Determining the recipe for manufacturing semiconductors

Abstract

To facilitate recipe definition, methods, systems, and computer programs are presented for determining recipes for semiconductor manufacturing using machine learning (ML). One general aspect includes a method including conducting experiments for processing components, each experiment controlled by a recipe from a set of recipes that specify parameters for manufacturing equipment. The method further includes conducting virtual simulations for processing the components, each simulation controlled by a recipe from the set of recipes. The ML model is obtained by training an ML algorithm using experimental results and virtual results from the virtual simulations. The method further includes receiving specifications for desired processing of the components and creating a new recipe for processing the components using the ML model based on the specifications.

Description

Technical Field

[0001] [Priority Claim] This application claims priority based on U.S. Provisional Patent Application No. 62 / 925,157, filed on October 23, 2019, entitled "Determination of Recipe for Manufacturing Semiconductor". This provisional application is hereby incorporated by reference in its entirety.

[0002] [Related Application] This application is related to U.S. Patent Application No. 16 / 260,870, filed on January 29, 2019, entitled "Fill Process Optimization Using Feature Scale Modeling", and is hereby incorporated by reference in its entirety.

[0003] The subject matter disclosed herein generally relates to methods, systems, and machine-readable storage media for manufacturing semiconductors.

Background Art

[0004] The background art described herein is for generally presenting the content of the present disclosure. The work of the inventors, currently named, is not admitted as prior art to the present disclosure, either explicitly or implicitly, in the scope described in this background art section and aspects of the description that do not fall within the prior art at the time of filing.

[0005] Chemical reaction apparatuses used in the deposition process development of semiconductor chips tend to have many interdependent subsystems (e.g., sensors, actuators, gas supply, power supply, coordination network). These subsystems are independently controlled by process parameters that follow a set of instructions included in the recipe of the control system. The operations of these subsystems collectively determine the output performance in the product wafer.

[0006] The increasing complexity of current process equipment means that the number of system components increases to address this complexity, and consequently, the number of process "knobs" (e.g., adjustable processing parameters) in the system increases. Various system conditions (pressure, temperature, flow rate setpoints, etc.) are factors that play a role in the desired output to the wafer, such as film step coverage, film non-uniformity, and etching depth.

[0007] Setting the correct process parameter values ​​is a critical challenge in the semiconductor equipment industry, and it often takes weeks or even months of process development to obtain a recipe that sets all these process parameters to obtain components that meet the desired criteria. [Overview of the Initiative]

[0008] To facilitate recipe definition, methods, systems, and computer programs for determining recipes for manufacturing semiconductors using machine learning (ML) are presented. One general embodiment includes a method comprising the steps of conducting experiments for processing components, each experiment being controlled by one recipe from a set of recipes that specify the parameters of the manufacturing equipment. This method further includes the steps of conducting virtual simulations for processing components, each simulation being controlled by one recipe from a set of recipes. The ML model is obtained by training an ML algorithm with the experimental results and virtual results from the virtual simulations. This method further includes the steps of receiving the specifications for the desired processing of the components and creating a new recipe for processing the components by the ML model based on those specifications. [Brief explanation of the drawing]

[0009] The various accompanying drawings merely represent exemplary embodiments of the present disclosure and should not be considered to limit its scope.

[0010] [Figure 1]A process for finding semiconductor manufacturing recipes, according to several exemplary embodiments.

[0011] [Figure 2] A diagram illustrating the use of a simulation tool to conduct virtual experiments using recipes, based on several exemplary embodiments.

[0012] [Figure 3] A diagram illustrating the use of machine learning to facilitate recipe definition, using several exemplary embodiments.

[0013] [Figure 4] Some of the features used by machine learning programs in several exemplary embodiments.

[0014] [Figure 5A] A diagram illustrating the training and use of a machine learning program by several exemplary embodiments.

[0015] [Figure 5B] Diagram illustrating the use of machine learning programs with active process control in several exemplary embodiments.

[0016] [Figure 6] An example of a Deposition-Inhibition-Deposition (DID) deposition process using Inhibition-Enhanced Control (ICE) filling, which can be optimized using a behavioral model.

[0017] [Figure 7] Etching chambers according to several exemplary embodiments.

[0018] [Figure 8] A flowchart illustrating a method for determining a recipe for manufacturing a semiconductor, based on several exemplary embodiments.

[0019] [Figure 9]A block diagram showing examples of machines on which embodiments of one or more of the exemplary processes described herein can be implemented or controlled.

Best Mode for Carrying Out the Invention

[0020] Exemplary methods, systems, and computer programs are directed to determining a recipe for manufacturing a semiconductor. The examples are merely illustrative of possible variations. Unless specified, components and functions are optional and may be combined or subdivided, and steps may be different in sequence or may be combined or subdivided. In the following description, for the purpose of explanation, some specific details are set forth to provide a thorough understanding of the exemplary embodiments. However, it will be apparent to those skilled in the art that the subject matter may be practiced without these specific details.

[0021] FIG. 1 shows a process for finding a semiconductor manufacturing recipe according to some exemplary embodiments. Current approaches to recipe development are based on historical learning from subject matter experts on the process engineering team.

[0022] Product requirements 102 are provided to subject matter expert 104 who plans experiments (step 106) based on past experience. An example of product requirements 102 is the required deposition amount for a feature.

[0023] The experiments include recipes created by subject matter expert 104 based on product requirements 102. However, finding the correct recipe can be a difficult process, especially when the product requirements are on the order of 1 nm accuracy.

[0024] The experiments are conducted in a laboratory 108, and the results are measured and compared 110 to the original product requirements 102. Some exemplary result criteria include uniform film thickness and resistivity of the conductor.

[0025] A comparison 112 is performed to determine if the result is suitable to meet product requirement 102. If the result is suitable, a recipe is found 114; otherwise, the process returns to step 106, where an expert fine-tunes the recipe based on the result to improve it to better meet product requirement 102.

[0026] As the number of processing parameters in a system increases, it becomes increasingly difficult to truly isolate and learn the effects of each individual factor and their impact on the desired outcome. Manufacturing tools are complex and have many configurable parameters, such as gas flow, plasma properties, and thermal properties. For example, a typical atomic layer deposition (ALD) process has 100 to 200 adjustable parameters.

[0027] Common experimental techniques such as single-variable testing (SVT) and design of experiments (DOE) (e.g., full factorial models, screening designs, response surface models, mixed models, Taguchi orthogonals) allow us to not only determine the influence of experimental factors on the outcome but also to learn the correlations between different factors. However, these methods require many experiments to determine the influence of various processing parameters. To obtain a near-perfect recipe, process engineers or their teams typically need to conduct many tests, which can take weeks (or months) and increase the cost of wafer development.

