example, in some embodiments, the output of the AI / ML model can be provided to a coupling block or module that receives the output of the AI / ML model and sends the output to the second model for use by the second model as input to the second model.

[0125] In 410, a determination can be made as to whether the digital twin is complete. The determination as to whether the digital twin is complete can be made in any appropriate manner based on any appropriate information. For example, the digital twin can be determined to be complete in response to a determination that the digital twin contains a model for each location or system in a set of process chamber locations and / or systems. In some embodiments, the set of process chamber locations and / or systems can include any appropriate number of chamber locations and / or systems specified as required for an accurate digital twin model of the process chambers.

[0126] It should be noted that in some embodiments, the digital twin can be considered complete when the models included in the digital twin are combined.

[0127] In 410, in response to the decision that the digital twin is not complete ("No" in 410), the process can loop back to 402, which can generate HFS model results for different locations and / or different classes of physical phenomena in the process chamber.

[0128] Conversely, in response to the decision at 410 that the digital twin is complete ("yes" at 410), the process can be terminated at 412.

[0129] Referring to Figure 4B, an example of a process for using a digital twin is shown according to several embodiments of the disclosed subject matter. In particular, Figure 4B shows a process for using a digital twin in one of the following scenarios: 1) design verification, 2) process verification, or 3) predictive maintenance. Note that predictive maintenance is applicable to process chambers that are already deployed and in use. In contrast, design verification and / or process verification may be applicable to process chambers or processes that are under design and not currently deployed.

[0130] In 450, a group of digital twin inputs related to one of the following can be identified: 1) design verification, 2) process verification, or 3) predictive maintenance.

[0131] In some embodiments, the inputs related to design verification may include structural specifications of the new or modified system, subsystem, or component being evaluated. In addition, in some embodiments, the inputs related to design verification may include structural specifications of other systems, subsystems, or components of the process chamber that are not being evaluated. For example, if a potentially new pedestal of the ESC is being evaluated, the inputs may also include the specifications of the potentially new pedestal, as well as the specifications of other systems, subsystems, or components of the process chamber that remain unchanged.

[0132] In some embodiments, the inputs related to process validation may include information indicating the process or recipe being carried out within the process chamber. For example, the information may include setpoints (e.g., temperature setpoint, pressure setpoint, etc.), the composition of the gas mixture, the gas flow rate, and so on.

[0133] In some embodiments, inputs related to predictive maintenance may include specifications for the deployed process chamber and / or specifications for recipes implemented for the deployed reactor. In some embodiments, the specifications for the deployed process chamber may include specifications for the process chamber system, subsystems, and / or components, such as model numbers of specific components, dimensions of any appropriate side of the system or subsystem (e.g., base size, base thickness, dimensions of the mesh within the base, chamber wall thickness, etc.), materials used in a particular system or subsystem, and / or any other appropriate specification information. In some embodiments, the recipe specifications may include information indicating the settings used in the recipe (e.g., temperature setting, pressure setting), the composition of the gas used, the gas flow rate, etc.

[0134] In 452, a group of digital twin inputs can be used to generate predicted wafer characteristics using a digital twin. In some embodiments, the predicted wafer characteristics can correspond to wafers fabricated using a process chamber when using digital twin inputs.

[0135] In 454, intermediate values ​​of the models included in the digital twin can be identified. In some embodiments, intermediate values ​​of the models may include values ​​generated by any model included in the digital twin that correspond to any location in the process chamber and / or represent any class of physical phenomena. For example, values ​​may be generated by values ​​from a thermal model of the pedestal, a thermal model of the showerhead, a plasma model of the gap between the pedestal and the showerhead, a structural model of the chamber wall, a hydrodynamic model of the gap between the pedestal and the showerhead, a chemical model of the wafer surface, and / or any other suitable model. Note that the models may be AI / ML, HFS, and / or closed-form solutions.

[0136] As a more specific example, the values ​​in the thermal model of the pedestal may include time-series values ​​that include simulated temperature measurements associated with various locations on the pedestal.

[0137] As another, more specific example, the plasma model values ​​for the gap between the pedestal and the showerhead could include time-series values ​​showing measurements of simulated plasma temperature, density, potential, and / or composition at various locations within the gap between the pedestal and the showerhead.

[0138] In 456, the information may be presented in one of the following scenarios: 1) design verification, 2) process verification, and 3) predictive maintenance. In some embodiments, the information can be generated and presented based on predicted wafer characteristics and / or intermediate values ​​of the model contained in the digital twin.

