Performance predictors for semiconductor manufacturing processes.

Abstract

Methods, systems, and computer programs for predicting the performance of semiconductor manufacturing equipment operations are presented. One method includes operations for obtaining machine learning (ML) models, each associated with predicting a performance metric for the operation of a semiconductor manufacturing tool. Furthermore, each ML model utilizes features that define inputs for the ML model. The method further includes operations for receiving a process definition for manufacturing a product at the semiconductor manufacturing tool. The one or more ML models are utilized to estimate the performance of the process definition used at the semiconductor manufacturing tool. Additionally, the method includes presenting results on a display that indicate an estimate of the performance of manufacturing the product. In some aspects, the use of hybrid models improves the predictive accuracy of the system by augmenting the capabilities of a data-driven model with the reinforcement provided by a physics-based model.

Description

[Technical field]

[0001] <Priority Claim> This application claims the benefit of priority to U.S. Patent Application No. 62 / 966,378, filed January 27, 2020, which is incorporated herein by reference in its entirety.

[0002] The subject matter disclosed herein generally relates to methods, systems, and machine-readable storage media for predicting performance of deposition, etch, and cleaning processes in semiconductor manufacturing tools. [Background technology]

[0003] The background description provided herein is intended to provide a general overview of the contents of the present disclosure. Work by the currently named inventors within the scope of what is described in this Background section, as well as aspects of the description that may not otherwise be regarded as prior art at the time of filing, are not admitted, expressly or impliedly, as prior art against the present disclosure.

[0004] To better understand the physical and chemical mechanisms between species phases on the substrate (e.g., in gas, aqueous, or organic solvent-solvent phases, solid state) and the associated surface dynamics, significant efforts are typically expended on high-fidelity modeling and simulation of semiconductor process reactors and device features. A thorough understanding and predictability of such systems is critical for improving product design, as well as optimizing process conditions, e.g., adjusting parameters of semiconductor manufacturing tools. Summary of the Invention

[0005] Exemplary methods, systems, and computer programs are directed to predicting the performance of semiconductor manufacturing equipment operations. The examples are merely representative of possible variations. Unless expressly stated otherwise, components and functions are optional and may be combined or sub-divided, and operations may be out of order or combined or sub-divided. In the following description, for purposes of explanation, numerous specific details are set forth to provide a thorough understanding of the exemplary embodiments. However, it will be apparent to one skilled in the art that the subject matter may be practiced without these specific details.

[0006] Hybrids of physics-based and data-driven models, also known as hybrid models, offer the opportunity to build relationships between various factors and response variables (generated in experimental or simulation data sets) that are not directly coupled or may have nonlinear dependencies. Because different models (high-fidelity simulations, data-driven models, reduced-order models, etc.) are used to build these relationships, effective hybrid models can identify previously unknown relationships. These hybrid models are computationally cheaper than high-fidelity physics-based simulations and can span a wide range of spatial and time scales. When hybrid models are augmented with experimental data, the predictive value of the results improves as uncertainties and assumptions that cause simulations to vary from experimental data are eliminated or their influence is reduced. This approach aims to reduce the time spent during the design, fabrication, and testing phases of a project, resulting in faster solutions and lower product costs.

[0007] The presented embodiments utilize various truncated data sets generated from physics-based simulations (e.g., high-fidelity CFD simulations, atomistic simulations), approximation methods (e.g., reduced order models, approximation models), experimental data, and other data sources (either as individual models or stack models) to determine the collective impact on processing of semiconductor features. Data obtained from these sources can be analyzed using machine learning (ML) techniques to better predict process behavior and expected outcomes. ML techniques include any combination of statistical modeling, deep neural networks, recurrent neural networks, convolutional neural networks, kriging, dynamic mode decomposition, appropriate orthogonal decomposition, etc., to generate physically constrained data-driven models.

[0008] One method includes acts for obtaining machine learning (ML) models, each model associated with predicting a performance metric for operation of a semiconductor manufacturing tool. Additionally, each ML model utilizes features that define inputs for the ML model. The method further includes acts for receiving a process definition for manufacturing a product at the semiconductor manufacturing tool. The one or more ML models are utilized to estimate performance of the process definition used at the semiconductor manufacturing tool. Additionally, the method includes presenting results on a display indicative of an estimate of performance of manufacturing the product.

[0009] Another general aspect relates to a system including a memory including instructions and one or more computer processors. The instructions, when executed by the one or more computer processors, cause the one or more computer processors to perform operations including obtaining machine learning (ML) models, each model associated with predicting a performance metric for operation of a semiconductor manufacturing tool. Further, each ML model utilizes features that define inputs for the ML model. A process definition for manufacturing a product at a semiconductor manufacturing tool is received, and the one or more ML models are utilized to estimate performance of the process definition used at the semiconductor manufacturing tool. Additionally, results indicative of an estimate of the performance of the process are presented on a display.

[0010] In yet another general aspect, a machine-readable storage medium (e.g., a non-transitory storage medium) includes instructions that, when executed by a machine, cause the machine to perform operations including obtaining machine learning (ML) models, each model associated with predicting a performance metric for operation of a semiconductor manufacturing tool. Further, each ML model utilizes features that define inputs for the ML model. A process definition for manufacturing a product at a semiconductor manufacturing tool is received, and one or more ML models are utilized to estimate performance of the process definition used at the semiconductor manufacturing tool. Additionally, results indicative of an estimate of the performance of the process are presented on a display. [Brief description of the drawings]

[0011] The various accompanying drawings depict only exemplary embodiments of the present disclosure and are not to be considered as limiting its scope.

[0012] [Figure 1] FIG. 1 illustrates the complexity of equipment and process design for semiconductor manufacturing.

[0013] [Diagram 2] FIG. 2 is an etch chamber according to some example embodiments.

[0014] [Diagram 3] FIG. 3 is a diagram illustrating feature modeling at multiple levels in semiconductor manufacturing, in accordance with some example embodiments.

[0015] [Figure 4A] FIG. 4A illustrates an interaction between multiple machine learning models for predicting process behavior, according to some example embodiments.

[0016] [Figure 4B] FIG. 4B is a table illustrating examples of modeling algorithms at different levels, according to some illustrative embodiments.

[0017] [Diagram 5] FIG. 5 is a diagram illustrating the training and use of a machine learning (ML) program, according to some example embodiments.

[0018] [Figure 6] FIG. 6 is a diagram illustrating a process for plasma reduced order model (ROM) simulation, in accordance with some example embodiments.

[0019] [Figure 7] FIG. 7 illustrates a process for plasma ROM using simulations and experiments, according to some example embodiments.

[0020] [Figure 8] FIG. 8 is a diagram illustrating some of the inputs to the ML model, according to some example embodiments.

[0021] [Figure 9] FIG. 9 is a table containing some example features used by the ML model. [Figure 10] FIG. 10 is a table containing some example features used by the ML model.

[0022] [Figure 11] FIG. 11 illustrates a showerhead design using ML, according to some example embodiments.

[0023] [Figure 12] FIG. 12 is a flow chart of a method for predicting performance of semiconductor manufacturing equipment operation according to some example embodiments for performing damage simulation.

[0024] [Figure 13] FIG. 13 is a block diagram illustrating one example of a machine capable of implementing or controlling one or more exemplary process embodiments described herein. DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS

[0025] Depending on the application, there are several types of modeling techniques: stress modeling, thermal modeling, computational fluid dynamics (CFD), plasma modeling, Monte Carlo simulation, etc. These models predict species concentrations, temperature profiles, plasma density and related distributions, flow pressure and velocity fields, calculation of related dimensionless numbers, etc.

[0026] Depending on the domain of interest, spatial scales can range from nanometers to meters, and time intervals can range from picoseconds to minutes, implying a very wide range of spatial and time scales. Because multiple design conditions must be tested and operating ranges must be mapped, modeling these systems using physics-based, chemical-based, or quantum-based methods can be very costly and intractable from industrial environments. For example, individual process executions can take a week or more, and during the design iteration phase, design engineers may be limited to using only two or three such modeling efforts before finalizing the design.

[0027] Furthermore, the fidelity of a physics-based model generally depends on the ability to accurately predict or characterize input parameters or boundary conditions for the model. Often such parameters cannot be directly measured and must be inferred from experimental results. Such challenges are particularly acute when attempting to model plasma environments common in, for example, semiconductor wafer process equipment. Such plasmas are time-varying both by design and due to the accumulation or removal of products or by-products within the process equipment chamber, and the inclusion of a measurement probe in the plasma environment may have undesirable effects on the properties of the plasma being examined. Also, by way of example, measurements of surface or interface properties such as thermal contact resistance or emissivity may be impossible or impractical with sufficient resolution to fully inform a physics-based high-fidelity model.