[0028] Furthermore, when testing one parameter, there is no optimization for other parameters, and the experiment falls into a minimum value. Therefore, when multiple parameters are specified simultaneously, even experts may be unable to develop the recipe in the right direction.

[0029] The limitations of this method are that it is highly manual, requiring extended tool time to conduct tests and implement the method, and additional time for analysis to make changes for subsequent sets of tests. This results in multiple learning cycles, which can amount to several months in complex cases. Typically, the factors tested do not include all possible variables, and the correlation coefficients of the models indicate the degree of predictability of the models. Blocking and screening designs exist to test more factors in the experimental setup, but designing the experimental setup requires expertise. Ultimately, the results are the same: many tests are needed, which is costly and time-consuming.

[0030] Figure 2 illustrates the use of a simulation tool to conduct virtual experiments using recipes, through several exemplary embodiments. The embodiments presented here describe a method for finding process recipes using machine learning (ML) and the simulation tool 206.

[0031] Process simulation is used to test exemplary recipe 202 in order to minimize the costs incurred in physical testing and related methods, and to shorten process development time (reduce the experiment and learning cycle). An example of a simulation tool is SEMulator3D by Lam Research Corporation, which provides a voxel model of a semiconductor process based on predicted behavior simulated by behavior model 210. However, similar principles may apply to other simulation tools.

[0032] Behavioral model 210 explains the process output based on the analysis formula. While behavioral model 210 qualitatively captures the process, it does not provide instructions for process development in terms of the quantities that define the process parameters.

[0033] For example, consider a deposition process with two surfaces B, one of which is selectively deposited and the other is not. A model measures the total deposition thickness across the surface. For example, some areas have a deposition thickness of 0.72, while others have a thickness of 0.71. A behavioral model investigates the behavior of particles within a chamber, but rather measures the result of that behavior (e.g., deposition thickness).

[0034] For example, an ALD system may have hundreds of parameters 204 that define recipe 202. A non-exclusive list of these parameters 204 includes general tool parameters (e.g., number of ALD cycles, presence or absence of immersion, tool process mode, etc.), fluid parameters (e.g., flow rates of various gases, fluid concentrations, dilution gases, non-reactive gases for station isolation, etc.), chamber adjustment or pre-coat parameters (e.g., pre-coat temperature, pre-coat time, etc.), pressure parameters (e.g., chamber pressure, reservoir pressure, throttle valve angle, precursor ampoule pressure, vacuum clamp pressure, etc.), nucleating chemical parameters (e.g., dose time, dose flow, chemical concentration, cycles, etc.), temperature parameters (e.g., base temperature, ampoule temperature, chamber temperature, showerhead temperature, etc.), ALD time adjustment parameters (e.g., dose time for precursor A, purge time, dose time for precursor B, preheating time, etc.), and various other parameters. These parameters may be used as features of the ML model, as described below with respect to Figure 4.

[0035] In some examples, there are linear dependencies between parameters regarding the predicted output, but in certain cases, there are nonlinear dependencies. Predicting these dependencies and building appropriate models is not straightforward. Furthermore, the methods used today do not involve placement dependencies (i.e., up-process aggregation steps).

[0036] The simulation tool 206 constructs a three-dimensional model of what happens on the substrate when recipe 202 is executed through the process and generates simulation results measured by measurement method 214. Measurement method 214 provides simulation results 212, including items such as layer thickness and resistivity. Image analysis may be used to examine the simulation results 212, but other types of measurement methods 214 may be used.

[0037] Physical model 208 describes the physical behavior on the substrate and is usually based on the first principle, but may also be used empirically with ML and statistical methods based on physics and chemistry.

[0038] Physical model 208 incorporates chamber parameters such as pressure, temperature, and seed flux (number of particles crossing a unit area per second). Physical model 208 analyzes these parameters to predict the behavior of physical particles (e.g., seed flux) that affect the process. For example, the flux value influences the deposition thickness, meaning that higher flux values ​​result in thicker deposition than lower velocity values. These parameters may be used as features of the ML model, as described below with respect to Figure 4.

[0039] In some exemplary embodiments, several correlation methods generate a bridge linking the behavioral model 210 and real-world process recipes via the physical model 208. These correlation methods include multivariate regression, neural networks, decision trees, and support vector machines (SVMs). In some exemplary embodiments, the physical model 208 may be a combination of multiple models representing different aspects of the physical model. Furthermore, in some exemplary embodiments, the behavioral model 210 may be a combination of multiple models, each containing a different behavioral aspect.

[0040] The simulation tool 206 generates simulation results 212 using the behavioral model 210 and the physical model 208. Since no actual experiment was conducted, the simulation results 212 are also called virtual results.

[0041] The simulation result 212 is compared to product requirement 102 to determine whether recipe 202 satisfies product requirement 102. If simulation result 212 satisfies the requirement, the work recipe is found. Otherwise, a new simulation may be performed with a different recipe 202 to continue searching for the correct recipe 202.

[0042] A successful simulation means fewer tests are required, leading to savings in time and cost. For unsuccessful simulations, the results are fed back into the model to improve the accuracy of future predictions.

[0043] Figure 3 illustrates the use of machine learning to facilitate recipe definition in several exemplary embodiments. ML is an application that gives computer systems the ability to perform tasks without being explicitly programmed by inferring based on patterns found in data analysis. ML relies on data from which it can learn in order to infer.

[0044] In some exemplary embodiments, the data for ML includes experimental results 306 from actual experiments 302 performed with semiconductor manufacturing tools, and virtual results 308 from simulations 304. Experiments 302 and simulations 304 may use the same or different recipes 202. This method can be applied to multiple semiconductor manufacturing processes, such as deposition, etching, and cleaning.

[0045] Performing actual experiments 302 is costly and time-consuming. However, performing simulations 304 is much faster and cheaper. Therefore, many simulations 304 may be performed (e.g., 10 to 1000 or more) to obtain many hypothetical results 308 that can be used to train machine learning algorithms. For example, simulations 304 may be performed by changing the values ​​of parameters of interest to enable predictions as to how the parameters of interest can be tuned to create new recipes 316.

[0046] In step 310, the ML algorithm is trained using data from experimental results 306 and virtual results 308. The result of the training in step 310 is an ML model 314 configured to accept desired components 312 (e.g., product requirements) and create a new recipe 316. In some exemplary embodiments, the number of experiments is 10 to 100, but other numbers are also possible. Furthermore, the number of simulations 304 is 100 to 100,000, but other numbers are also possible.

