Method and system for monitoring plasma free radical substance flux

Abstract

A method of monitoring a plasma-based process in a processing chamber includes: measuring a first temperature at a first location associated with the processing chamber during the plasma-based process; and determining a value indicative of a first free radical species flux associated with the plasma-based process based on the first temperature. The method includes a trained machine learning model to determine whether a value representative of the free radical substance flux satisfies a free radical substance flux drift threshold.

Description

Technical Field

[0001] The embodiments disclosed herein relate to manufacturing. Specifically, this disclosure relates to a system and method for monitoring plasma free radical flux. Background Technology

[0002] Traditionally, manufacturing formulations processed through processing chambers are static formulations applied mechanically and do not respond to in-situ conditions. Furthermore, the timing of maintenance and reuse of the processing chamber is statically determined based on pre-set schedules and predetermined formulations. Processing chambers typically lack any autonomy or ability to make their own decisions regarding processing formulations, maintenance, tooling, etc.

[0003] The processing chamber (i.e., the plasma chamber) is ideally maintained in a steady state to preserve the processing uniformity of the substrate (e.g., a wafer) being processed within the chamber. Steady-state conditions depend on maintaining good chamber health. Chamber health can include parameters such as accurate power input, functional heat exchange and other thermal components, leak-free operation, and the like.

[0004] Currently, it is difficult to monitor the processing environment of the chamber to ensure that everything works as expected. Instead, the metering of the processed substrate is used to determine when the chamber experiences a shift from the desired steady-state conditions. This can lead to improper substrate processing, thereby increasing costs and reducing throughput. Therefore, an improved method for monitoring the processing chamber is needed. Summary of the Invention

[0005] This document describes a method for monitoring a processing chamber. In some embodiments, the method may include: measuring a first temperature at a first location associated with the processing chamber during a plasma-based process; and determining a value representing a first free radical flux associated with the plasma-based process based on the first temperature.

[0006] In some embodiments, a method may include receiving or generating a training dataset associated with a plasma-based process performed at a processing chamber, the training dataset comprising a plurality of data items. Each of the plurality of training data items may include a temperature measurement generated by a temperature sensor at a location associated with the processing chamber and a label indicating the state of the processing chamber. The method further includes using the training dataset to train a machine learning model to generate a trained machine learning model trained to receive the temperature measurement generated by the temperature sensor at that location and, based on the temperature, determine a value representing the free radical mass flux utilizing the plasma-based process.

[0007] In some embodiments, a system includes: a processing chamber configured to perform a plasma-based process; a temperature sensor at a first location associated with the processing chamber, the temperature sensor being configured to generate one or more temperature measurements during the plasma-based process; and a computing device. The computing device is configured to: receive the first temperature measurement generated by the temperature sensor during the plasma-based process; and determine a value representing a first free radical flux associated with the plasma-based process based on the first temperature.

[0008] Several other features are provided according to these and other embodiments disclosed herein. These other features and embodiments will become more apparent from the following description, the claims, and the accompanying drawings. Attached Figure Description

[0009] This disclosure is illustrated by way of example in the accompanying drawings and is not intended to be limiting, in which like references designate similar components. It should be noted that different references to "an" or "one" embodiments in this disclosure are not necessarily the same embodiments, and such references imply at least one.

[0010] Figure 1A This is a plan view illustration of a processing tool having a remote plasma source (RPS) and an indirect process monitoring sensor, according to an embodiment.

[0011] Figure 1B This is a plan view illustration of a processing tool having an RPS and a plurality of indirect process monitoring sensors, according to an additional embodiment;

[0012] Figure 2 A cross-sectional view of a processing tool including a processing chamber according to an embodiment is depicted;

[0013] Figure 3 This is a flowchart of a method for monitoring and processing a chamber according to an embodiment.

[0014] Figure 4 This is a flowchart of a method for performing actions by means of a processing tool and / or a substrate processing system according to another embodiment.

[0015] Figure 5 This is a flowchart illustrating the method for monitoring the processing chamber and determining when to perform maintenance on it.

[0016] Figure 6 This is a flowchart of a method for automatically determining when to perform maintenance on a processing chamber, according to an embodiment.

[0017] Figure 7This is a flowchart of a method for automatically determining when to put a processing chamber back into use after maintenance has been performed, according to an embodiment.