[0028] Purely data-driven models (e.g., machine learning models) are sometimes utilized to predict the behavior of complex systems. Data-driven models are statistical in nature and do not explicitly require knowledge of the physical system being predicted. However, such models generally exhibit improved predictive value as the amount of training data increases, and require the generation of large experimental or simulated data sets to achieve a useful model. The creation of such large data sets may be impractical or impossible, especially when the purpose of the model is to predict the behavior of a novel process.

[0029] In order to design better and faster semiconductor equipment and semiconductor manufacturing processes, reliable methods and systems are needed that can rapidly predict the behavior of various elements in a semiconductor manufacturing process.

[0030] Figure 1 shows the complexity of designing equipment and processes for semiconductor manufacturing. To manufacture a semiconductor, both the equipment and the processes for making the semiconductor must be designed. Based on a product specification 102, the equipment and processes are designed before the product is brought into production 108.

[0031] Equipment design 104 typically involves designing 110 many of the components of the semiconductor manufacturing equipment, such as the geometry of the chamber, showerhead, pedestal, power supplies, etc. After the components are designed, the system is tested in operation 112 and performance is measured. Typically, there are multiple cycles to perfect the design 110 of the components until the test results are satisfactory for the purpose of producing the product.

[0032] One of the challenges for the design team is that the equipment (e.g., chambers) must be designed before the process is determined. Thus, the equipment is designed to meet a general need, not the specific requirements for a given product. In some cases, the design process can generate feedback to the equipment design 104, and the equipment is redesigned based on testing of the process. However, redesigning the equipment according to a new process design can be expensive and time consuming.

[0033] On the process design 106 side, a recipe 118 is designed 114, which includes the definition of the process such as the workflow, fluid flows, temperature settings for the chamber, pressure in the chamber, duration of different steps, radio frequency (RF) power applied, concentrations of chemicals, electrical biases, etc. Recipe design is a technique where design experts define the recipe based on their past experience.

[0034] After a recipe 118 is designed, the recipe 118 is tested 116 on semiconductor manufacturing equipment. However, multiple cycles of recipe design 114 and testing 116 are typically required to find a satisfactory recipe 118 for producing a product during production 108.

[0035] Both equipment design 104 and process design 106 tend to be expensive and take a long time, such as weeks, months, or years, for each iteration. To accelerate design, simulation is sometimes used to perform testing without running actual tests on semiconductor manufacturing equipment.

[0036] For example, to predict uniformity over a substrate, the pedestal and showerhead are modeled to predict the interactions between them using physics-based models that make predictions based on the physical aspects of the chamber elements. However, physical models take a long time to run, and testing conditions using simulations can take hours or days.

[0037] Physics-based simulation may involve using the laws of physics to predict the behavior of elements and solving conservation equations, boundaries, or mesh nodes within a system. Physics-based simulation is typically used when it is not possible to observe the evolution and outcome of a process, such as observing what is happening inside a chamber.

[0038] There are also behavioral models that describe the output of a manufacturing process based on analytical formulations. One exemplary simulation tool is Lam Research's SEMulator3D®, which provides a voxel model of a semiconductor process based on the expected behavior simulated by a behavioral model. The behavioral model does not directly simulate the behavior of particles in the chamber, but rather measures the results of the operation (e.g., deposition thickness on a substrate).

[0039] Current approaches to solving the challenges of uniform deposition (verified on blanket wafers) and conformal deposition (verified on patterned wafers) are highly experimental. Modeling and simulation can aid in current workflows for component design and prediction of filling performance, but these are often time consuming (e.g., first-principles / physics-based computational methods are multiphysics problem and mesh size dependent, often taking days or weeks per simulation), do not always address all relevant physics (e.g., systems complexity and state-of-the-art technologies do not always have clearly defined physics), can be based on non-physical methods such as behavior-based simulation (e.g., not all behavior is captured in the modeling software), and generally require experimental data for calibration.

[0040] Behavioral model simulations for feature performance prediction are employed in the process. However, these models are limited by the fact that the models are calibrated based on experimental data and employ non-physical parameters to optimize the geometry of the features. Behavioral models cannot be used for the complete system, and the behavior may vary from structure to structure, requiring various calibrations, which means additional testing to capture data for the model.

[0041] The available experimental data also has limitations. While data from experiments is the ultimate ground truth of what is seen on the wafer and within the features, these models based on the data do not allow for the transfer of the obtained learning to other systems. For example, the macroscopic parameters (e.g., atmospheric pressure, wafer temperature, flow rate) do not account for the capture of the fundamental flow fields, charge density, flux transport, and other microscopic parameters that define the fundamental learning of the system. In some cases, there is no directly observable correlation between the experimental and simulation data, and such learning cannot be extended to other designs.

[0042] In summary, current methods employ disparate approaches (modeling, behavior-based simulation, experimentation) and have no way to "learn" from these separate data sources and extend the "insights" to other chamber designs. Experimental data is often the only reliable outcome requiring several costly design iterations, resulting in long product release cycles.

[0043] 2 is an etch chamber 200, according to one embodiment. Exciting an electric field between two electrodes is one way to obtain a radio frequency (RF) gas discharge in the etch chamber. When an oscillating voltage is applied between the electrodes, the resulting discharge is called a capacitively coupled plasma (CCP) discharge.

[0044] The plasma 202 can be generated utilizing a stable feedstock gas to obtain a wide variety of chemically reactive by-products generated by dissociation of various molecules caused by electron-neutral collisions. The chemical aspect of etching involves the reaction of neutral gas molecules and their dissociated by-products with molecules on the surface to be etched, as well as the generation of volatile molecules that can be expelled. When a plasma is generated, positive ions are accelerated out of the plasma across a space charge sheath that separates the plasma from the chamber walls, and impact the wafer surface with sufficient energy to remove material from the wafer surface. This is known as ion bombardment or ion sputtering. However, some industrial plasmas do not generate ions with enough energy to efficiently etch a surface by purely physical means.

[0045] A controller 216 manages the operation of the chamber 200 by controlling different elements in the chamber 200, such as an RF generator 218, a gas source 222, and a gas pump 220. In one embodiment, fluorocarbon gases such as CF4 and C-C4F8 are used in the dielectric etch process for their anisotropic and selective etching capabilities, although the principles described herein can be applied to other plasma generating gases. Fluorocarbon gases readily dissociate into chemically reactive by-products including smaller molecular and atomic radicals. These chemically reactive by-products etch away the dielectric material, which in one embodiment can be SiO2 or SiOCH for low-k devices.

[0046] Chamber 200 illustrates a processing chamber having an upper electrode 204 and a lower electrode 208. The upper electrode 204 may be grounded or coupled to an RF generator (not shown), and the lower electrode 208 is coupled to the RF generator 218 via a matching network 214. The RF generator 218 provides RF power at one, two, or three different RF frequencies. At least one of the three RF frequencies can be turned on or off according to the desired configuration of the chamber 200 for a particular operation. In the embodiment illustrated in FIG. 2, the RF generator 218 provides frequencies of 2 MHz, 27 MHz, and 60 MHz, although other frequencies are possible.

[0047] The chamber 200 includes a gas showerhead over the upper electrode 204 that inputs gases provided by a gas source 222 into the chamber 200, and a perforated confinement ring 212 that allows gases to be exhausted from the chamber 200 by a gas pump 220. In some exemplary embodiments, the gas pump 220 is a turbomolecular pump, although other types of gas pumps may be utilized.

[0048] When a substrate 206 is present in the chamber 200, a silicon focus ring 210 is positioned next to the substrate 206 such that a uniform RF field exists at the bottom of the plasma 202 for uniform etching on the surface of the substrate 206. The embodiment of Figure 2 shows a triode reactor configuration in which the top electrode 204 is surrounded by a symmetric RF ground electrode 224. An insulator 226 is a dielectric material that insulates the ground electrode 224 from the top electrode 204.

[0049] Each frequency can be selected for a specific purpose in the wafer fabrication process. In the example of Figure 2, RF power is provided at 2 MHz, 27 MHz, and 60 MHz, with the 2 MHz RF power providing ion energy control and the 27 MHz and 60 MHz power providing control of the plasma density and dissociation pattern of the chemical. This configuration, where each RF power can be turned on or off, allows for certain processes that use ultra-low ion energy on the substrate or wafer, and certain processes where the ion energy must be low (less than 200 or 200 eV) (e.g., soft etching of low-k materials).

[0050] In another embodiment, 60 MHz RF power is used on the top electrode 204 to obtain very low energy and very high density. This configuration allows for chamber cleaning with high density plasma when the substrate 206 is not in the chamber 200, while minimizing sputtering on the electrostatic chuck (ESC) surface. When the substrate 206 is not present, the ESC surface is exposed, and therefore any ion energy on the surface should be avoided, which is why the bottom 2 MHz and 27 MHz power supplies can be turned off during cleaning.