[0047] In some exemplary embodiments, the experimental results 306 are few in number, and the actual experimental results 306 are more accurate than the virtual results 308 obtained from the simulation 304; therefore, the experimental results 306 are given greater weight than the virtual results 308 for training (step 310). Furthermore, other data, such as data obtained from an experimental library, may be used for training. More details about ML are provided below in relation to Figure 5A.

[0048] In the search for the best recipe that satisfies product requirement 102, ML model 314 may be used several times. To improve the accuracy of ML model 314, a new recipe 316 may be used with the data used for experiments (real or virtual) and training the ML algorithm.

[0049] The embodiments described herein relate chamber setpoints and sensor data for a simulator behavior model calibrated based on real-world measurement data (image data, film property data, etc.). In some exemplary embodiments, prior knowledge can be used to determine a new design space, for example, by Bayesian estimation. Furthermore, guidance from physics-based modeling can be used in conjunction with the behavior model to further improve the accuracy of the correlation process between process parameters and model outputs.

[0050] The impact of process parameters on process performance is inferred based on virtual model results. This not only accelerates the understanding of the process based on the model, but also makes it easier for process engineers and technicians to understand the process by associating it with tool parameters, enabling process optimization on the tool. As a result, the presented solution reduces experimental costs and time. Fewer iterations lead to higher process efficiency.

[0051] In some exemplary embodiments, different ML models 314 are used for different semiconductor manufacturing processes. For example, one ML model is created for the deposition process (e.g., using deposition experiments and simulations), another model is created for the etching process, and yet another model is created for substrate cleaning.

[0052] One of the advantages of the ML model 314 is that it can explore not only process parameters but also physical parameters in the search for a new recipe 316.

[0053] One example of an application is deposition using suppression profiles. The behavioral models, calibrated by behavioral adjustment of a deposition-suppression-deposition model or a deposition-etching-deposition model, are described in U.S. Patent Application No. 16 / 260,870, filed January 29, 2019, entitled "Fill Process Optimization Using Feature Scale Modeling," which is incorporated herein by reference. The calibrated behavioral models are associated with experimental variables such as dose time, purge time, flow rates of various suppression chemicals, system pressure, wafer temperature, and molecular migration per recessed feature shape. These variables from a test set (a small sample in which the experiment is conducted) are used to train an ML model that bridges the gap between process results and simulated behavior, based on one or more ML models calibrated using optimization methods such as gradient descent. Key parameters that significantly impact the process are extracted and experimentally validated to obtain an ideal recipe based on the sample space in which the model is calibrated. This model incorporates structural geometry that is often overlooked by other analytical methods. Details of this process are provided below in relation to Figure 6.

[0054] Another deposition example concerns the packing and roughness control of 3D NAND WLs (word lines). Here, the reactants in the ALD system need to travel not only the full length of the high aspect ratio (HAR) structure but also laterally into the WL interior. In addition to molecular transport challenges, these processes are typically performed at extremely high speeds to match the throughput expected by the customer (less than 1 second per cycle for the precursor dose-purge-reducing agent dose-purge process).

[0055] Roughness due to particle growth can lead to pinch-off and void formation, which can be mitigated by suppressing growth in specific regions. The model predicts profile behavior based on reaction-diffusion models in the vertical and lateral structures. Experimental data allows for model calibration and correlation with tool parameters such as precursor dose time, chemical purge time, reducing agent dose time, inhibitor molecule dose time, system pressure, wafer temperature, and the geometry of the structure modeled in the simulator. Based on the sample set used for data training, the ML model is used to correlate the results of the optimal solution (e.g., void-free film, low-roughness film, possibility of post-deposition etch-back, etc.) based on process parameters.

[0056] Figure 4 shows some of the features 402 used in machine learning programs in several exemplary embodiments. Features are used by ML algorithms to represent data. A feature is an individual measurable characteristic of an observed event. The concept of a feature is related to the concept of explanatory variables used in statistical methods such as linear regression. In the effective process of ML in pattern recognition, classification, and regression, it is important to select informational, discriminative, and independent features. Features may be of different types, such as numerical features, strings, and graphs.

[0057] In some exemplary embodiments, the feature 402 of the ML algorithm used to find a process recipe includes a recipe feature 404, an experimental result feature 406, a hypothetical result feature 408, and a measurement feature 410. Other models may use additional features or some of these features.

[0058] Recipe feature 402 includes parameters associated with the recipe, such as workflow, gas flow, chamber temperature, chamber pressure, process duration, and radio frequency (RF) values ​​(e.g., frequency, voltage).

[0059] Experimental result feature 406 and hypothetical result feature 408 include values ​​measured from the resulting semiconductor (conmorphism, lateral ratio, isotropy, deposition depth, global adhesion coefficient, surface-dependent adhesion coefficient, delayed thickness, neutral particle-ion ratio, ion angular distribution function, etc.).

[0060] Measurement features 410 include criteria used by measurement methods such as imaging methods (e.g., scanning electron microscope (SEM), transmission electron microscope (TEM)), thickness measurements (e.g., X-ray fluorescence analysis (XRF), polarization analysis), sheet resistance, surface resistivity, stress measurement, and other analytical methods used to determine layer thickness. These other analytical methods include one or more of the following: X-ray diffraction (XRD), X-ray reflectivity (XRR), prior electron diffraction (PED), electron energy loss spectroscopy (EELS), energy dispersive X-ray spectroscopy (EDS), and secondary ion mass spectrometry (SIMS).

[0061] In some exemplary embodiments, the measurement method includes time-series data, which includes sensor measurements taken over time for specific parameters, such as how chamber pressure develops over time during the manufacturing process.

[0062] Figure 5A illustrates the training and use of a machine learning program in several exemplary embodiments. In some exemplary embodiments, a machine learning program (MLP), also called a machine learning algorithm or machine learning tool, is used to carry out steps related to determining a recipe for manufacturing semiconductors.

[0063] Machine learning, also referred to herein as a tool, is an algorithm that explores and constructs algorithms capable of learning from existing data and making predictions about new data. Such a machine learning algorithm operates by constructing an ML model 314 from exemplary training data 512 to make predictions or decisions on data expressed as outputs or assessments (e.g., the discovery of a new recipe 316). Exemplary embodiments are presented with respect to several machine learning tools, but the principles presented herein may also be applicable to other machine learning tools.

[0064] ML has two common modes: supervised ML and unsupervised ML. Supervised ML learns the relationship between inputs and outputs using prior knowledge (e.g., associating inputs with outputs or results). The goal of supervised ML is to learn the ability to bring training inputs and outputs as close together as possible given some training data, so that the ML model can provide the same relationship when given inputs to produce corresponding outputs. Unsupervised ML is the training of ML algorithms using information that has not been classified or labeled, allowing the algorithm to act on that information without guidance. Unsupervised ML is effective in exploratory analysis because it can automatically identify the structure of the data.