[0139] For example, if predicted wafer characteristics are used for design or process verification, an indicator can be provided to show whether the predicted wafer characteristics contain a specific defect. Another example is the ability to indicate whether a wafer corresponding to the predicted wafer characteristics does not meet any appropriate quality criteria.

[0140] As another example, when predicted maintenance information is presented, intermediate values ​​of models included in the digital twin can be used to identify potential failures in a simulated process chamber system, subsystem, or component. As a more specific example, it is possible to identify a potential failure of a pedestal in response to a determination that the values ​​of the pedestal's thermal model are outside of normal operating conditions. As yet another more specific example, it is possible to identify a potential defect in a fabricated wafer in response to a determination that the values ​​of the plasma model of the gap between the pedestal and the showerhead are outside of normal operating conditions.

[0141] In some embodiments, in response to the identification of a potential failure, a digital twin can be used to identify one or more recommendations to mitigate that potential failure. For example, in some embodiments, it is possible to identify changes to recipe parameters. As another example, in some embodiments, it is possible to identify the replacement of process chamber components. In some such embodiments, it is possible to evaluate the recommendations by rerunning the digital twin with updated input values ​​representing the modifications.

[0142] The process can be terminated at 458.

[0143] Purpose Using the techniques described herein for generating a digital twin of a process chamber, it is possible to generate a digital twin that balances accurate simulation of the process chamber with the use of computational resources. For example, the model type of individual models can be selected (e.g., representing a specific location in the process chamber and / or a specific class of physical phenomena), thereby allowing computationally intensive models such as HFS models to be used in situations where the accuracy of HFS simulation would be particularly beneficial. Conversely, less computationally intensive models such as AI / ML models and / or closed-form solutions can be used in situations where an HFS model cannot be used (e.g., because the complexity of the situation cannot be simulated by an HFS model), where an AI / ML model can be trained, and / or where it can be represented using one or more closed-form solutions.

[0144] By combining models of different locations within the process chamber and representing different classes of physical phenomena so that the different models interact with each other, it is possible to represent the complexity of the entire process chamber using a digital twin.

[0145] By simulating the entire process chamber, new designs of process chamber systems, subsystems, and / or components, as well as new processes or recipes, can be evaluated before deployment. Simulation of the design and / or process can make it possible to identify potential problems in the design or process before costly failures occur after deployment.

[0146] In addition, simulation of the entire process chamber can potentially identify potential failures in the process chamber system, subsystems, and / or components in advance, thereby enabling proactive maintenance planning, component replacement, and / or recipe parameter modifications that can mitigate potential failures. Such predictive maintenance can save costs and reduce downtime for semiconductor manufacturing equipment.

[0147] Context of the disclosed computational embodiments Certain embodiments disclosed herein relate to a computing system for generating and / or using various computational models. Certain embodiments disclosed herein relate to a method for generating and / or using computational models implemented for such a system. A system for generating computational models can also be configured to receive data and instructions, such as program code representing physical processes occurring during semiconductor device fabrication operations. In this way, computational models are generated or programmed for such a system.

[0148] Many types of computing systems having any of the various computer architectures can be used as disclosed systems for implementing computational models and algorithms for generating and / or optimizing such models. For example, the system may include software components that run on one or more general-purpose processors, or on specially designed processors such as application-specific integrated circuits (ASICs) or programmable logic devices (e.g., field-programmable gate arrays (FPGAs)). Furthermore, the system may be implemented on a single device or distributed across multiple devices. The functions of the computational elements may be integrated with each other or further divided into multiple submodules.

[0149] In some embodiments, the code executed during the generation or execution of a computational model on a properly programmed system can be embodied in the form of software elements that can be stored on a non-volatile storage medium (e.g., optical discs, flash memory devices, mobile hard disks, etc.) and contain a large number of instructions for creating computer devices (e.g., personal computers, servers, network equipment, etc.).

[0150] At a certain level, software elements are implemented as a set of commands prepared by the programmer / developer. However, module software that can be executed by computer hardware is executable code committed to memory using a specific set of machine language instructions, or "machine code," selected from "native instructions," designed for the hardware processor. The machine language instruction set, or native instruction set, is known to the hardware processor and is essentially built into the hardware processor. It is the "language" for system and application software to communicate with the hardware processor. Each native instruction is a distinct piece of code that is recognized by the processing architecture and can specify specific registers for arithmetic, addressing, or control functions, specific memory locations or offsets, and specific addressing modes used to interpret operands. More complex operations are constructed by combining these simpler native instructions, which are executed sequentially or as directed by control flow instructions.