[0018] Figure 8 A graphical representation of a machine in the form of an example computing device is shown, in which a set of instructions can be executed to cause the machine to perform any one or more of the methods discussed herein. Detailed Implementation

[0019] The embodiments described herein relate to methods and / or systems for monitoring the amount of plasma flux (e.g., radical flux) during plasma-based processes. Monitoring the radical flux of plasma is challenging when performing plasma-based processes. Currently, there is no robust solution for reproducibly monitoring the radical flux generated by a plasma source. While optical emission spectroscopy has been attempted to monitor radicals, this technique is typically inefficient and expensive. The methods and systems described herein address these problems using non-optical sensing components (e.g., by using non-optical sensors whose measurements correlate with radical flux concentration) and allow for the detection of the amount of radical flux associated with a plasma-based process that is efficient and inexpensive. The systems and methods further enable the detection of conditions that alter radical flux, such as radical flux drift. The methods and systems disclosed herein provide a low-cost, non-optical technique and system for monitoring the plasma health of in-situ or remote plasma sources.

[0020] The methods and systems according to embodiments of this disclosure utilize temperature sensors that may already be present in the processing chamber and / or in exhaust systems or fore-line components associated with the processing chamber to detect free radical flux. Because the methods and systems utilize temperature sensors, they can be easily and cost-effectively integrated into the processing chamber. The inventors of this disclosure have discovered that small temperature changes correlate with changes in free radical flux. When the free radical flux changes, one or more parameters of the plasma-based process (e.g., etch rate, deposition rate, film growth rate, etc.) may change with the density of free radicals in the process.

[0021] In an embodiment, a method is provided that includes: measuring a first temperature at a first location associated with a processing chamber during a plasma-based process; and determining a value representing a first free radical flux associated with the plasma-based process based on the first temperature. In some embodiments, the measured temperature is input into a trained machine learning model that has been trained to estimate the value associated with the free radical flux based on the temperature. The trained machine learning model can then output a current estimate associated with the free radical flux based on the temperature. In some embodiments, the machine learning model outputs an actual estimated free radical flux. In some embodiments, the value output by the trained machine learning model is proportional to the free radical flux but does not indicate the actual amount of free radical flux. In some embodiments, the processing chamber may have any chamber configuration known in the art. For example, the chamber may be a plasma etching chamber, a plasma deposition chamber, etc., which may use a remote plasma source or an in-situ plasma source. In some embodiments, the plasma-based process may include plasma etching, plasma-assisted chemical vapor deposition (PECVD), plasma-enhanced atomic layer deposition, plasma etching, etc.

[0022] In some embodiments, the first location for placing the temperature sensor may include or be the inlet of the processing chamber through which plasma flows during plasma-based processes. In some embodiments, the first location is a bend or restriction in a vacuum line that supplies plasma (e.g., from a remote plasma source) to the processing chamber. In some embodiments, the first location is inside the processing chamber. In some embodiments, the first location is at the exhaust line of the processing chamber. Other locations may also be used for the temperature sensor.

[0023] In some embodiments of the method, determining a value representing the free radical flux includes inputting temperature into a trained machine learning model that outputs a value representing the free radical flux.

[0024] In this embodiment, the trained machine learning model is an edge-based model executed on the processing chamber itself, rather than on a remote computing device. Training of the machine learning model can be performed remotely, and the trained model can then be transferred to or executed on the processing chamber. Retraining or training updates of the machine learning model can be performed periodically or continuously on the processing chamber. By executing and / or training (including retraining) the machine learning model on the processing chamber, the latency between the generation of sensor measurements and the decision-making based on those measurements can be significantly reduced. This improves the ability to make real-time decisions on the processing chamber. Furthermore, shifting decision-making to the processing chamber reduces the amount of data transmitted over the network, increases efficiency, and increases the speed at which decisions can be made. For example, in an embodiment including a machine learning model trained to detect the end of an etching process, a decision on when to stop the etching process can be made within seconds or fractions of a second upon receiving sensor data that triggers such a decision.

[0025] In some embodiments, the method may further include: measuring a second temperature at a second location associated with a processing chamber during a plasma-based process; and determining a second value representing a second free radical flux associated with the plasma-based process based on the second temperature. In some embodiments, the second location may include an exhaust line to the processing chamber.