[0065] Common tasks in supervised machine learning 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 several items (e.g., by assigning scores to several input values). Some examples of commonly used supervised machine learning algorithms include logistic regression (LR), naive Bayes, random forest (RF), neural networks (NN), deep neural networks (DNN), matrix factorization, and support vector machines (SVM).

[0066] Some common tasks in unsupervised machine learning include clustering, representation learning, and density estimation. Some examples of commonly used unsupervised ML algorithms include k-means clustering, principal component analysis, and autoencoders.

[0067] In some embodiments, an exemplary machine learning algorithm determines a new recipe 316 for manufacturing a semiconductor. The machine learning algorithm uses training data 512 to find correlations between identified features 402 that influence the outcome. In one exemplary embodiment, the features may be of different types and may comprise the features 402 described above with respect to Figure 4.

[0068] During the training process 310, the ML algorithm analyzes the training data 512 based on identified features 402 and configuration parameters 511 defined for training. The result of the training process 310 is an ML model 314 that can be input to create an assessment.

[0069] Typically, training an ML algorithm involves analyzing large amounts of data (e.g., gigabytes to terabytes or more) to find data correlations. The ML algorithm uses the training data 512 to find correlations between identified features 402 that influence outcomes or assessments (e.g., a new recipe 316). In some exemplary embodiments, the training data 512 includes labeled data that is known data for one or more identified features 402 and one or more outcomes (e.g., measurement values).

[0070] ML algorithms typically explore many possible features and parameters before finding what they identify as the best correlation in the data. Therefore, training may require significant computing resources and time.

[0071] Many ML algorithms have 511 configuration parameters, and the more complex the ML algorithm, the more parameters 511 become available to the user. The 511 configuration parameters define the variables of the ML algorithm in the search for the best ML model. The configuration parameters include model parameters and hyperparameters. Model parameters are learned from the training data, but hyperparameters are not learned from the training data and are instead provided to the ML algorithm.

[0072] Some examples of model parameters include the maximum model size, the maximum number of training passes, the type of data swap, regression coefficients, and decision tree splitting positions.

[0073] In some exemplary embodiments, the representative model parameters are scalar attributes or contextual attributes. Examples of scalar attributes are determined physical or behavioral model parameters, such as deposition rate or suppression depth. Furthermore, contextual attributes are attributes that depend on other attributes (e.g., context) and may have physical, statistical, and machine learning-based relationships. In particular, the physical context may be the deposition rate relative to the aspect ratio.

[0074] Examples of statistical contexts include parameter reduction using principal component analysis (PCA) or linear discriminant analysis (LDA). PCA is a dimensionality reduction method used to reduce the dimensionality of a large set of data by transforming a large set of variables into a smaller set that contains most of the information of the large set. The smaller set of data is much easier and faster to explore, visualize, and analyze for machine learning algorithms, without variables irrelevant to the process.

[0075] Linear discriminant analysis (LDA) is a method used to find a linear combination of features that characterize or separate two or more classes of objects or events. The resulting combination may be used as a linear classifier or for dimensionality reduction before classification.

[0076] Examples of machine learning contexts include autoencoders, neural networks, or training regressors. These scalar or contextual attributes are representative model parameters when validated as representative of experimental data and are used as input to subsequent modeling work. When these parameters are the result of simulation work, they are called "virtual results" or "simulation results."

[0077] Hyperparameters may include the number of hidden layers in the neural network, the number of hidden nodes in each layer, the learning speed (perhaps through various adaptations for learning speed), regularization parameters, and the type of nonlinear activation function. Finding the correct (or best) set of hyperparameters can be a very time-consuming task requiring significant computer resources.

[0078] When ML model 314 is used to perform an assessment, specification 518 is provided as input to ML model 314, and ML model 314 creates a new recipe 316 as output.

[0079] Figure 5B illustrates the use of machine learning programs employing active process control in several exemplary embodiments. In some exemplary embodiments, the goal of the final recipe is active process control. Active process control is a process correlation method for compensating for incoming changes, downstream changes, or environmental changes. For example, the number of accumulation cycles can be increased in response to earlier outputs while working on a particular structure to compensate for work performed on a larger structure.

[0080] The trained ML model 314 may be deployed to determine which process parameters satisfy the controlled object. The input 520 includes a recipe and control specifications for the desired active process control. The resulting new recipe with control parameters 522 may include setpoints that depend on local control requirements in process execution time.

[0081] Figure 6 shows an example of a Delayed Inhibition (DID) deposition process using suppressed-controlled-enhanced (ICE) filling, which can be optimized using a behavioral model. In some exemplary embodiments, the behavioral model uses process abstractions to predict the structural details of components manufactured by one or more semiconductor device manufacturing processes. Examples of behavioral models are described in U.S. Patents 9,015,016 and 9,659,126, which are incorporated herein by reference.

[0082] The pre-filling stage 606 shows an unfilled component 602. The component 602 may be formed in one or more layers on a semiconductor substrate and may optionally have one or more layers along the side walls and / or bottom of the component 602. The purpose of this is to prevent voids in the filling of the component 602.

[0083] Step 608 shows the component 602 after the initial deposition of the filler material to form a layer of material 604 to fill the component 602. Examples of material 604 include tungsten, cobalt, molybdenum, and ruthenium, but the techniques described herein also include oxides (e.g., SiO2). x It may be used to optimize the filling of any suitable material 604, including other conductors and dielectrics such as AbO3, nitrides (e.g., SiN, TiN) and carbides (e.g., SiC).

[0084] Step 610 shows the components 602 after the suppression treatment. The suppression treatment is a treatment that has the effect of suppressing subsequent deposition on the treated surface 614. The suppression may include a variety of mechanisms, depending on various factors including the surface being treated, the suppression chemical, and whether the suppression is a thermal or plasma process. In one example, tungsten deposition resulting from tungsten nucleation is suppressed by exposure to a nitrogen-containing chemical. This may include the generation of activated nitrogen-containing species, for example, by a remote plasma generator or a direct plasma generator, or by exposure to ammonia vapor in an example of a thermal (non-plasma) process.

[0085] Examples of inhibition mechanisms may include chemical reactions between an active species and the constituent surface for forming a thin layer of composite material such as tungsten nitride (WN) or tungsten carbide (WC). In some embodiments, inhibition may include surface effects such as adsorption that passivate the surface without forming a layer of composite material.