[0051] 3 illustrates feature modeling at multiple levels in semiconductor manufacturing, according to some exemplary embodiments. The embodiments presented herein describe a method for building tools to predict the behavior and performance of semiconductor manufacturing equipment.

[0052] This type of tool for predicting the behavior of physical entities is called a digital twin, also referred to herein as a system model. Digital twins are digital replicas of living or non-living physical entities and refer to digital replicas of potential and actual physical assets (physical twins), processes, people, places, systems, and devices that can be used for a variety of purposes. Digital twins emphasize the connection between a physical model and a corresponding virtual model or virtual counterpart, and this connection can be augmented by using sensors to generate real-time data.

[0053] A digital twin learns and updates itself from multiple sources, representing its near real-time status, working condition, or location. This learning system learns from itself (e.g., using sensor data that communicates various aspects of its operating conditions), from human experts (e.g., design engineers), from other machines, and from the environment of which it may be a part. A digital twin also integrates historical data from past machine usage to weave into its digital model.

[0054] In some exemplary embodiments, a digital twin for semiconductor manufacturing equipment utilizes various disconnected data sets generated from first-principles physics-based simulations, approximate empirical methods (reduced-order models, approximate models, etc.), experimental data, and other data sources (e.g., chamber sensors). Additionally, the digital twin employs ML techniques such as statistical modeling, deep neural networks, recurrent neural networks, and convolutional neural networks to generate physics-constrained, data-driven, surrogate hybrid models that are predictive, high-fidelity, and accurate.

[0055] Digital twins can be applied, for example, to analyze the impact of etch chamber design on wafer etch uniformity. For example, chamber flow simulations provide information on fluid flow fields, pressure fields, concentration gradients, etc. Plasma simulations provide charged species densities and fluxes to the wafer, as well as plasma parameters in the process window of interest.

[0056] Experimental data provides partial sensor information that can be correlated to upstream simulation data and inputs, including output variables, which are on-wafer uniformity on blanket wafers and / or feature profile characteristics across the wafer (e.g., depth, critical dimension (CD), tilt, selectivity, mask loss). Individually, high-fidelity simulations are not expected to exactly match on-wafer experimental performance metrics, and experimental data alone does not provide chamber physics such as flow fields, plasma properties, etc. Hybrid model solutions can help provide insights from these decoupled but related data sets to aid product design, engineering, and development.

[0057] The system hybrid model may include separate sub-models covering different aspects of the equipment, which may interact with each other as described below with reference to FIG. 4A.

[0058] 3, in some exemplary embodiments, the sub-models include a chamber model 302, a plasma model 304, a sheath model 306, a wafer feature model 308, an atomistic model 310, and an electronic state model 312. Each model uses its respective features and training data to predict its respective behavior. Further details on the construction of the models are provided below with reference to FIG. 5, and further details on each of the features are provided below with reference to FIGS. 9-10. Additionally, each of the sub-models may also be split into separate models.

[0059] The chamber model 302 is for predicting chamber related phenomena such as chamber geometry, flow rates, thermal information, structural information, and electromagnetics, etc. The plasma model 304 is for predicting plasma performance such as electromagnetic fields, plasma chemistry, and reaction data.

[0060] The sheath model 306 is for predicting sheath data such as RF voltage, electromagnetic field, reaction cross section, and reaction path, etc. The wafer feature model 308 is for predicting wafer features including layout (e.g., design layout, mask layers) and chemistry (e.g., on-wafer flux, material properties).

[0061] The atomistic model 310 predicts phenomena at the atomic level, such as lattice structure, species distribution, and atomic dynamics including diffusion coefficients. The electronic state model 312 is for predicting states for energy calculations, such as energy states and cluster configurations.

[0062] The presented embodiments provide the advantage of determining constitutive relationships between different models generated in isolation for a specific problem on a given system or process. These separate data sets are then combined to build a system model that can be used for design and process optimization in semiconductor manufacturing.

[0063] These models encompass different spatial scales (e.g., on the order of meters to nanometers) and time scales (e.g., from transient processes such as pulsed systems to steady-state conditions under equilibrium). System models result in cost savings (e.g., reduced chemical use, code iterations) as well as development time savings (e.g., reducing design time for new reactors and processes from 18-24 months to less than a year) because systems can be designed and tested much faster than with previous methodologies.

[0064] 4A illustrates interactions between multiple machine learning models for predicting process behavior, according to some exemplary embodiments. A system model 402 comprises multiple ML models ML1-ML13 at different levels of prediction. Each level includes one or more ML models that can interact with each other and with ML models in other layers. In some exemplary embodiments, the models include behavioral models and physics-based models.

[0065] The model includes inputs 404 (e.g., pressure, temperature, flow) and generates outputs related to behavior. The output of the system model 402 is the system behavior 408. The inputs 404 may also include metrology 406 data obtained from sensors. Additionally, the inputs 404 may also include metrology 406 data from an incoming substrate prior to application of the process being modeled. The metrology 406 provides experimental measurements and includes items such as layer thickness, resistivity, film property data, etc. Image analysis may be used to examine experimental results, although other types of metrology 406 may be used.

[0066] Metrology 406 includes one or more of imaging techniques (e.g., scanning electron microscopy (SEM)), transmission electron microscopy (TEM)), typical thickness measurements (e.g., X-ray fluorescence (XRF), ellipsometry), sheet resistance, surface resistivity, stress measurements, and other analytical methods used to determine layer thickness, composition, grain orientation, etc. These other analytical methods include one or more of X-ray diffraction (XRD), X-ray reflectivity (XRR), X-ray photoelectron spectroscopy (XPS), precessional electron diffraction (PED), electron energy loss spectroscopy (EELS), energy dispersive X-ray spectroscopy (EDS), secondary ion mass spectrometry (SIMS), etc.

[0067] In some embodiments, the raw metrology data 406 may be conditioned by a dedicated ML algorithm for each metrology source, or in some embodiments, by an ML algorithm that takes input from multiple metrology sources. Metrology sources in the semiconductor industry can generate a huge amount of data, and distinguishing useful data from individual sources is often a difficult problem. The difficulty grows exponentially when data from many sources must be considered together. Tasking a dedicated ML algorithm to tune the incoming metrology data allows for the construction of a more efficient system model 402. The output of each metrology tuning algorithm is a ROM, and each component ML in the system model can select only the output parameters that are relevant to its own operation.

[0068] In some exemplary embodiments, the metrology includes time series data, which includes sensor measurements taken over time for a given parameter, such as how the pressure in a chamber changes over time during a manufacturing process.

[0069] The inputs 404 can include controls for pressure, flow, power, temperature, etc., all of which affect the behavior of the chamber model 302. At the chamber model 302 level, there are flow fields, ion density, neutron density, etc. Then there is the plasma model 304 level, which includes items such as electric fields, current density, chemistry, etc. When there is no plasma, different species are present in the chamber. Additionally, when there is power applied to the bias, there is a sheath when there is plasma deposition, and controls are used to modify the redistribution function.

[0070] Additionally, there are levels of wafer feature models 308 at which wafer features become observable. Typically, a model is provided for a single level, but some models may include multiple levels. Each model takes into account certain features to obtain its respective predictions. The output of a model can be used by other models operating at the same level or other levels.

[0071] During chamber design, various options can be explored (e.g., changing the distribution of holes in the showerhead, the gap between the wafer and the showerhead), with the goal being to achieve a particular metric on the wafer, such as having a uniform profile. The designer develops the chamber configuration (e.g., chamber knobs, geometry) and recipe, and the system model 402 makes predictions of the system behavior 408, including metrics on the wafer performance. The system model 402 can identify not only the performance on the wafer, but also information about what is happening in the chamber (e.g., plasma, temperature is too high or too low, etc.).

[0072] If performance is not satisfactory, the designer can change certain parameters, such as changing one or more variables at a time (e.g., pressure, chemicals, timing), and repeat the process to see the effect. However, with an ML model, this process is quick (minutes or hours) instead of waiting weeks or months for results. The system model 402 can evaluate the impact on performance when making changes to the hardware or process recipe.

[0073] One advantage of combining models is to find correlations between multiple models generated in isolation for a particular problem on a given system / subsystem / process, and combine these separate data sets to build a system model 402 that can be used for design and process optimization in the features of interest.

[0074] Using such techniques, it is possible to build models from different spatial scales (from chambers ~m order to features ~nm order) and time scales (from transient processes such as pulsed systems to steady state conditions under equilibrium). This allows for a better understanding of the impact on the final desired state (e.g., process fill) based on the design conditions. This could not be done accurately in complex deposition chambers because the path from chamber level simulation to on-wafer surface dynamics is not direct. Using a system model 402 saves wafers and chemicals from a process perspective and minimizes costly design iterations.