[0026] In some embodiments, the plasma flux may be stable when the temperature at the inlet of the processing chamber and / or the temperature at the exhaust line of the processing chamber are stable (e.g., do not change over time or change only minimally). In some embodiments, the processing chamber may have reached a steady state and the processing substrate may be ready when the temperature at the inlet of the processing chamber is approximately the same as the temperature at the outlet of the processing chamber. In some embodiments, the location of one or more temperature sensors whose measurements are used to determine the free radical flux (e.g., at the inlet and / or exhaust line of the processing chamber and / or surrounding area) may include a catalytic material. In some embodiments, the temperature sensors may be mounted to the catalytic material. In other embodiments, the chamber components may be made of stainless steel, nickel, or another catalytic material. In some embodiments, the location of one or more temperature sensors is relative to the location where the free radical material may collide with other locations and / or bends in the processing chamber. The catalytic material may include stainless steel, platinum, nickel, gold, and / or another material. The catalytic material may increase the reaction of the free radical material with the site where the temperature sensor is located, thereby increasing the sensitivity of the detection of the free radical flux. In some embodiments, the temperature sensor disclosed herein may be a thermocouple, a resistance thermometer, a resistance temperature device, a thermistor, and / or other temperature sensors.

[0027] In some embodiments, the method may further include determining whether the determined value representing the free radical flux meets a criterion. The method may further include scheduling maintenance in response to the determination that the value representing the free radical flux meets the criterion.

[0028] In some embodiments, the plasma-based process is used to process a quaternization process in a chamber. In some embodiments of the method, temperature measurement may be performed at a first time. The method may further include measuring a second temperature at a first location at a second time during the plasma-based process. The method may further include determining a second value representing the flux of a second free radical substance based on the second temperature. The method may further include comparing the second value with a first value to identify a difference between the second value and the first value. The method may further include determining whether the difference between the second value and the first value meets a criterion. The method may further include stopping the quaternization process in response to determining that the difference meets the criterion.

[0029] In another embodiment of the method, temperature measurement is performed at a first time. The method may further include measuring a second temperature at a first location at a second time during a second plasma-based process, wherein the second plasma-based process is the same as the plasma-based process. The method may further include determining a second value representing the flux of a second free radical substance based on the second temperature. The method may further include comparing the second value with a first value to identify a difference between the second value and the first value. The method may further include determining whether the difference between the second value and the first value meets a criterion. The method may further include scheduling maintenance in response to the determination that the difference meets the criterion.

[0030] In some embodiments, the standard may include a free radical flux drift threshold and may be satisfied in response to a difference reaching or exceeding the free radical flux drift threshold.

[0031] In another embodiment, an additional method is provided. The method includes receiving or generating a training dataset associated with a plasma-based process performed in a processing chamber under known favorable conditions. The training dataset may include a plurality of data items, each of which may include a temperature measurement generated by a temperature sensor at a location associated with the processing chamber and a label indicating the state of the processing chamber. The method further includes using the training dataset to train a machine learning model to generate a trained machine learning model that is trained to receive the temperature measurement generated by the temperature sensor at that location and, based on the temperature, determine a value representing the free radical mass flux associated with the plasma-based process. Different machine learning models may be trained to process temperature measurements obtained by different sensors at different locations.

[0032] In some embodiments, the machine learning model may be further trained to predict, based on temperature, when to perform a seasonalization process on the processing chamber after a maintenance event has been completed. In one embodiment, the machine learning model is trained based on data from one or more chambers in a known target state. The trained machine learning model may then be monitored until sensor indication signals match the known target state within a tolerance (e.g., + / - 10-20%). Conditioning or seasonalization of the chamber may be completed when the sensors match the trained sensor values ​​within a specified tolerance. In another embodiment, the machine learning model may be further trained to predict, based on temperature, when to perform maintenance associated with the processing chamber.

[0033] In another embodiment, a system is provided. The system includes a processing chamber configured to perform a plasma-based process. The system further includes a temperature sensor and a computing device at a first location associated with the processing chamber, wherein the temperature sensor is configured to generate one or more temperature measurements during the plasma-based process. The computing device may be configured to receive, during the plasma-based process, the first temperature measurement generated by the temperature sensor and to determine, based on the first temperature, a value representing a first free radical flux associated with the plasma-based process. In some embodiments, the temperature sensor disclosed herein may be a thermocouple, a resistance thermometer, a resistance temperature device, a thermistor, or another type of temperature sensor.