[0086] The suppression may be characterized by a suppression depth 616 and a suppression gradient. That is, the suppression may vary with depth such that it is greater at the opening of the component 602 than at its bottom, and extends only partway down the component 602. In the example shown in the figure, the suppression depth 616 is about half the total depth of the component 602. In addition, the suppression treatment is stronger at the top of the component 602. Because deposition is suppressed near the opening of the component 602, during the second deposition stage Dep2 612, the material preferentially deposits at the bottom of the component 602, with little to no deposition at the opening of the component 602. This can prevent the formation of voids and seams inside the filled component 602. Therefore, during the second deposition stage Dep2 612, the material 604 may be filled in a manner characterized by bottom-up filling rather than conformal first deposition stage Dep1 filling.

[0087] As deposition continues, the inhibitory effect may be removed so that deposition on the lightly treated surface is no longer inhibited. This is represented in step 612, where the treated surface 614 is a narrower area than before step Dep2. In the example shown in the figure, as the second deposition Dep2 progresses, the inhibitor is eventually removed across the entire surface, and the components are completely filled with material 604, as shown in step 614.

[0088] Although only one suppression cycle is shown, this process may include several deposition-suppression cycles. Behavioral modeling is used to fine-tune the recipe to control deposition and suppression parameters so that voids in the filler are eliminated and the filler material meets the requirements. Measurement methods are used to measure different criteria for deposition and suppression, including the appearance of voids in the filler.

[0089] Figure 7 shows an etching chamber 700 according to one embodiment. Exciting an electric field between two electrodes is one method for obtaining a radio frequency (RF) gas discharge in the etching chamber. When an oscillating voltage is applied between the electrodes, the resulting discharge is called a capacitively coupled plasma (CCP) discharge.

[0090] Plasma 702 may be generated using a stable source gas to obtain various chemically reactive byproducts resulting from the ionization of various molecules caused by electron-neutral collisions. The chemical aspects of etching include the reaction of neutral gas molecules and their separation byproducts with molecules on the surface to be etched, as well as the generation of volatile molecules that may be emitted. When the plasma is generated, positive ions are accelerated from the plasma across a space-charge sheath that separates the plasma from the chamber wall and collide with the wafer surface with enough energy to remove material from the wafer surface. This is known as ion collision or ion sputtering. However, some industrial plasmas do not generate ions with enough energy to efficiently etch the surface by physical means alone.

[0091] The controller 716 manages the process in chamber 700 by controlling different elements of chamber 700 (such as the RF generator 718, gas source 722, and gas pump 720). In one embodiment, fluorocarbon gases (such as CF4 and C-C4F8) are used in dielectric etching processes due to their anisotropic and selective etching capabilities, but the principles described herein can also be applied to other plasma generating gases. Fluorocarbon gases are readily ionized into chemically reactive byproducts containing smaller molecular and atomic radicals. These chemically reactive byproducts etch dielectric materials, which in one embodiment may be SiO2 or SiOCH for low dielectric constant devices.

[0092] Chamber 700 represents a processing chamber having an upper electrode 704 and a lower electrode 708. The upper electrode 704 may be grounded or connected to an RF generator (not shown), and the lower electrode 708 is connected to an RF generator 718 via a matching network 714. The RF generator 718 provides RF power at one, two, or three different RF frequencies. Depending on the desired configuration of chamber 700 for a particular process, at least one of the three RF frequencies may be turned on or off. In the embodiment shown in Figure 7, the RF generator 718 provides frequencies of 2 MHz, 27 MHz, and 60 MHz, but other frequencies are also possible.

[0093] The chamber 700 includes a gas shower head on the upper electrode 704 for injecting gas supplied from the gas source 722 into the chamber 700, and a perforated confinement ring 712 from which the gas can be discharged from the chamber 700 by a gas pump 720. In some exemplary embodiments, the gas pump 720 is a turbomolecular pump, but other types of pumps may be used.

[0094] When the substrate 706 is inside the chamber 700, the silicon focus ring 710 is positioned next to the substrate 706 so that a uniform RF field exists at the bottom of the plasma 702 for uniform etching of the surface of the substrate 706. The embodiment in Figure 7 shows a triode reactor configuration in which the upper electrode 704 is surrounded by symmetrical RF ground electrodes 724. The insulator 726 is a dielectric that separates the ground electrode 724 from the upper electrode 704.

[0095] Each frequency may be selected for a specific purpose in the wafer manufacturing process. In the example in Figure 7, with RF powers provided at 2 MHz, 27 MHz, and 60 MHz, the 2 MHz RF power provides ion energy control, while the 27 MHz and 60 MHz powers provide plasma density control and chemical ionization patterns. This configuration, in which each RF power can be turned on or off, enables certain processes using ultra-low ion energy on the substrate or wafer, and certain processes requiring low ion energy (below 700 eV or 200 eV) (e.g., soft etching of low dielectric constant materials).

[0096] In another embodiment, 60 MHz RF power is used at the upper electrode 704 to achieve ultra-low energy and ultra-high density. This configuration allows for chamber cleaning with high-density plasma when the substrate 706 is not in the chamber 700, while minimizing sputtering on the electrostatic chuck (ESC) surface. The ESC surface is exposed when the substrate 706 is not present, and any ionic energy on the surface must be avoided. This is why the 2 MHz and 27 MHz power supplies at the bottom are turned off during cleaning.

[0097] Figure 8 is a flowchart of Method 800 for determining a recipe for semiconductor manufacturing by several exemplary embodiments. Various steps of this flowchart are presented and described in order, but those skilled in the art will recognize that some or all of the steps may be performed in a different order, or may be combined or omitted, or may be performed simultaneously.

[0098] In process 802, several experiments are conducted to process the components. Each experiment is controlled by one of several recipes that specify parameters for the manufacturing apparatus.

[0099] Method 800 proceeds from step 802 to step 804 for performing multiple virtual simulations to process the components. Each simulation is controlled by one of several recipes.

[0100] Method 800 proceeds from step 804 to step 806, in which an ML model is obtained by training an ML algorithm using experimental results and virtual results from virtual simulations.

[0101] Method 800 proceeds from step 806 to step 808 to receive specifications for the desired processing of the components. In step 810, the ML model creates a new recipe for processing the components based on the specifications.

[0102] In one example, an ML model is based on multiple features, including recipe features, experimental result features, hypothetical result features, and measurement features.

[0103] For example, the measurement features include one or more of the following: imaging methods, transmission electron microscopy, thickness measurement, sheet resistance, surface resistivity, stress measurement, and analytical methods used to determine the thickness, composition, particles, or orientation of the layer.

[0104] In one example, recipe features include workflow, gas flow, chamber temperature, chamber pressure, process duration, and radio frequency (RF) values.

[0105] In one example, virtual simulation is performed by a simulation tool based on behavioral modeling.