[0075] 4B is a table 412 showing examples of modeling algorithms at different levels, according to some exemplary embodiments. The different areas include etching / atomic layer etching, electrofill, plasma-enhanced (PE) chemical vapor deposition (CVD) and atomic layer deposition (ALD), cleaning and stripping, physical vapor deposition, and chemical / mechanical polishing. Although some of the embodiments are described with respect to etching, the same principles may be applied to other areas.

[0076] 5 illustrates training and use of a machine learning program, according to some exemplary embodiments. In some exemplary embodiments, a machine learning program (MLP), also referred to as a machine learning algorithm or tool, is utilized to perform search-related operations, such as a job search.

[0077] Machine learning (ML) is an application that provides computer systems with the ability to perform tasks without being explicitly programmed by making inferences based on patterns found in the analysis of data. Machine learning explores the study and construction of algorithms, also referred to herein as tools, that can learn from existing data and make predictions about new data. Such machine learning algorithms operate by constructing ML models 510 from exemplary training data 506 to make data-driven predictions or decisions, represented as outputs or evaluations 514. Although exemplary embodiments are presented with respect to some machine learning tools, the principles presented herein may be applied to other machine learning tools.

[0078] Data representation refers to a way of organizing data for storage on a computer system, including structures for identified features and their values. In ML, it is typical to represent data in vectors or matrices of two or more dimensions. When dealing with large amounts of data and many features, data representation is important so that training can identify correlations in the data.

[0079] There are two common modes of ML: supervised ML and unsupervised ML. Supervised ML uses prior knowledge (e.g., examples correlating inputs to outputs or outcomes) to learn the relationships between inputs and outputs. The goal of supervised ML is to learn a function that best approximates the relationships between training inputs and outputs given some training data, so that an ML model can implement the same relationships given an input to produce a corresponding output. Unsupervised ML is the training of an ML algorithm using information that is not classified or labeled, allowing the algorithm to act on that information without guidance. Unsupervised ML is useful for exploratory analysis because it can automatically identify structures in the data.

[0080] Common tasks for supervised ML are classification and regression problems. Classification problems, also called categorization problems, aim to classify an item into one of several categorical values ​​(e.g., is this object an apple or an orange?). Regression algorithms aim to quantify some item (e.g., by providing a score to the value of some inputs). Some examples of commonly used supervised ML algorithms are logistic regression (LR), naive Bayes, random forest (RF), neural networks (NN), deep neural networks (DNN), matrix factorization, and support vector machines (SVM).

[0081] Some common tasks for unsupervised ML include clustering, representation learning, and density estimation. Some examples of commonly used unsupervised ML algorithms are K-means clustering, principal component analysis, and autoencoders.

[0082] Training data 506 includes example values ​​for features 502. In some example embodiments, training data 506 includes labeled data having example values ​​for features 502 and labels indicating results such as plasma properties, gas density, gas flow, etch rate, etch uniformity, etc.

[0083] In some demonstrative embodiments, training data 506 is obtained by performing experiments on semiconductor manufacturing equipment, and the data obtained from the experiments is used for ML training 508. Additionally, training data 506 may also be obtained by performing simulations (e.g., physics-based simulations, behavior-based simulations), and the results from the simulations (e.g., etch uniformity on wafers, uniform deposition on wafers) are used for ML training 508.

[0084] Machine learning algorithms utilize training data 506 to find correlations between identified features 502 that influence the outcome. Features 502 are individual, measurable properties of the phenomenon being observed. The concept of features is related to the concept of explanatory variables used in statistical techniques such as linear regression. Selecting features that are informative, discriminatory, and independent is important for the effective operation of ML in pattern recognition, classification, and regression. Features may be of different types, such as numerical features, strings, and graphs. Further details about the features 502 used by the ML model 510 are provided below with reference to Figures 9 and 10.

[0085] During training 508, the ML algorithm analyzes training data 506 based on the identified features 502 and configuration parameters 504 defined for training 508. The result of training 508 is an ML model 510 that can take inputs to generate an evaluation.

[0086] In some embodiments, the exemplary ML model provides estimates at different levels of the semiconductor manufacturing process, as described above with reference to FIGS.

[0087] Training an ML algorithm involves analyzing large amounts of data (e.g., gigabytes to a terabyte or more) to find data correlations. The ML algorithm utilizes training data 506 to find correlations between identified features 502 that affect outcomes or evaluations 514. In some exemplary embodiments, the training data 506 includes labeled data, which is known data about one or more identified features 502 and one or more outcomes.

[0088] ML algorithms typically explore many possible functions and parameters before finding what they identify as the best correlations in the data, and therefore training can require significant computing resources and time.

[0089] Many ML algorithms include configuration parameters 504, and the more complex the ML algorithm, the more parameters are available to the user. The configuration parameters 504 define the variables of the ML algorithm when searching for the best ML model. Training parameters include model parameters and hyperparameters. Model parameters are learned from training data, while hyperparameters are not learned from training data, but are instead provided to the ML algorithm.

[0090] Some examples of model parameters include maximum model size, maximum number of passes through training data, data shuffle type, regression coefficients, split locations in decision trees, etc. Hyperparameters can include the number of hidden layers in the neural network, the number of hidden nodes in each layer, learning rate (possibly using various adaptation schemes for the learning rate), regularization parameters, type of nonlinear activation function, etc. Finding the correct (or best) set of hyperparameters can be a very time-consuming task requiring large amounts of computer resources.

[0091] When the ML model 510 is used to perform the evaluation, inputs 512 are provided to the ML model 510, which generates as output an evaluation 514. For example, when analyzing gas flow distribution for a showerhead, the output may indicate a time series of gas density across the surface of the wafer.

[0092] For example, an ML model 510 for detecting complete coverage when filling deep trenches on a wafer can provide an estimate indicating the level of coverage for the trench, with 100% meaning coverage and anything less meaning there are voids in the fill.

[0093] Feature extraction is the process of reducing the amount of resources required to describe large data sets. When performing analysis of complex data, one of the major problems stems from the number of variables involved. Analyses with a large number of variables generally require large amounts of memory and computational power, and classification algorithms may overfit to the training samples and generalize poorly to new samples. Feature extraction involves constructing combinations of variables to avoid these problems with large data sets while still describing the data with sufficient accuracy for the desired purpose.

[0094] In some exemplary embodiments, feature extraction starts from an initial set of measurement data and builds derived values ​​(features) that are intended to be informative and non-redundant, facilitating subsequent learning and generalization steps. Additionally, feature extraction is associated with dimensionality reduction, such as reducing large vectors (possibly with very sparse data) into smaller vectors that capture the same or similar amount of information.

[0095] In some exemplary embodiments, feature extraction involves obtaining feature values ​​(as described in FIGS. 9 and 10) and optionally converting those values ​​into vectors or matrices. In some cases, the vectors or matrices are processed to reduce their dimensionality in order to provide information to the ML model 510 in a more compact manner. Additionally, data from multiple features can be combined, such as by concatenating vectors from multiple inputs into a single vector.

[0096] 6 illustrates a process for plasma reduced order model (ROM) simulation, according to some exemplary embodiments. Although the exemplary embodiments are described with reference to plasma properties, the same principles may be applied to other areas, such as thermal processes. In some exemplary embodiments, the goal is to use physics-based simulation, behavior-based simulation, wafer data log (WDL) (including time series information generated from sensors), and ML models to derive a reduced order model (ROM), also referred to herein as a digital twin, for a semiconductor manufacturing chamber.

[0097] ROMs are simplified versions of complex models that capture the behavior of a system, allowing engineers to quickly study the dominant effects in the system using minimal computational resources. ROMs enable engineers to achieve shorter design cycles and produce higher quality products.

[0098] The chamber knobs 602 are configurable parameters that are used as inputs to the simulation 604. The chamber knobs 602 include any of the configurable parameters for a manufacturing process, such as pressure, flow rate, transformer coupled plasma (TCP), bias, transformer coupled capacitive tuning (TCCT), chemical information, etc.

[0099] A design of experiments (DOE) matrix is ​​created for the simulation 604. DOE is a branch of applied statistics that deals with the planning, conducting, analyzing, and interpreting controlled trials to evaluate factors that control the value of a parameter or group of parameters. DOE is a powerful data collection and analysis tool that can be used in a variety of experimental situations.

[0100] In some exemplary embodiments, the simulation 604 is a physics-based simulation, although other embodiments may utilize other types of simulations, such as behavior-based simulations. Typically, multiple simulations are performed, and the result is an output 606 that includes physical quantity data as well as derived values ​​obtained in the simulation 604. For example, the simulation may generate predictions including flow fields, ion densities, energy distributions, species distributions, and the like.

[0101] The results of the simulation 604 are used as training data for the ML training 508 to create a ROM 608. The ROM 608 receives the configuration of the chamber knobs 602 as input and generates predictions regarding the performance of a particular design and process.