[0034] In another embodiment, the non-transitory computer-readable medium includes instructions that, when executed by a processor, cause the processor to perform operations including: measuring a first temperature at a first location associated with a processing chamber during a plasma-based process, and determining a value representing a first free radical flux associated with the plasma-based process based on the first temperature.

[0035] Generally, the embodiments disclosed herein are applicable to use with any processing tool employing plasma-based processes. For example, embodiments can be used with semiconductor processing tools employing plasma-based processes. Examples of processing tools employing plasma-based processes include plasma etching reactors, plasma cleaners, and plasma-based deposition chambers (e.g., CVD chambers, ion-assisted deposition (IAD) chambers, ALD chambers, etc.). In some embodiments, the processing tool includes a remote plasma source (RPS). In some embodiments, the processing tool includes an in-situ plasma source. In one example, the processing tool is used for rapid thermal processing (RTP), such as oxidation processes, plasma-enhanced thermal oxidation, and / or nitriding processes. In this embodiment, an array of heating lamps may be provided above the substrate. A reflector plate may be provided below the substrate to reflect thermal energy back to the substrate. In embodiments, the RPS or in-situ plasma source may be coupled to the chamber to provide improved oxidation efficiency.

[0036] It will be understood that additional temperature sensors can be added to the processing chamber, and their measurements can also be used to detect the free radical flux at different locations. The location of the temperature sensor can be selected based on knowledge of where free radicals can impact within the chamber and / or areas associated with the chamber, such as at the inlet and outlet of the processing chamber. It has been found that increased amounts of free radicals can impact any location within a processing chamber having bends or at an angle of at least about 10° (e.g., about 90°), such as a corner of the processing chamber. Furthermore, increased amounts of free radicals can impact locations where there are confinements in the flow.

[0037] Please refer to the diagram below. Figure 1A This is a plan view illustration of a processing tool 100 according to an embodiment. In an embodiment, the processing tool 100 may include a chamber 105. The chamber 105 may have any chamber configuration. For example, the chamber 105 may be suitable for low-pressure or near-atmospheric pressure environments. The chamber may be an oxidation chamber, an etching chamber, a deposition chamber, or any other type of chamber. In an embodiment, the chamber 105 may include a support for holding a substrate 107. The substrate 107 may be a semiconductor substrate, such as a silicon wafer or the like. The substrate 107 may have any suitable external dimensions (e.g., 200 mm, 300 mm, 450 mm, etc.). The substrate may replace a ground glass substrate or another type of substrate.

[0038] In one embodiment, the processing tool 100 may be a rapid thermal processing (RTP) tool. The processing tool 100 may be mounted on the substrate 107 (i.e., on...). Figure 1AAn array of thermal lamps (not shown) is provided outside the plane of the substrate 107. The lamps are suitable for rapidly increasing the temperature of the substrate 107 to enable thermally driven processes such as thermal oxidation. In an embodiment, a reflector plate 108 may be provided below the substrate 107. The reflector plate 108 can reflect heat energy upwards back to the substrate 107.

[0039] Chamber 105 may also include additional components. For example, a slit valve 106 may be provided along the sidewall of chamber 105. Slit valve 106 may be an opening through which substrate 107 inserts into and retracts from chamber 105. In embodiments, chamber 105 may also include an exhaust device 104. Exhaust device 104 may be an outlet for removing gas or other byproducts from chamber 105. Exhaust device 104 may include conduits, pumps, and the like.

[0040] A remote plasma source (RPS) 115 is provided as part of a semiconductor processing tool. The RPS 115 generates plasma outside a chamber 105, and a connector 112 fluidly couples the chamber to the RPS 115. The connector 112 may be a ceramic-lined stainless steel component. For example, the ceramic may include quartz or the like. In an embodiment, the RPS 115 may be (at least partially) controlled by an RPS connector 117. It will be understood that components such as magnetrons and generators (not shown) may also be used to control the RPS 115. The RPS connector 117 may include forward power settings, stub settings, and the like. In an embodiment, a mass flow meter (MFM) 116 may be provided along a gas line 118 supplying the RPS 115. Additional sensors, such as pressure sensors and optical sensors (not shown), may also be used to monitor the performance of the RPS 115.