[0106] For example, the experimental results include values ​​measured from the processing of the constituent elements, which include one or more of the following: lateral ratio, isotropy ratio, deposition depth, global adhesion coefficient, surface-dependent adhesion coefficient, delayed thickness, neutral particle-ion ratio, and ion angular distribution function.

[0107] In one example, each experiment is performed on semiconductor manufacturing equipment based on an experimental recipe, and one experiment is performed to measure the effect of changing one parameter value from a previous recipe used in a previous experiment.

[0108] In one example, the processing of the constituent elements is for deposition processing using a suppression profile.

[0109] In one example, the processing of the components is for deposition in 3D NAND word line (WL) filling.

[0110] Another common embodiment relates to a system comprising memory containing instructions and one or more computer processes. When the instructions are executed by one or more computer processors, the instructions cause one or more computer processors to perform the following steps: perform a series of experiments for processing a component, each experiment being controlled by one of a series of recipes that specify parameters of a manufacturing apparatus; perform a series of virtual simulations for processing the component, each simulation being controlled by one of a series of recipes; obtain an ML model by training an ML algorithm using the experimental results and virtual results of the virtual simulations; receive specifications for the desired processing of the component; and create new recipes for processing the component by the ML model based on the specifications.

[0111] In yet another common embodiment, a machine-readable storage medium (e.g., a non-temporary storage medium) includes an instruction that causes the machine to perform several experiments for processing a component, each experiment being controlled by one of several recipes specifying parameters of a manufacturing apparatus; several virtual simulations for processing the component, each simulation being controlled by one of several recipes; an ML model being obtained by training an ML algorithm using the experimental results and virtual results of the virtual simulations; receiving specifications for a desired processing of the component; and creating a new recipe for processing the component by the ML model based on the specifications.

[0112] Figure 9 is a block diagram representing an example of a machine 900 on which one or more exemplary processes described herein are performed or controlled. In other embodiments, machine 900 may operate as a standalone device or may be connected to other machines (e.g., networked). In a networked configuration, machine 900 may operate as a server machine, a client machine, or both in a server-client network environment. In the example, machine 900 may function as a peer machine in a peer-to-peer (P2P) (or other distributed) network environment. Furthermore, although only a single machine 900 is shown, the term “machine” may also be interpreted to include a group of machines that individually or collaboratively perform a set (or more sets) of instructions for performing one or more methods described herein (e.g., by cloud computing, software as a service (SaaS), or other computer cluster architectures).

[0113] The examples described herein may include, or be operated by, logic, several components, or mechanisms. A network is a set of circuits implemented in a tangible object that includes hardware (e.g., simplex circuits, gates, logic). Components of a network may be flexible with respect to time and to changes in the underlying hardware. A network includes elements that, individually or in combination, can perform a specific operation when in operation. In the examples, the hardware of a network may be designed (built into the hardware) immutably to perform a specific operation. In the examples, the hardware of a network may include variable-connected physical components (e.g., execution devices, transistors, simplex circuits) that include a computer-readable medium that has been physically modified (e.g., magnetically, electrically, or by immutable dense particles) to encode instructions for a specific operation. When physical components are connected, the basic electrical properties of the hardware elements are changed (e.g., from insulator to conductor, or vice versa). The instructions allow the embedded hardware (e.g., execution devices or loading mechanisms) to form elements of the hardware network by variable connections in order to perform a portion of a specific operation when in operation. Accordingly, the computer-readable medium is communicatively connected to other components of the network when the device is operating. In the example, any physical element may be used in one or more elements of one or more networks. For example, the execution device may, at a given point in time, be used in a first circuit of a first network, then again in a second circuit of the first network, or again in a third circuit of a second network at a different time.

[0114] The machine (e.g., a computer system) 900 may include a hardware processor 902 (e.g., a central processing unit (CPU), a hardware processor core, or a combination thereof), an image processing unit (GPU) 903, main memory 904, and static memory 906, some or all of which can communicate with each other via an interlink (e.g., a bus) 908. The machine 900 may further include a display device 910, an alphanumeric input device (e.g., a keyboard) 912, and a user interface (UI) navigation device (e.g., a mouse) 914. In this example, the display device 910, the alphanumeric input device 912, and the UI navigation device 914 may be touchscreens. The machine 900 may further include a mass storage device (e.g., a drive unit) 916, a signal generator (e.g., a speaker) 918, a network interface device 920, and one or more sensors 921 (such as a Global Positioning System (GPS) sensor, a compass, an accelerometer, or another sensor). The machine 900 may include an output control device 928 (such as a series connection (e.g., Universal Serial Bus (USB)), a parallel connection, or other wiring connection, or a wireless connection (e.g., infrared (IR), near-field communication (NFC))) for communicating with or controlling one or more peripheral devices (e.g., a printer, a card reader).

[0115] The mass storage device 916 may include a machine-readable medium 922 in which one or more data structures or instructions 924 (e.g., software) are stored, which embody or are used by one or more of the technologies or functions described herein. The instructions 924 may reside, all or at least partially, in the main memory 904, static memory 906, hardware processor 902, or GPU 903 during their execution by the machine 900. In the example, one or any combination of the hardware processor 902, GPU 903, main memory 904, static memory 906, and mass storage device 916 may constitute the machine-readable medium.

[0116] Although the machine-readable medium 922 is shown as a single medium, the term “machine-readable medium” may include a single or multiple medium configured to store one or more instructions 924 (e.g., a centralized or distributed database, and / or associated caches and servers).

[0117] The term “machine-readable medium” may include any medium capable of storing, encoding, or executing instructions 924 for execution by machine 900, any medium capable of causing machine 900 to implement one or more of the techniques of this disclosure, or any medium capable of storing, encoding, or executing data structures used by or associated with such instructions 924. Non-limiting examples of machine-readable mediums may include solid memory, as well as optical and magnetic media. In the examples, the group of machine-readable mediums includes a machine-readable medium 922 comprising a plurality of particles having an immutable (e.g., stationary) mass. Thus, the group of machine-readable mediums is not a transient propagating signal. Specific examples of the group of machine-readable mediums may include non-volatile memory (such as semiconductor memory devices (e.g., erasable read-only memory (EPROM), rewritable read-only memory (EEPROM)), and flash memory devices), magnetic disks (such as internal hard disks and remoteable disks), magneto-optical disks, and CD-ROM and DVD-ROM disks.

[0118] Instruction 924 may further be transmitted to or received by the communication network 926 via the network interface device 920 using a transmission medium.