[0102] For example, two designs are considered for a showerhead. Each design may be defined, for example, by its respective computer-aided design (CAD) geometry (e.g., number and distribution of holes in the showerhead) and gas flows. The two designs are input into ROM 608 and results are obtained. The performance parameters considered for showerhead improvement are then compared to see which showerhead design is better.

[0103] Because comparing performance using ROM 608 does not require actual experiments in a chamber, results are obtained very quickly (e.g., within minutes or hours) rather than having to wait days or weeks to get results for each design. The very fast testing of designs allows design engineers to test multiple designs by tweaking one or more parameters at a time until a satisfactory design is found.

[0104] Additionally, the ROM 608 can be verified (e.g., its accuracy in formulating the predictions can be checked) by comparing the results generated by the simulation 604 with the predictions 610 generated by the ROM 608 when using the same configuration of the chamber knobs 602. In this manner, the ROM 608 can be continually enhanced until the predictions 610 are substantially similar to the results of the simulation 604. It should be noted that, in general, the ROM 608 is much faster than a simulation 604 (e.g., a physics-based simulation), which requires solving complex mathematical equations with many involved variables. Thus, having an accurate ROM 608 greatly accelerates the development process.

[0105] FIG. 7 illustrates a process for plasma ROM using simulations and experiments, according to some exemplary embodiments. Although the exemplary embodiments are described with reference to plasma properties, the same principles may be applied to other areas, such as thermal processes. The method described above with reference to FIG. 6 can be enhanced by using actual experiments 702 performed using a semiconductor manufacturing tool. Of course, using actual experiments is more expensive than simulations 604 and requires more time to set up and run. Using experimental data improves the quality of the predicted results.

[0106] The experimental output 704 resulting from the experiment 702 may also be used as additional data for ML training 508 to obtain the ROM 608. Additionally, data from the sensors 706 obtained during the experiment 702 can be used for training and as input to the ROM 608. The ROM 608 then makes predictions 710 based on the chamber knobs 602 used as input. Thus, the ROM 608 benefits from the training data 506 derived from the experiment 702 and the simulation 604. Additionally, the data from the experiment 702 is validated 807 with the output of the predictions 710.

[0107] Similar to simulation 604, results from experiments can be used to validate predictions 710 of ROM 608. Based on the validation, ROM 608 can be fine-tuned, such as by changing hyperparameters for the configurations used during ML training 508.

[0108] For example, a study of the impact of a design on film non-uniformity across a wafer would include simulation data with information on fluid flow fields, pressure fields, concentration gradients, etc. The experimental data would include upstream simulation data and partial sensor information that can be correlated to inputs, but would primarily include output variables, which would be the film non-uniformity or step coverage variation across a 300 mm wafer.

[0109] In some implementations, a behavioral model based on process understanding can be used. Other data inputs such as chamber sensor information and process recipe conditions can be provided to further enhance the model. The generated model is part of an existing data set calculated for a different purpose, and the model can be trained online based on new experimental and simulation data. By itself, high-fidelity simulations cannot match on-wafer experimental performance such as non-uniformity matching, and by itself, experimental data cannot understand the physical properties of the system such as flow fields, fluxes, etc. The use of ROM 608 assists in learning from these decoupled data sets and aids in design engineering for future systems.

[0110] Additionally, other applications include design optimization of thermal fluid systems with conjugate heat transfer such as found in high temperature showerhead / pedestal applications, mass transport and bias optimization in wet processing chambers, plasma driven process systems, or other atomic layer deposition applications.

[0111] 8 illustrates some of the inputs to the ML model, according to some example embodiments. Several inputs are used as data for the algorithm 802 used for system modeling. The inputs include experimental data 804 obtained from recipe information and on-wafer results, experimental data 806 resulting from hardware testing, chamber sensor data 808 obtained during semiconductor manufacturing operations, chamber scale simulations 810, feature scale simulations 812, atomistic / quantum simulations 814, behavior-based models at the feature scale 816, and reduced order models 608.

[0112] The result is a physically constrained ML model 818 that predicts equipment and process performance. The physically constrained ML model 818 allows designers to set controls on the chamber to get the desired results, which means greater predictability over wafer properties and reduced development time.

[0113] 9 and 10 show tables containing some exemplary features used by ML models. The first column indicates the scale (e.g., the level of the hierarchy described above with reference to FIG. 3), the second column indicates the phenomenon being observed, the third column represents the input, and the fourth column represents the output. Each row represents a respective ML model.

[0114] The first model is for the geometry of the chamber with inputs including the design dimensions of the chamber (e.g., chamber diameter, wafer diameter). The output includes the specific geometry design for the chamber, which can be in the form of a CAD file or some other geometry description such as a mesh of the geometry shape.

[0115] The following model is for flow in a chamber. Inputs include at least flow rates, chamber pressure, chemistry, and flow transients. Outputs include at least pressure and velocity fields, species concentrations throughout the chamber, and diffusive fluxes.

[0116] The following model is for a thermal study in a chamber, where the inputs include at least heat flux, heat transfer coefficient, thermal conductivity, heat capacity, and contact area. The outputs are temperature information throughout the chamber, heat flux (e.g., radiation, conduction, and convection), and heat loss.

[0117] The following model is for a structural analysis of a chamber, where the inputs include at least structural loads, reactions, tolerances, temperatures, gradients, and pressures. The outputs include structural stresses, strains, deflections, creep rates, fatigue, and elasticity / plasticity.

[0118] The final model for the chamber is for electromagnetic estimation within the chamber, with inputs including at least voltage, current, frequency, inductance, capacitance, and impedance, and outputs including electric field, magnetic field B, induced current density, and power density.

[0119] The plasma model is for analyzing plasma behavior, and inputs include at least electric field, magnetic field B, current density, chemicals, reaction cross sections, reaction paths, material properties (e.g., conductivity, dielectric constant, emission coefficient), RF frequency, RF voltage, and RF bias. Outputs include charged species density and flux, ambipolar field, electron temperature, electron energy distribution function (EEDF), ion energy angular distribution (IEAD), on-wafer flux, charge density (surface and volume) source, and species loss terms.

[0120] As mentioned above, some outputs from one model can be used as inputs to other models, for example, outputs from the chamber electromagnetic model can be used as inputs to the plasma behavior model (e.g., electric field, magnetic field B, current density).

[0121] The first model in Figure 10 is for analyzing the performance of the sheath, where the inputs include at least the RF voltage, the electric field, the source terms, the reaction collision cross sections, and the reaction paths. The outputs include the flux on the surface, the ion energy and angular distributions, the conduction and displacement currents, the ion transit times, and the charge density.

[0122] The next model is for the layout of wafer features, where the input includes at least the design layout, mask layers, and initial steps of the application. The output includes a geometric description of the feature, such as a CAD file or a mesh configuration.

[0123] The next model is for the feature chemistry, and the inputs include at least on-wafer fluxes, material properties, reaction pathways, reaction rates, ion angular yields, etch thresholds, sticking coefficients, and accommodation coefficients. The outputs include geometry evolution and front tracking, species distribution within the feature, ion energy and angle distribution within the feature.

[0124] The following model is for dynamics at the atomistic level, where inputs include lattice structure, material IDs, and interatomic potentials. Outputs include lattice layout, species distributions, radial distribution functions, diffusion coefficients, and reaction kinetics.

[0125] The final model in the table is for the electronic state and predicts statistics for energy calculations. Inputs include at least material / species ID, cluster configuration, and electronic structure information. Outputs include energy states, cluster configuration, reaction energetics, reaction paths, bond angles, and bond lengths.

[0126] 11 shows a design of a showerhead 1102 using ML, according to some example embodiments. For example, the showerhead 1102 is for use in a low fluorine tungsten reactor, but can also be used in other types of reactors.

[0127] In one example, the showerhead design did not meet the requirements for fill gap, WF6 gas consumption, and Rs NU. Lower gas consumption reduces the cost of consumables (CoC) per wafer. In addition, the gap from the pedestal to the showerhead could not be accurately set by the Automated Gap System (AGS) at high temperatures such as 430 °C. In addition, the planarity from the pedestal to the showerhead could not be accurately set using AGS wafers because AGS wafers are not measured at the outer edge of the pedestal. This leads to tool-to-tool variations and process impacts. An improved design of the showerhead was needed.

[0128] The showerhead 1102 presented thermal instability. The design team tried to manually change the variables of the showerhead 1102, but the process was slow and tedious, requiring a lot of effort and long development time. DOE was used to learn from simulations, which required a long time for calculations. The design process was improved using simulations, but it was still time consuming, and the results of the actual experiments did not really match the results from the simulations. This is why an ML model for the showerhead can speed up the design process. Also, because the ML model can be improved as additional data becomes available, the accuracy of the ML model can improve over time, improving and accelerating the design of the showerhead 1102.