[0041] RPS 115 can be coupled to connector 112 via gasket 113 (such as an O-ring or similar). Gasket 113 can be a worn component that deteriorates during use of semiconductor processing tool 100. For example, gasket 113 can be a common leakage source for semiconductor processing tool 100. However, it will be understood that leakage can also occur at other locations.

[0042] A plurality of sensors may be provided within the semiconductor processing tool 100. For example, a first sensor 121 may be provided in chamber 105. More specifically, the first sensor 121 may be configured to detect the temperature of reflector plate 108. The first sensor 121 may be any suitable sensor type. For example, the first sensor 121 may be a thermocouple or the like. In some embodiments, the first sensor 121 may be in direct contact with reflector plate 108. In some embodiments, the first sensor 121 may be located in the inlet of processing chamber 105 through which plasma flows during plasma-based processes.

[0043] In embodiments, the plurality of sensors may further include a second sensor 122, which is provided on or in the connector 112 or in the inlet gas line (e.g., at a bend or elbow in the inlet gas line, which may or may not be close to the processing chamber 105). The second sensor 122 may also be a temperature sensor. The second sensor 122 can provide a measurement of the temperature of the connector 112 or the gas line (internal or external). In some embodiments, the second sensor 122 may be provided on or near the exhaust line of the processing chamber 105. The temperature sensor may also be located at other locations associated with the processing chamber 105, such as at various locations within the processing chamber 105, at the inlet of the processing chamber, at the outlet of the processing chamber, at one or more bends or other points in the gas line connected to the inlet of the processing chamber, at one or more bends or other points in the gas line connected to the outlet of the processing chamber, and so on.

[0044] In some embodiments, a plurality of sensors (such as first sensor 121 and second sensor 122) may be temperature sensors. Temperature sensors may include temperature control devices, resistance temperature devices, thermocouples, thermistors, or combinations thereof. In some embodiments, temperature sensors may include or be located near a catalytic material. Catalytic materials may include stainless steel, nickel (Ni), platinum (Pt), gold (Au), or combinations thereof. Catalytic materials exhibit an increased responsiveness to free radical fluxes, thereby increasing sensitivity to free radical fluxes.

[0045] Free radical flux refers to the flow rate or rate of free radical substances in a chemical reaction or process. Free radicals are highly reactive chemical substances containing unpaired electrons. They frequently participate in various chemical reactions, such as combustion, polymerization, and oxidation processes.

[0046] In chemical reactions, free radicals can be generated through various mechanisms, such as the homolytic cleavage of covalent bonds or the transfer of single electrons. Once formed, free radicals can react with other molecules, thereby initiating a chain reaction by generating new free radicals.

[0047] Free radical flux is a measure of the amount of free radicals produced or consumed per unit time in a given reaction or process. It provides information about the kinetics and kinematics of free radical reactions and is frequently used to study free radical chain reactions. By quantifying free radical flux, scientists can gain deeper insights into the mechanisms, reaction rates, and overall behavior of free radical-based processes.

[0048] Due to its unique properties and ability to generate and maintain high concentrations of free radicals, plasma is often associated with free radical matter. Plasma is considered a fourth state of matter, distinct from solids, liquids, and gases. It is an ionized gas composed of a mixture of ions, electrons, neutral atoms, and molecules.

[0049] Plasma can be generated by applying energy to a gas, causing ionization and the formation of reactive substances, including free radicals. Energy can be supplied through various methods, such as electrical discharges, electromagnetic fields, or powerful laser beams. When sufficient energy is input, electrons are stripped from atoms or molecules, resulting in the formation of positive ions and free electrons. Collisions between these charged particles and gas molecules can lead to the production of reactive substances, including free radicals.

[0050] In a plasma environment, free radicals can be formed through processes such as electron-bombardment dissociation, ion-molecule reactions, or reactions involving excited substances. The high energy and reactivity of the plasma environment promote the formation and propagation of free radicals, leading to complex and often non-equilibrium chemical reactions.

[0051] Plasma-generated free radicals are used in a variety of applications, such as plasma chemistry, surface modification, plasma etching, and plasma polymerization. These free radicals can initiate and drive chemical reactions that would otherwise be difficult to achieve under normal conditions. Furthermore, plasma free radicals can be used to degrade contaminants, sterilize, and synthesize advanced materials. The amount of free radical flux in plasma can affect processing parameters such as etching rate, deposition rate, and polymerization rate. Therefore, understanding the generation, behavior, and flux of free radicals in plasma systems helps optimize plasma-based processes and leverage the unique reactivity of free radicals in various applications.