[0151] The relationship between executable software instructions and the hardware processor is structural. In other words, the instruction itself is a set of symbols or numbers. A set of symbols or numbers inherently conveys no information. It is the processor, pre-configured by design to interpret the symbols / numbers, that gives meaning to the instruction.

[0152] The models used herein may be configured to run on a single machine in a single location, on multiple machines in a single location, or on multiple machines in multiple locations. When multiple machines are used, each machine can be tuned to its specific tasks. For example, operations requiring large code blocks and / or considerable processing power may be performed on larger machines and / or stationary machines.

[0153] In addition, certain embodiments relate to tangible and / or non-temporary computer-readable media or computer program products containing program instructions and / or data (including data structures) for performing various computer implementation operations. Examples of computer-readable media include, but are not limited to, semiconductor memory devices, phase-change devices, magnetic media such as disk drives, optical media such as magnetic tapes and CDs, magneto-optical media, and hardware devices specifically configured to store and execute program instructions, such as read-only memory devices (ROM) and random-access memory (RAM). Computer-readable media may be directly controlled by the end user, or the media may be indirectly controlled by the end user. Examples of directly controlled media include media located at the user's facility and / or media not shared with other entities. Examples of indirectly controlled media include media that are indirectly accessible to the user via an external network and / or via a service providing shared resources such as a “cloud”. Examples of program instructions include both machine code, such as that generated by a compiler, and files containing high-level code that can be executed by a computer using an interpreter.

[0154] In various embodiments, the data or information used in the disclosed methods and apparatus is provided in electronic form. Such data or information may include design layouts, simulation values, sensor values, etc. When used herein, data or other information provided in electronic form is available for storage on machines and transmission between machines. Conventionally, data in electronic form is provided digitally and can be stored as bits and / or bytes in various data structures, lists, databases, etc. The data can be embodied electronically, optically, etc.

[0155] In some embodiments, the computational model can be seen as application software that interfaces with user and system software. System software typically interfaces with computer hardware and associated memory. In some embodiments, system software includes operating system software and / or firmware, as well as any middleware and drivers installed on the system. System software provides basic, task-independent functionality for the computer. In contrast, modules and other application software are used to accomplish specific tasks. Each native instruction for a module is stored in a memory device and represented numerically.

[0156] An exemplary computer system 500 is illustrated in Figure 5. As shown, the computer system 500 includes an input / output subsystem 502 which can implement an interface for interacting with a human user and / or other computer systems, depending on the application. Embodiments of the present disclosure may be implemented in program code on system 500 having an I / O subsystem 502 used to receive input program statements and / or data from a human user (e.g., via a GUI or keyboard) and to display them to the user. The I / O subsystem 502 may include, for example, a keyboard, mouse, graphical user interface, touchscreen, or other input interface, and, for example, an LED or other flat-screen display, or other output interface.

[0157] The communication interface 507 may include any suitable components or circuits used for communication over any suitable communication network (e.g., the Internet, an intranet, a wide area network (WAN), a local area network (LAN), a wireless network, a virtual private network (VPN), and / or any other suitable type of communication network). For example, the communication interface 507 may include a network interface card circuit, a wireless communication circuit, and the like.

[0158] Program code may be stored in a non-temporary medium such as secondary memory 510 or memory 508, or both. In some embodiments, secondary memory 510 may be persistent memory. One or more processors 504 read program code from one or more non-temporary media and execute the code, enabling the computer system to achieve methods implemented by the embodiments herein, such as methods relating to the generation or use of models described herein. Those skilled in the art will understand that a processor may accept source code, such as statements for performing training and / or modeling operations, interpret or compile the source code into machine code understandable at the processor's hardware gate level. A bus 505 connects the I / O subsystem 502, the processors 504, peripheral devices 506, communication interface 507, memory 508, and secondary memory 810.

[0159] conclusion The description has included numerous specific details to provide a complete understanding of the presented embodiments. The disclosed embodiments can be practiced without some or all of these specific details. In other examples, well-known process behaviors are not described in detail so as not to unnecessarily obscure the disclosed embodiments. Although the disclosed embodiments have been described in conjunction with specific embodiments, it will be understood that the specific embodiments are not intended to limit the disclosed embodiments.

[0160] Unless otherwise indicated, the method operations and device features disclosed herein involve techniques and apparatus commonly used in metrology, semiconductor device fabrication techniques, software design and programming, and statistics, which are within the scope of the art.