[0129] Features for the showerhead ML are defined to cover multiple facets of the showerhead 1102 (arrows in the figure showing some example areas) and include one or more of the following:

[0130] - contact conductance between the plates;

[0131] - contact conductance, gap conductance, and radiation between the manifold and the backplate;

[0132] - contact conductance, gap conductance, and radiation between the edge ring 1110 and the manifold;

[0133] -Gap conductance and radiation between the pedestal ring and the showerhead edge;

[0134] -Contact conductance and gap conductance between the showerhead backplate and the manifold;

[0135] - contact conductance, gap conductance, and radiation between the pedestal range and the pedestal;

[0136] -Radiance and gap conductance between the wafer and showerhead and between the wafer and pedestal;

[0137] -Gap conduction in the showerhead cavity;

[0138] - Radiation to the outer surface of the pedestal and the surrounding area relative to the pedestal ring;

[0139] -Gap conduction and radiation between the surface of the showerhead and the edge ring;

[0140] - Radiation to the surrounding area against the outer surface of the edge ring; and

[0141] - Radiation to the surrounding area against the outer surface of the manifold.

[0142] Additionally, zones, such as the illustrated zones Z1-Z4, are defined within the showerhead, and parameters associated with each zone are used as characteristics, such as temperature or conductivity over the zone.

[0143] Multiple experiments are performed varying different parameters as inputs. The inputs may include the number of holes in the showerhead, the distribution of the holes, different types of gases used, the pressure of the gas, the delivery cycle, etc. The results, along with the inputs, are used as training data for the showerhead model. The results may include the thickness of the wafer across its surface.

[0144] The showerhead model is then used to estimate respective outputs regarding the showerhead's performance based on various inputs, which can include the measured uniformity across the wafer as well as metrology data obtained from sensors in the chamber.

[0145] In this way, the showerhead design team can quickly model different showerhead configurations and determine which configuration will work best for a given product requirement.

[0146] The resulting showerhead 1102, designed with the aid of the showerhead model, has a reduced showerhead internal cavity volume, a reduced pedestal-to-showerhead volume, a smaller pedestal-to-showerhead gap, and a different faceplate hole pattern, which reduces gas consumption while meeting the required performance parameters such as uniformity, temperature requirements, etc.

[0147] In some exemplary embodiments, the showerhead gap and wafer distance measurements were taken using a laser mounted on the showerhead facing the atmosphere side. The laser light passed through a vacuum sealed sapphire window. The laser light was then reflected off the pedestal surface and the timing was used to measure the distance.

[0148] 12 is a flowchart of a method 1200 for predicting performance of semiconductor manufacturing equipment operations, according to some example embodiments. Although the various operations of this flowchart are presented and described in sequence, those skilled in the art will understand that some or all of the operations may be performed in a different order, combined or omitted, or performed in parallel.

[0149] In operation 1202, a plurality of machine learning (ML) models are obtained, each model associated with predicting a performance metric for the operation of a semiconductor manufacturing tool, and each ML model utilizes a plurality of features that define an input for the ML model.

[0150] From operation 1202, the method 1200 moves to operation 1204 for receiving, by the one or more processors, a process definition for manufacturing the product at a semiconductor manufacturing tool.

[0151] Further, from operation 1204, the method 1200 moves to operation 1206, where the one or more processors utilize one or more ML models from the plurality of ML models to estimate the performance of a process definition used in a semiconductor manufacturing tool.

[0152] From operation 1206, the method 1200 moves to operation 1208, where results indicative of an estimate of the performance of the process are presented on a display.

[0153] In one example, creating a machine learning (ML) model from multiple ML models includes obtaining training data for the ML model, where the training data provides feature values ​​for the ML model, and training an ML algorithm to obtain the ML model.

[0154] In one example, obtaining training data for the ML model includes conducting an experiment at a semiconductor manufacturing tool, measuring feature values ​​for the experiment, and using the measured values ​​for the training data.

[0155] In one example, obtaining training data for the ML model includes obtaining the training data by performing a physics-based simulation of the operation of a semiconductor manufacturing tool.

[0156] In one example, the multiple ML models include a chamber model, a plasma model, a sheath model, a wafer feature model, an atomistic model, and an electronic state model.

[0157] In one example, the chamber ML model is for estimating the geometry of a chamber in a semiconductor manufacturing tool with inputs including design dimensions and outputs including a definition of the chamber's geometry.

[0158] In one example, the Plasma ML model is for analyzing plasma behavior with inputs including one or more of electric field, magnetic field B, current density, chemicals, reaction cross sections, reaction pathways, material properties, RF frequency, RF voltage, and RF bias, and outputs of the Plasma ML model include one or more of charged species densities and fluxes, ambipolar fields, electron temperature, electron energy distribution function (EEDF), ion energy angular distribution (IEAD), on-wafer flux, charge density (surface and volume) sources, and species loss terms.

[0159] In one example, the Sheath ML model is for analyzing the performance of a sheath using inputs including one or more of radio frequency (RF) voltage, electric field, source terms, reaction collision cross sections, and reaction paths, and outputs of the Sheath ML model include one or more of flux, ion energy and angular distribution, conduction and displacement currents, ion transit times, and charge density on the wafer surface.

[0160] In one example, the wafer feature ML model is for analyzing the layout of a wafer feature using input including one or more of a design layout, a mask layer, and an initial step, and the output of the wafer feature ML model includes a geometric description of the wafer feature.

[0161] In one example, the wafer chemical ML model is for analyzing the chemistry of a wafer feature using inputs including one or more of on-wafer fluxes, material properties, reaction pathways, reaction rates, ion angular yields, etch thresholds, sticking coefficients, and accommodation coefficients, and outputs of the wafer chemical ML model include one or more of geometry evolution and front tracking, distribution of species within the wafer feature, and distribution of ion energy and angles within the wafer feature.

[0162] FIG. 13 is a block diagram illustrating an example of a machine 1300 that can perform or control one or more exemplary process embodiments described herein. In alternative embodiments, the machine 1300 may operate as a standalone device or may be connected (e.g., networked) to other machines. In a network deployment, the machine 1300 can operate in the capacity of a server machine, a client machine, or both, in a server-client network environment. In one example, the machine 1300 can operate as a peer machine in a peer-to-peer (P2P) (or other distributed) network environment. Furthermore, although only a single machine 1300 is shown, the term "machine" should also be construed to include any collection of machines that individually or jointly execute a set (or sets) of instructions to perform any one or more of the methodologies discussed herein, such as via cloud computing, software as a service (SaaS), or other computer cluster configurations.

[0163] The examples described herein may include or be operated by logic, some components, or mechanisms. A circuit set is a collection of circuits implemented in a tangible entity that includes hardware (e.g., simple circuits, gates, logic, etc.). The membership of a circuit set can flexibly accommodate the passage of time and the variability of the underlying hardware. A circuit set includes members that, alone or in combination, can perform a particular operation when operated. In one example, the hardware of a circuit set may be immutably designed (e.g., hardwired) to perform a particular operation. In one example, the hardware of a circuit set may include variably connected physical components (e.g., execution units, transistors, simple circuits) that include a computer-readable medium that has been physically modified (e.g., magnetically, electrically, by a movable arrangement of immutable mass particles, etc.) to encode instructions for a particular operation. In connecting the physical components, the underlying electrical properties of the hardware components are changed (e.g., from insulator to conductor or vice versa). The instructions enable the embedded hardware (e.g., execution units or loading mechanisms) to create the members of the circuit set in the hardware through the variable connections to perform some of the particular operations when operated. Thus, the computer-readable medium is communicatively coupled to other components of the circuit set when the device is operating. In one example, any of the physical components may be used in multiple members of multiple circuit sets. For example, during operation, an execution unit may be used in a first circuit of a first circuit set at one time and reused at another time by a second circuit in the first circuit set or by a third circuit in the second circuit set.

[0164] The machine (e.g., computer system) 1300 may include a hardware processor 1302 (e.g., a central processing unit (CPU), a hardware processor core, or any combination thereof), a graphics processing unit (GPU) 1303, a main memory 1304, and a static memory 1306, some or all of which may communicate with each other via an interlink (e.g., a bus) 1308. The machine 1300 may further include a display device 1310, an alphanumeric input device 1312 (e.g., a keyboard), and a user interface (UI) navigation device 1314 (e.g., a mouse). In one example, the display device 1310, the alphanumeric input device 1312, and the UI navigation device 1314 may be touch screen displays. The machine 1300 may further include a mass storage device (e.g., a drive unit) 1316, a signal generating device 1318 (e.g., a speaker), a network interface device 1320, and one or more sensors 1321 (such as a Global Positioning System (GPS) sensor, a compass, an accelerometer, or another sensor). The machine 1300 may include an output controller 1328, such as a serial (e.g., Universal Serial Bus (USB)), parallel, or other wired or wireless (e.g., infrared (IR), near field communication (NFC)) connection, for communicating with or controlling one or more peripheral devices (e.g., printer, card reader).