[0052] In embodiments, a plurality of temperature sensors (e.g., thermocouples, RTDs, thermistors, semiconductor-based sensors, etc.) may be used to detect drift or shift in free radical flux associated with a particular process. In one embodiment, the plurality of sensors may provide temperature readings that can be compared with values ​​representing free radical flux. If the measured temperature exceeds a predetermined threshold near that value, an indication of drift or shift in free radical flux can be determined. In embodiments, the reference temperature and threshold are determined via machine learning or artificial intelligence (AI) applications trained on data from temperature sensors at corresponding locations on the processing tool 100. A more detailed explanation of the machine learning or AI process is described below.

[0053] See now Figure 1BThe illustration shows a plan view of a processing tool 100 according to an additional embodiment. In this embodiment, in addition to a plurality of sensors, Figure 1B The processing tool 100 in the middle can be substantially similar to Figure 1A The processing tool 100 in the embodiment. In this embodiment, the processing tool 100 may include, in addition to Figure 1A Additional sensors besides those shown. For example, a third sensor 123 may be provided within chamber 105. The third sensor 105 may be a temperature sensor. The third sensor 123 may be used to measure the temperature of the sidewalls or other surfaces within chamber 105.

[0054] In some embodiments, a fourth sensor 124 may also be used. The fourth sensor 124 may also be a temperature sensor. Figure 1B As shown, the fourth sensor 124 may be located within the exhaust system 104. The fourth sensor 124 may be located at any point within the exhaust system 104. For example, the fourth sensor 124 may be at the inlet of the exhaust system 104, at the pump, or after the pump.

[0055] In some embodiments, a fifth sensor 125 may also be used. In some embodiments, the fifth sensor 125 may be a temperature sensor. The fifth sensor 125 may be located within the RPS 115. For example, the fifth sensor 125 may measure the sidewall temperature of the chamber of the RPS 115.

[0056] Although described as a temperature sensor, additional sensors 123-125 may also include other types of sensors. For example, sensors 123-125 may include pressure sensors, optical sensors, and the like. Similar to the measurements from the temperature sensor, measurements from one or more other types of sensors may also be correlated with free radical flux, such as through the use of machine learning.

[0057] As shown, multiple different sensor locations can be used for detecting shifts in free radical flux. Including more sensors allows for improved drift detection. That is, depending on the mechanism causing the drift, some parts of the semiconductor processing tool 100 may drift before other parts of the semiconductor processing tool 100. For example, if the reflector plate 108 becomes dirty or a redeposited coating forms over the reflector plate 108, a temperature change in the first sensor 121 can be an initial indicator of drift before other temperatures begin to change.

[0058] It will be understood that the RPS can be located at various positions relative to the chamber. For example, in some embodiments, a top RPS configuration may be used. In this embodiment, the plasma enters the chamber from above. In other embodiments, a crossflow configuration may be used. In this embodiment, the RPS is located on the side of the chamber, and the plasma flows across the chamber. Although specific embodiments have been illustrated, it will be understood that any RPS configuration can be used in conjunction with the embodiments described herein.

[0059] In an embodiment, one or more sensors 121-125 are connected to a computing device that executes one or more trained machine learning models. The trained machine models can be trained to estimate values ​​representing free radical flux based on sensor measurements (such as temperature measurements at one or more locations). The machine learning model can be trained using data from specific sensors at specific locations and during specific types of processes. This training can be performed using data from known steady-state and / or health tools. In some embodiments, training is performed using different data associated with different known free radical fluxes. Based on the training, the machine learning model can be trained to output an accurate estimate of the free radical flux based on the input temperature. Alternatively, the machine learning model can be trained to output a value associated with the free radical flux. For example, this value can increase as the free radical flux increases and decrease as the free radical flux decreases.

[0060] In one embodiment, one or more systems in the trained machine learning model use regression models trained using regression. Instances of regression models use regression models trained using linear regression or Gaussian regression. Given known values ​​of variable X, the regression model predicts the value of Y. The regression model can be trained using regression analysis, which may include interpolation and / or extrapolation. In one embodiment, the parameters of the regression model are estimated using least squares. Alternatively, Bayesian linear regression, percentage regression, least absolute bias, nonparametric regression, scene optimization, and / or distance metric learning may be performed to train the regression model.