[0161] Unless otherwise defined herein, all technical and scientific terms used herein have the same meaning as commonly understood by those skilled in the art. Various scientific dictionaries containing the terms included herein are well known and available to those skilled in the art. Any methods and materials similar to or equivalent to those described herein will find use in the practice or testing of the embodiments disclosed herein, although only a few methods and materials are described.

[0162] A numerical range includes the number that defines the range. Any maximum numerical limit given throughout this specification is intended to include all lower numerical limits as if such lower numerical limits were explicitly stated herein. Any minimum numerical limit given throughout this specification includes all higher numerical limits as if such higher numerical limits were explicitly stated herein. Any numerical range given throughout this specification includes all narrower numerical ranges that fall within such wider numerical ranges as if all such narrower numerical ranges were explicitly stated herein.

[0163] The headings provided herein are not intended to limit this disclosure.

[0164] As used herein, the singular terms “a,” “an,” and “the” include plural references unless otherwise clearly indicated by the context. As used herein, the term “or” refers to a non-exclusive “or” unless otherwise indicated.

[0165] Various computing elements, including processors, memory, instructions, routines, models, or other components, may be described or claimed to be "configured" to perform one or more tasks. In such contexts, the phrase "configured to" is used to imply a structure by indicating that the component contains a structure (e.g., stored instructions, circuits, etc.) that performs one or more tasks while in operation. Thus, a unit / circuit / component can be said to be configured to perform tasks even if the specified component is not necessarily currently in operation (e.g., not turned on).

[0166] The constructs used with the language "configured to..." can refer to hardware, such as circuits, or memory that stores executable program instructions to perform an operation. In addition, "configured to..." can refer to a general structure (e.g., a general circuit) that is operated by software and / or firmware (e.g., an FPGA, or a general-purpose processor running the software) to operate in a manner capable of performing the described task. Furthermore, "configured to..." can refer to one or more memories or memory elements that store computer executable instructions for performing the described task. Such memory elements can 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 manufacturing facility) to produce a device (e.g., an integrated circuit) adapted to implement or perform one or more tasks.

Claims

1. A digital twin of a process chamber for semiconductor manufacturing equipment, A first model of the first location of the process chamber, The second model of the second location of the process chamber and A computer-readable non-temporary recording medium comprising one or more computer-readable non-temporary recording media having logic configured to implement, The first model of the first location of the process chamber is coupled to the second model of the second location of the process chamber. The first model of the first location of the process chamber and the second model of the second location of the process chamber are each of a model type that is one of 1) an AI / ML model, 2) an HFS model, or 3) a closed-form solution. The first model of the first location of the process chamber and the second model of the second location of the process chamber each represent a class of physical phenomena which is one of the following: 1) thermal properties, 2) plasma properties, 3) hydrodynamics, 4) structural properties, or 5) chemical reactions. The first model of the first location of the process chamber is of a different model type than the second model of the second location of the process chamber, and the first model of the first location of the process chamber represents a different class of physical phenomena than the second model of the second location of the process chamber. Digital twin.

2. A digital twin according to claim 1, The first location is a digital twin, which is one of the following: 1) the base of the ESC, 2) the showerhead, 3) the gap between the base and the showerhead, 4) the chamber wall, or 5) the surface of the wafer fabricated by the process chamber.

3. A digital twin according to claim 1, A digital twin comprising the first model of the first location of the process chamber, coupled to the second model of the second location of the process chamber, the first model of the first location of the process chamber, which provides the output used by the second model of the second location of the process chamber to the second model of the second location of the process chamber.

4. A digital twin according to claim 3, A digital twin comprising a first model of a process chamber, coupled to a second model of a second location of the process chamber, the first model of a first location of the process chamber, which receives outputs used by the first model of the first location of the process chamber from the second model of the second location of the process chamber.

5. A computer program for generating a digital twin of a process chamber, A function for generating multiple HFS values ​​for a first location in a process chamber using a high-fidelity simulation (HFS) model of the first location in the process chamber, A function for receiving a plurality of sensor measurements corresponding to the first location in the process chamber, A function for training an artificial intelligence / machine learning (AI / ML) model of the first location in the process chamber using the plurality of HFS values ​​and at least one of the plurality of sensor measurements, A computer program that enables the function of combining the trained AI / ML model of the first location of the process chamber with the model of the second location of the process chamber, wherein the digital twin of the process chamber consists of the trained AI / ML model of the first location of the process chamber and the model of the second location of the process chamber.

6. A computer program according to claim 5, The model of the second location of the process chamber is a computer program which is one of 1) an AI / ML model, 2) an HFS model, or 3) a closed-form solution.