[0165] The mass storage device 1316 may include a machine-readable medium 1322. The machine-readable medium 1322 stores one or more sets of data structures or instructions 1324 (e.g., software) that embody or are utilized by any one or more of the techniques or functions described herein. The instructions 1324 may also reside, completely or at least partially, within the main memory 1304, within the static memory 1306, within the hardware processor 1302, or within the GPU 1303 during execution by the machine 1300. In one example, the machine-readable medium may be constituted by any one of the hardware processor 1302, the GPU 1303, the main memory 1304, the static memory 1306, or the mass storage device 1316, or any combination thereof.

[0166] Although the machine-readable medium 1322 is depicted as a single medium, the term “machine-readable medium” may include a single medium or multiple media (e.g., a centralized or distributed database, and / or associated caches and servers) configured to store one or more instructions 1324.

[0167] The term "machine-readable medium" may include any medium capable of storing, encoding, or carrying instructions 1324 for execution by the machine 1300 and causing the machine 1300 to perform any one or more of the techniques of this disclosure, or any medium capable of storing, encoding, or carrying data structures used by or related to such instructions 1324. Non-limiting examples of machine-readable media may include solid-state memory, optical media, and magnetic media. In one example, a high-capacity machine-readable medium includes a machine-readable medium 1322 having a plurality of particles having an unchanging (e.g., stationary) mass. Thus, a high-capacity machine-readable medium is not a transitory propagating signal. Specific examples of high-capacity machine-readable media may include non-volatile memory such as semiconductor memory devices (e.g., electrically programmable read-only memory (EPROM), electrically erasable programmable read-only memory (EEPROM)) and flash memory devices, magnetic disks such as internal hard disks and removable disks, magneto-optical disks, and CD-ROM and DVD-ROM disks.

[0168] Further, the instructions 1324 may be transmitted or received over a communications network 1326 via the network interface device 1320 using a transmission medium.

[0169] Throughout this specification, components, operations, or structures described as a single instance may be implemented by multiple instances. Although individual operations of one or more methods are illustrated and described as separate operations, one or more of the individual operations may be performed simultaneously, and each operation need not be performed in the order illustrated. Structures and functions presented as separate components in example configurations may be implemented as combined structures or components. Similarly, structures and functions presented as single components may be implemented as separate components. These and other variations, modifications, additions, and improvements are included within the scope of the subject matter of this specification.

[0170] The illustrated embodiments have been described in sufficient detail to enable those skilled in the art to practice the disclosed teachings. Other embodiments may be utilized and derived from the teachings disclosed herein, such that structural and logical substitutions and changes can be made without departing from the scope of the disclosure. Therefore, this detailed description is not to be taken in a limiting sense, and the scope of the various embodiments is defined only by the appended claims, along with the full scope of equivalents to which such claims are entitled.

[0171] As used herein, the term "or" may be interpreted in an inclusive or exclusive sense. Moreover, multiple instances may be provided for a resource, operation, or structure described herein as a single instance. In addition, boundaries between various resources, operations, modules, engines, and data stores are somewhat arbitrary, and certain operations are illustrated in the context of certain exemplary configurations. Other allocations of functionality are contemplated and may be included within the scope of various embodiments of the disclosure. In general, structures and functions presented as separate resources in an exemplary configuration may be implemented as a combined structure or resource. Similarly, structures and functions presented as a single resource may be implemented as separate resources. These and other variations, modifications, additions, and improvements are included within the scope of the embodiments of the disclosure as expressed by the appended claims. The specification and drawings are therefore to be regarded in an illustrative, and not a restrictive, sense. The present disclosure can also be realized in the following forms. [Form 1] 1. A method comprising: obtaining a plurality of machine learning (ML) models, each of the models relating to predicting a performance metric for operation of a semiconductor manufacturing tool, each of the ML models utilizing a plurality of features that define an input for the ML model; receiving, by one or more processors, a process definition for manufacturing a product at the semiconductor manufacturing tool; utilizing, by the one or more processors, one or more ML models from the plurality of ML models to estimate performance of the process definition for use in the semiconductor manufacturing tool; presenting on a display a result indicative of the estimate of the performance of the production of the product; and A method comprising: [Form 2] 2. The method of claim 1, further comprising: Creating a machine learning (ML) model from the plurality of ML models includes: obtaining training data for the ML model, the training data providing values ​​of the features for the ML model; training an ML algorithm to obtain the ML model; A method comprising: [Form 3] 3. The method of claim 2, further comprising: Obtaining the training data for the ML model includes: conducting an experiment at the semiconductor manufacturing tool; measuring values ​​of said characteristic for said experiment; using the measured values ​​for the training data; A method comprising: [Form 4] 4. The method of claim 3, further comprising: Obtaining the training data for the ML model includes: training a second-order ML model to produce a reduced order model (ROM) for the measurement data; using the output of the secondary ML model as additional training data; The method further comprising: [Form 5] 3. The method of claim 2, further comprising: Obtaining the training data for the ML model includes: Obtaining the training data by performing a physics-based simulation of the operation of the semiconductor manufacturing tool. A method comprising: [Form 6] 2. The method of claim 1, further comprising: The method, wherein the plurality of ML models include a chamber model, a process matrix model, a substrate scale model, a wafer feature model, an atomistic model, and an electronic state model. [Form 7] 2. The method of claim 1, further comprising: A method, wherein a chamber ML model is for estimating the geometry of the chamber in the semiconductor manufacturing tool using inputs including design dimensions and outputs including a definition of the geometry of the chamber. [Form 8] 2. The method of claim 1, further comprising: The process matrix ML model is for analyzing the behavior of the environment of a substrate during processing with inputs including one or more of electric field, magnetic field B, current density, chemicals, reaction cross sections, reaction pathways, material properties, RF frequency, RF voltage, temperature, and RF bias, and outputs of the plasma ML model include one or more of charged species density and flux, ambipolar field, electron temperature, electron energy distribution function (EEDF), ion energy angular distribution (IEAD), on-wafer flux, charge density (surface and volume) source, and species loss or creation terms. [Form 9] 2. The method of claim 1, further comprising: The method includes a substrate-level ML model for analyzing the performance of a sheath close to the substrate using inputs including one or more of radio frequency (RF) voltage, electric field, source terms, reaction collision cross sections, solution concentrations, and reaction paths, and outputs of the sheath ML model include one or more of flux, ion energy and angular distribution, conduction and displacement currents, ion transit times, surface functionalization, and charge density on the wafer surface. [Form 10] 2. The method of claim 1, further comprising: A method, comprising: a wafer feature ML model for analyzing a layout of a wafer feature using input including one or more of a design layout, a mask layer, and an initial step; and an output of the wafer feature ML model including a geometric description of the wafer feature. [Form 11] 2. The method of claim 1, further comprising: The method includes a wafer chemical ML model for analyzing the chemicals of a wafer feature using inputs including one or more of on-wafer fluxes, material properties, reaction pathways, reaction rates, ion angular yields, etch thresholds, sticking coefficients, and accommodation coefficients, and outputs of the wafer chemical ML model include one or more of geometry evolution and front tracking, distribution of species within the wafer feature, and distribution of ion energy and angles within the wafer feature. [Form 12] 1. A system comprising: a memory containing instructions; one or more computer processors, the instructions, when executed by the one or more computer processors, providing the system with: obtaining a plurality of machine learning (ML) models, each model relating to predicting a performance metric for operation of a semiconductor manufacturing tool, each ML model utilizing a plurality of features that define an input for the ML model; receiving a process definition for manufacturing a product at the semiconductor manufacturing tool; utilizing one or more ML models from the plurality of ML models to estimate performance of the process definition for use in the semiconductor manufacturing tool; presenting on a display a result indicative of the estimate of the performance of the production of the product. one or more computer processors for performing operations including A system comprising: [Form 13] 13. The system according to claim 12, Creating a machine learning (ML) model from the plurality of ML models includes: obtaining training data for the ML model, the training data providing values ​​of the features for the ML model; training an ML algorithm to obtain the ML model; Including, the system. [Form 14] 14. The system according to claim 13, Obtaining the training data for the ML model includes: conducting an experiment at the semiconductor manufacturing tool; measuring values ​​of said characteristic for said experiment; using the measured values ​​for the training data; and Obtaining additional training data by performing a physics-based simulation of the semiconductor manufacturing tool; Including, the system. [Form 15] 13. The system according to claim 12, A system, wherein the chamber ML model is for estimating the geometry of the chamber in the semiconductor manufacturing tool using inputs including design dimensions and outputs including a definition of the geometry of the chamber. [Form 16] 13. The system according to claim 12, The process matrix ML model is for analyzing the behavior of the environment of a substrate during processing using inputs including one or more of electric field, magnetic field B, current density, chemicals, reaction cross sections, reaction pathways, material properties, temperature, mass transport, RF frequency, RF voltage, and RF bias, and outputs of said process matrix ML model include one or more of charged species density and flux, ambipolar field, electron temperature, electron energy distribution function (EEDF), ion energy angular distribution (IEAD), on-wafer flux, charge density (surface and volume) source, and species loss or generation terms. [Form 17] When executed by a machine, the machine: obtaining a plurality of machine learning (ML) models, each of the models relating to predicting a performance metric for operation of a semiconductor manufacturing tool, each of the ML models utilizing a plurality of features that define an input for the ML model; receiving a process definition for manufacturing a product at the semiconductor manufacturing tool; utilizing one or more ML models from the plurality of ML models to estimate performance of the process definition for use in the semiconductor manufacturing tool; presenting on a display a result indicative of the estimate of the performance of the production of the product. A machine-readable storage medium for performing operations including: [Form 18] 18. The machine-readable storage medium according to claim 17, Creating a machine learning (ML) model from the plurality of ML models includes: obtaining training data for the ML model, the training data providing values ​​of the features for the ML model; training an ML algorithm to obtain the ML model; 1. A machine-readable storage medium comprising: [Form 19] 19. The machine-readable storage medium of claim 18, Obtaining the training data for the ML model includes: conducting an experiment at the semiconductor manufacturing tool; measuring values ​​of said characteristic for said experiment; using the measured values ​​for the training data; and Obtaining additional training data by performing a physics-based simulation of the semiconductor manufacturing tool; 1. A machine-readable storage medium comprising: [Form 20] 18. The machine-readable storage medium according to claim 17, A machine-readable storage medium, wherein a chamber ML model is for estimating the geometry of the chamber in the semiconductor manufacturing tool with inputs including design dimensions and outputs including a definition of a geometry of the chamber.