[0119] Throughout this specification, multiple examples may include components, processes, or structures described as a single example. While individual steps of one or more methods are described as separate steps, these individual steps may be performed simultaneously, and the steps do not need to be performed in the order illustrated. Structures and functions presented as separate components in the illustrative configurations may be implemented as an integrated structure or component. Similarly, structures and functions presented as single components may be implemented as separate components. These and other changes, modifications, additions, and improvements fall within the scope of this specification.

[0120] The embodiments described herein are described in sufficient detail to enable those skilled in the art to perform the teachings of the disclosure. Other embodiments may be used or derived therefrom so as to be structural and logical substitutions and modifications without departing from the scope of this disclosure. Therefore, the forms for carrying out the invention should not be taken restrictively, and the scope of the various embodiments is defined only by the appended claims and the entire scope equivalent to the scope of such claims.

[0121] The term “or” as used herein may be interpreted either in an inclusive or exclusive sense. Furthermore, multiple examples may be provided for a resource, process, or structure described herein as one example. In addition, the boundaries between various resources, processes, modules, engines, and data storage are somewhat arbitrary, and a particular process is described in terms of a particular exemplary configuration. The distribution of other functions is speculative and may fall within the scope of various embodiments of this disclosure. Generally, structures and functions presented as separate resources in an exemplary configuration may be implemented as an integrated structure or resource. Similarly, structures and functions presented as a single resource may be implemented as separate resources. These and other changes, modifications, additions, and improvements fall within the scope of embodiments of this disclosure as set forth in the appended claims. Accordingly, the specification and drawings are not restrictive but illustrative. [Application Example 1] A method, Multiple experiments were conducted to process the components, and each experiment was controlled by one recipe from several recipes that identified parameters for the manufacturing apparatus. Multiple virtual simulations are performed to process the aforementioned components, and each simulation is controlled by one of the multiple recipes. By training a machine learning (ML) algorithm using the experimental results and the virtual results of the virtual simulation, an ML model is obtained. Having received the specifications for the desired processing of the aforementioned components, Based on the above specifications, create a new recipe for processing the components using the ML model. A method that includes [a certain feature]. [Application Example 2] The method described in Application Example 1, The ML model is a method based on multiple features, including recipe features, experimental result features, hypothetical result features, and measurement features. [Application Example 3] The method described in Application Example 2, The measurement features include one or more of the following: imaging methods, transmission electron microscopy, thickness measurement, sheet resistance, surface resistivity, stress measurement, and analytical methods used to determine the thickness, composition, particles, or orientation of a layer. [Application Example 4] The method described in Application Example 2, The aforementioned recipe features include a method comprising workflow, gas flow, chamber temperature, chamber pressure, process duration, and radio frequency (RF) value. [Application Example 5] The method described in Application Example 2, A method comprising an ML model that includes active process control for determining process parameters to satisfy a control objective, wherein the input to the ML model includes the recipe and the control objective for the desired active process control. [Application Example 6] The method described in Application Example 1, The aforementioned virtual simulation is performed by a simulation tool based on behavioral modeling. [Application Example 7] The method described in Application Example 1, The experimental results include values ​​measured from the processing of the components, wherein the values ​​include one or more of the following: lateral ratio, isotropy ratio, deposition depth, global adhesion coefficient, surface-dependent adhesion coefficient, delayed thickness, neutral particle-ion ratio, and ion angular distribution function. [Application Example 8] The method described in Application Example 1, A method in which each experiment is carried out in a semiconductor manufacturing apparatus based on the recipe for the experiment, and one experiment is carried out to measure the effect of changing the value of one parameter from a previous recipe used in a previous experiment. [Application Example 9] The method described in Application Example 1, The method for processing the aforementioned components is for a deposition process using an inhibitory profile. [Application Example 10] The method described in Application Example 1, The method for processing the aforementioned components is for deposition in 3D NAND word line (WL) filling. [Application Example 11] A system in which, Memory containing instructions, The instruction comprises one or more computer processors, and is executed by the one or more computer processors. The process involves conducting multiple experiments to process the components, with each experiment controlled by one recipe from several recipes that identify parameters for the manufacturing equipment. A process for performing multiple virtual simulations to process the aforementioned components, and each simulation being controlled by one recipe from the multiple recipes, The process involves obtaining an ML model by training a machine learning (ML) algorithm using experimental results and virtual results from the virtual simulation, A step of receiving specifications for the desired processing of the aforementioned components, A system that causes the system to perform the steps of creating a new recipe for processing the components using the ML model based on the specifications described above. [Application Example 12] The system described in Application Example 11, The aforementioned ML model is a system based on multiple features, including recipe features, experimental result features, hypothetical result features, and measurement features. [Application Example 13] The system described in Application Example 12, The measurement features include one or more of the following: imaging methods, transmission electron microscopy, thickness measurement, sheet resistance, surface resistivity, stress measurement, and analytical methods used to determine the thickness, composition, particles, or orientation of a layer. [Application Example 14] The system described in Application Example 12, The aforementioned recipe features include the workflow, gas flow, chamber temperature, chamber pressure, process duration, and radio frequency (RF) values ​​of the system. [Application Example 15] The system described in Application Example 11, The experimental results include values ​​measured from the processing of the components, the values ​​including one or more of the following: lateral ratio, isotropy ratio, deposition depth, global adhesion coefficient, surface-dependent adhesion coefficient, delayed thickness, neutral particle-ion ratio, and ion angular distribution function. [Application Example 16] The system described in Application Example 11, Each experiment is performed in a semiconductor manufacturing apparatus based on the recipe for the experiment, and one experiment is performed to measure the effect of changing the value of one parameter from a previous recipe used in a previous experiment. [Application Example 17] A tangible machine-readable storage medium is available, When performed by a machine, The process involves conducting multiple experiments to process the components, with each experiment controlled by one recipe from several recipes that identify parameters for the manufacturing equipment. The process involves performing multiple virtual simulations to process the aforementioned components, and each simulation is controlled by one recipe from the multiple recipes. The process involves obtaining an ML model by training a machine learning (ML) algorithm using experimental results and virtual results from the virtual simulation, A step of receiving specifications for the desired processing of the aforementioned components, A step of creating a new recipe for processing the components using the ML model based on the above specifications, A tangible, machine-readable storage medium containing an instruction to cause the machine to perform the action. [Example 18] A tangible machine-readable storage medium as described in Example 17, The ML model is a tangible, machine-readable storage medium based on multiple features, including recipe features, experimental result features, hypothetical result features, and measurement features. [Application Example 19] A tangible machine-readable storage medium as described in Application Example 18, A tangible, machine-readable storage medium, wherein the measurement features include one or more of the following: imaging methods, transmission electron microscopy, thickness measurement, sheet resistance, surface resistivity, stress measurement, and analytical methods used to determine the thickness, composition, particles, or orientation of a layer. [Example 20] A tangible machine-readable storage medium as described in Example 18, The aforementioned recipe features include a tangible, machine-readable storage medium containing the workflow, gas flow, chamber temperature, chamber pressure, process duration, and radio frequency (RF) values. [Example 21] A tangible machine-readable storage medium as described in Example 17, The experimental results include values ​​measured from the processing of the components, the values ​​including one or more of the following: lateral ratio, isotropy ratio, deposition depth, global adhesion coefficient, surface-dependent adhesion coefficient, delayed thickness, neutral particle-ion ratio, and ion angular distribution function, in a tangible, machine-readable storage medium.