7. A computer program according to any one of claims 5 or 6, Both the HFS model and the AI / ML model of the first location in the process chamber are computer programs that model the same class of physical phenomena.

8. A computer program according to any one of claims 5 or 6, The trained AI / ML model for the first location of the process chamber and the model for the second location of the process chamber are computer programs that each model a class of physical phenomena.

9. A computer program according to claim 8, A computer program in which the class of physical phenomena is one of thermal properties, plasma properties, fluid dynamics, structural properties, or chemical reactions.

10. A computer program according to claim 8, The trained AI / ML model for the first location of the process chamber and the model for the second location of the process chamber are computer programs that model different classes of physical phenomena.

11. A computer program according to any one of claims 5 or 6, A computer program that generates simulation values ​​for the HFS model of the first location in the process chamber at a shorter time step than the time step of the AI / ML model of the first location in the process chamber.

12. A computer program according to any one of claims 5 or 6, A computer program wherein the first location in the process chamber is one of the following: 1) the base of an electrostatic chuck (ESC), 2) a shower head, 3) the gap between the shower head and the base, 4) the chamber wall, or 5) the surface of a wafer produced by the process chamber.

13. A computer program according to any one of claims 5 or 6, A computer program having a function for combining the trained AI / ML model of the first location of the process chamber with the model of the second location of the process chamber, the function for providing a plurality of outputs of the trained AI / ML model of the first location of the process chamber with the model of the second location of the process chamber.

14. A computer program according to claim 13, The function for providing the plurality of outputs of the trained AI / ML model at the first location of the process chamber to the model at the second location of the process chamber is: A function for waiting until the multiple outputs of the trained AI / ML model at the first location of the process chamber are received, A function for sending the plurality of outputs to the model at the second location of the process chamber. A computer program that includes the following features.

15. A computer program according to any one of claims 5 or 6, A computer program comprising combining the trained AI / ML model of the first location of the process chamber with the model of the second location of the process chamber, wherein the program provides a plurality of outputs of the model of the second location of the process chamber to the trained AI / ML model of the first location of the process chamber.

16. A computer program according to any one of claims 5 or 6, Furthermore, a computer program that includes the trained AI / ML model of the first location of the process chamber in the digital twin, and then uses a computer to implement a function for verifying the performance of the trained AI / ML model of the first location of the process chamber.

17. A computer program according to claim 16, Verifying the performance of the trained AI / ML model is Data is generated using the digital twin, which includes the trained AI / ML model of the first location of the process chamber and the model of the second location of the process chamber. The simulated data is compared with experimental data collected using multiple sensors associated with the physical process chamber. A computer program that includes the following features.

18. A computer program according to any one of claims 5 or 6, A computer program further comprising computer-executable instructions for replacing the HFS model of the second location in the process chamber with a trained AI / ML model of the second location in the digital twin, wherein the model of the second location in the process chamber is an HFS model.

19. A computer program for using a digital twin of a process chamber, A function for identifying multiple inputs to a digital twin of a process chamber, wherein the digital twin comprises a first model of a first location of the process chamber and a second model of a second location of the process chamber, the first model of the first location of the process chamber and the second model of the second location of the process chamber are coupled, and the multiple inputs represent the operating conditions of the process chamber. A function for providing the aforementioned multiple inputs to the digital twin, A computer program that enables a computer to generate predicted wafer characteristics of a simulated wafer using the aforementioned digital twin.

20. A computer program according to claim 19, A computer program comprising a first model of the first location of the process chamber, the first model of the first location of the process chamber, the first model of the first location of the process chamber, and further comprising computer-executable instructions for verifying the specifications of the components based on the predicted wafer characteristics.

21. A computer program according to any one of claims 19 or 20, The plurality of inputs include parameters of a recipe implemented by the process chamber, and further, a computer program that enables a computer to perform a function to verify at least one parameter of the recipe based on the predicted wafer characteristics.

22. A computer program according to any one of claims 19 or 20, The predicted wafer characteristics are a computer program that includes an index of defects in the simulated wafer.

23. A computer program according to any one of claims 19 or 20, Furthermore, a computer program that enables a computer to implement a function for identifying recommendations to modify at least one of the operating conditions based on the predicted wafer characteristics.

24. A computer program according to claim 23, The recommendation is identified in response to a determination that the predicted wafer characteristics indicate defects in the simulated wafer, and is a computer program.

25. A computer program according to claim 23, The recommendation is identified in response to a determination that at least one of the first model and the second model has generated a value indicating abnormal operating conditions for the process chamber, and is a computer program.

Images