Claims

1. 1. A method comprising: obtaining a plurality of machine learning (ML) models, each of the models relating to predicting a performance metric for operation of a semiconductor manufacturing tool, each of the ML models utilizing a plurality of features defining an input for the ML model; receiving, by one or more processors, a process definition for manufacturing a product at the semiconductor manufacturing tool; estimating, by the one or more processors, performance of the process definition for use in the semiconductor manufacturing tool using one or more ML models from the plurality of ML models; presenting on a display a result indicative of the estimate of the performance of the production of the product; and A method comprising:

2. 10. The method of claim 1 further comprising: generating a machine learning (ML) model from the plurality of ML models; Creating a machine learning (ML) model from the plurality of ML models includes: obtaining training data for the ML model, the training data providing values ​​of the features for the ML model; training an ML algorithm to obtain said ML model; A method comprising:

3. 3. The method of claim 2, Obtaining the training data for the ML model comprises: conducting an experiment at the semiconductor manufacturing tool; measuring values ​​of said characteristic for said experiment; using the measured values ​​for the training data; A method comprising:

4. 4. The method of claim 3, Obtaining the training data for the ML model comprises: Training a second-order ML model to produce a reduced order model (ROM) for the measurement data; using the output of the secondary ML model as additional training data; The method further comprising:

5. 3. The method of claim 2, Obtaining the training data for the ML model comprises: Obtaining the training data by performing a physics-based simulation of the operation of the semiconductor manufacturing tool. A method comprising:

6. 2. The method of claim 1 , The method, wherein the plurality of ML models include a chamber model, a process matrix model, a substrate scale model, a wafer feature model, an atomistic model, and an electronic state model.

7. 2. The method of claim 1 , A method, wherein a chamber ML model is for estimating the geometry of the chamber in the semiconductor manufacturing tool using inputs including design dimensions and outputs including a definition of a geometry of the chamber.

8. 2. The method of claim 1 , The process matrix ML model is for analyzing the behavior of the environment of a substrate during processing with inputs including one or more of electric field, magnetic field B, current density, chemicals, reaction cross sections, reaction pathways, material properties, RF frequency, RF voltage, temperature, and RF bias, and outputs of the plasma ML model include one or more of charged species density and flux, ambipolar field, electron temperature, electron energy distribution function (EEDF), ion energy angular distribution (IEAD), on-wafer flux, charge density (surface and volume) source, and species loss or creation terms.

9. 2. The method of claim 1 , The method includes a substrate-level ML model for analyzing the performance of a sheath close to the substrate using inputs including one or more of radio frequency (RF) voltage, electric field, source terms, reaction collision cross sections, solution concentrations, and reaction paths, and outputs of the sheath ML model include one or more of flux, ion energy and angular distribution, conduction and displacement currents, ion transit times, surface functionalization, and charge density on the wafer surface.

10. 2. The method of claim 1 , A method, comprising: a wafer feature ML model for analyzing a layout of a wafer feature using input including one or more of a design layout, a mask layer, and an initial step; and an output of the wafer feature ML model including a geometric description of the wafer feature.

11. 2. The method of claim 1 , A method in which a wafer chemical ML model is for analyzing the chemicals of a wafer feature using inputs including one or more of on-wafer fluxes, material properties, reaction pathways, reaction rates, ion angular yields, etch thresholds, sticking coefficients, and accommodation coefficients, and outputs of the wafer chemical ML model include one or more of geometry evolution and front tracking, distribution of species within the wafer feature, and distribution of ion energy and angles within the wafer feature.

12. 1. A system comprising: a memory containing instructions; one or more computer processors, the instructions, when executed by the one or more computer processors, providing the system with: obtaining a plurality of machine learning (ML) models, each model relating to predicting a performance metric for operation of a semiconductor manufacturing tool, each ML model utilizing a plurality of features defining an input for the ML model; receiving a process definition for manufacturing a product at the semiconductor manufacturing tool; utilizing one or more ML models from the plurality of ML models to estimate performance of the process definition for use in the semiconductor manufacturing tool; presenting on a display a result indicative of the estimate of the performance of the production of the product. one or more computer processors for performing operations including A system comprising:

13. 13. The system of claim 12, further comprising: generating a machine learning (ML) model from the plurality of ML models; Creating a machine learning (ML) model from the plurality of ML models includes: obtaining training data for the ML model, the training data providing values ​​of the features for the ML model; training an ML algorithm to obtain said ML model; Including, the system.

14. 14. The system of claim 13, Obtaining the training data for the ML model comprises: conducting an experiment at the semiconductor manufacturing tool; measuring values ​​of said characteristic for said experiment; using the measured values ​​for the training data; and Obtaining additional training data by performing a physics-based simulation of the semiconductor manufacturing tool; Including, the system.

15. 13. The system of claim 12, A system, wherein the chamber ML model is for estimating the geometry of the chamber in the semiconductor manufacturing tool using inputs including design dimensions and outputs including a definition of the geometry of the chamber.

16. 13. The system of claim 12, The process matrix ML model is for analyzing the behavior of the environment of a substrate during processing with inputs including one or more of electric field, magnetic field B, current density, chemicals, reaction cross sections, reaction pathways, material properties, temperature, mass transport, RF frequency, RF voltage, and RF bias, and outputs of said process matrix ML model include one or more of charged species density and flux, ambipolar field, electron temperature, electron energy distribution function (EEDF), ion energy angular distribution (IEAD), on-wafer flux, charge density (surface and volume) source, and species loss or creation terms.

17. When executed by a machine, the machine: obtaining a plurality of machine learning (ML) models, each of the models relating to predicting a performance metric for operation of a semiconductor manufacturing tool, each of the ML models utilizing a plurality of features defining an input for the ML model; receiving a process definition for manufacturing a product at the semiconductor manufacturing tool; utilizing one or more ML models from the plurality of ML models to estimate performance of the process definition for use in the semiconductor manufacturing tool; presenting on a display a result indicative of the estimate of the performance of the production of the product. A machine-readable storage medium for performing operations including:

18. 20. The machine-readable storage medium of claim 17, causing the machine to further create a machine learning (ML) model from the plurality of ML models; Creating a machine learning (ML) model from the plurality of ML models includes: obtaining training data for the ML model, the training data providing values ​​of the features for the ML model; training an ML algorithm to obtain said ML model; 1. A machine-readable storage medium comprising:

19. 20. The machine-readable storage medium of claim 18, Obtaining the training data for the ML model comprises: conducting an experiment at the semiconductor manufacturing tool; measuring values ​​of said characteristic for said experiment; using the measured values ​​for the training data; and Obtaining additional training data by performing a physics-based simulation of the semiconductor manufacturing tool; 1. A machine-readable storage medium comprising:

20. 20. The machine-readable storage medium of claim 17, A machine-readable storage medium, wherein a chamber ML model is for estimating the geometry of the chamber in the semiconductor manufacturing tool with inputs including design dimensions and outputs including a definition of a geometry of the chamber.

Images