Claims

1. It is a method, Multiple experiments were conducted to process and form semiconductor devices, each experiment being controlled by one semiconductor device processing recipe from a set of semiconductor device processing recipes that specify parameters for the manufacturing equipment used to process the semiconductor devices. Multiple virtual simulations are performed to process the semiconductor device, and each virtual simulation is controlled by one semiconductor device processing recipe from the multiple semiconductor device processing recipes. By training a machine learning (ML) algorithm using experimental results and the virtual results of the aforementioned multiple virtual simulations, an ML model is obtained, and the ML model is based on multiple features including recipe features, experimental result features, virtual result features, and measurement features. The specifications for the desired processing of the semiconductor device are received, Based on the above specifications, create a new recipe for processing the semiconductor device using the ML model. A method that includes [a certain feature].

2. The method according to claim 1, A method wherein the measurement features include one or more criteria used by imaging methods, transmission electron microscopy, thickness measurement, stress measurement, and analytical methods used to determine the thickness, composition, particles, or orientation of a layer.

3. The method according to claim 1, The recipe features include a method comprising a workflow, gas flow, chamber temperature, chamber pressure, process duration, and radio frequency (RF) values.

4. The method according to claim 1, A method wherein the ML model is used for active process control to determine process parameters in order to satisfy a control objective, and the input to the ML model includes a recipe for processing the semiconductor device and the control objective for the desired active process control.

5. The method according to claim 1, The aforementioned virtual simulation is performed by a simulation tool based on behavioral modeling.

6. The method according to claim 1, The experimental results include values ​​measured from the processing of the semiconductor device, wherein the values ​​include one or more of the following: lateral ratio, isotropy ratio, deposition depth, global adhesion coefficient, surface-dependent adhesion coefficient, delayed thickness, neutral particle-ion ratio, and ion angular distribution function.

7. The method according to claim 1, A method wherein each experiment is carried out in a semiconductor manufacturing apparatus based on the semiconductor device processing recipe for the experiment, and one experiment is carried out to measure the effect of changing the value of one parameter from a previous semiconductor device processing recipe used in a previous experiment.

8. The method according to claim 1, The method for processing the aforementioned semiconductor device is a deposition process that uses an suppression profile to inhibit deposition.

9. The method according to claim 1, The method for processing the aforementioned semiconductor device is for deposition in 3D NAND word line (WL) filling.

10. It is a system, Memory containing instructions, The instruction comprises one or more computer processors, and is executed by the one or more computer processors. A process of conducting multiple experiments for processing and forming semiconductor devices, each experiment being controlled by one semiconductor device processing recipe from a plurality of semiconductor device processing recipes that specify parameters for the manufacturing equipment used to process the semiconductor device, A process for performing multiple virtual simulations for processing the semiconductor device, and each virtual simulation is controlled by one semiconductor device processing recipe from the multiple semiconductor device processing recipes. The process involves obtaining an ML model by training an ML algorithm using experimental results and virtual results from the aforementioned multiple virtual simulations, and the ML model being based on multiple features including recipe features, experimental result features, virtual result features, and measurement features. A step of receiving specifications for the desired processing of the semiconductor device, A system that causes the system to perform the steps of creating a new recipe for processing the semiconductor device using the ML model based on the specifications described above.

11. The system according to claim 10, The measurement features include one or more criteria used by imaging methods, transmission electron microscopy, thickness measurement, stress measurement, and analytical methods used to determine the thickness, composition, particles, or orientation of a layer.

12. The system according to claim 10, The aforementioned recipe features include the workflow, gas flow, chamber temperature, chamber pressure, process duration, and radio frequency (RF) values ​​of the system.

13. The system according to claim 10, The experimental results include values ​​measured from the processing of the semiconductor device, wherein the values ​​include one or more of the following: lateral ratio, isotropy ratio, deposition depth, global adhesion coefficient, surface-dependent adhesion coefficient, delayed thickness, neutral particle-ion ratio, and ion angular distribution function.

14. The system according to claim 10, Each experiment is performed in a semiconductor manufacturing apparatus based on the semiconductor device processing recipe for the experiment, and one experiment is performed to measure the effect of changing the value of one parameter from a previous semiconductor device processing recipe used in a previous experiment.

15. There is a tangible machine-readable storage medium, When performed by a machine, A process of conducting multiple experiments for processing and forming semiconductor devices, each experiment being controlled by one semiconductor device processing recipe from a plurality of semiconductor device processing recipes that specify parameters for the manufacturing equipment used to process the semiconductor device, The process involves performing multiple virtual simulations for processing the semiconductor device, and each virtual simulation is controlled by one semiconductor device processing recipe from the multiple semiconductor device processing recipes. The process involves obtaining an ML model by training an ML algorithm using experimental results and virtual results from the aforementioned multiple virtual simulations, and the ML model being based on multiple features including recipe features, experimental result features, virtual result features, and measurement features. A step of receiving specifications for the desired processing of the semiconductor device, A step of creating a new recipe for processing the semiconductor device using the ML model based on the above specifications, A tangible, machine-readable storage medium containing an instruction to cause the machine to perform the action.

16. A tangible machine-readable storage medium according to claim 15, A tangible, machine-readable storage medium, wherein the measurement features include one or more criteria used by imaging methods, transmission electron microscopy, thickness measurement, stress measurement, and analytical methods used to determine the thickness, composition, particles, or orientation of a layer.

17. A tangible machine-readable storage medium according to claim 15, The aforementioned recipe features include a tangible, machine-readable storage medium containing the workflow, gas flow, chamber temperature, chamber pressure, process duration, and radio frequency (RF) values.

18. A tangible machine-readable storage medium according to claim 15, The experimental results include values ​​measured from the processing of the semiconductor device, wherein the values ​​include one or more of the following: lateral ratio, isotropy ratio, deposition depth, global adhesion coefficient, surface-dependent adhesion coefficient, delayed thickness, neutral particle-ion ratio, and ion angular distribution function, in a tangible, machine-readable storage medium.

Images