[0061] In one embodiment, the trained machine learning model includes one or more decision trees, random forests, support vector machines, or other types of machine learning models.

[0062] In one embodiment, one or more artificial neural networks (also simply referred to as neural networks) are used in the trained machine learning model. The artificial neural network may be, for example, a convolutional neural network (CNN) or a deep neural network. In one embodiment, the processing logic performs supervised machine learning to train the neural network.

[0063] Artificial neural networks typically include feature representation components with classifier or regression layers that map features to a target output space. For example, a convolutional neural network (CNN) manages multiple layers of convolutional filters. Pooling is performed at lower layers, typically with multiple layers of classifiers attached on top, and nonlinearities are resolved, thereby mapping the top-level features extracted by the convolutional layers to a decision (e.g., a classification output). Neural networks can be deep networks with multiple hidden layers or shallow networks with zero or a few (e.g., 1-2) hidden layers. Deep learning is a class of machine learning algorithms that use cascades of multiple layers of nonlinear processing units for feature extraction and transformation. Each successive layer uses the output of the previous layer as input. Neural networks can learn in a supervised (e.g., classification) and / or unsupervised (e.g., pattern analysis) manner. Some neural networks (e.g., such as deep neural networks) include layers, where different layers learn different levels of representation corresponding to different levels of abstraction. In deep learning, each level learns to transform its input data into a slightly more abstract and complex representation.

[0064] One or more components of a trained machine learning model can be a recurrent neural network (RNN). An RNN is a type of neural network that includes memory to enable it to capture time-dependent information. An RNN learns an input-output mapping that depends on the current input and past inputs. The RNN resolves past and future measurements and makes predictions based on this continuous measurement information. For example, sensor measurements can be acquired continuously during a manufacturing process, and these sets of measurements can be sequentially fed into an RNN. Current and previous sensor measurements can influence the current output of the trained machine learning model. One type of RNN that can be used is a long short-term memory (LSTM) neural network.

[0065] Trained machine learning models for the processing tool can be used in multiple different processing chambers that share a common chamber type, containing sensors (e.g., temperature sensors) at the same locations, and are used to perform the same or similar processes. For example, both the first and second processing chambers can be etching chambers performing the same etching process. The trained machine learning models can be used to determine when to schedule maintenance for each of the first and second processing chambers, when to stop the seasonalization process after maintenance, and so on.

[0066] Some trained machine learning models use all sensor measurements generated by and / or used in the processing chamber (e.g., for processes performed on the processing chamber). Other trained machine learning models use a subset of the generated sensor measurements. For example, a trained machine learning model designed to determine the endpoint of a quaternization formulation might receive measurements from one or more temperature sensors as input.

[0067] In one embodiment, a trained machine learning model processes temperature sensor measurements periodically (e.g., every 50-100 milliseconds) during a process such as a plasma process. For each input, the trained machine learning model may output a value representing the free radical flux. The method may then determine whether the value representing the free radical flux meets a criterion, wherein corrective actions (e.g., such as maintaining or stopping the quaternization process) may be scheduled or determined in response to the determination that the value representing the free radical flux meets the criterion. In one embodiment, the machine learning model is trained to make such a decision, and instead of outputting a value associated with the free radical flux, the machine learning model outputs suggestions, notifications, and / or instructions to perform corrective actions (e.g., performing maintenance or stopping the quaternization process). The criterion may include a free radical flux drift threshold, and the criterion may be met in response to a determined value representing the free radical flux reaching or exceeding the free radical flux drift threshold. In one embodiment, the trained machine learning model is a recurrent neural network (RNN). In another embodiment, the trained machine learning model is a neural network (e.g., a CNN). In one embodiment, the trained machine learning model is a linear regression model, and in another embodiment, the machine learning model is a Gaussian regression model. In one embodiment, the trained machine learning model is a random forest.

[0068] Training a neural network can be achieved through supervised learning, which involves feeding the network a training dataset consisting of labeled inputs, observing its outputs, defining the error (by measuring the difference between the output and the labeled values), and using techniques such as deep gradient descent and backpropagation to tune the network's weights across all layers and nodes to minimize the error. In many applications, this process is repeated on numerous labeled inputs in the training dataset to produce a network that can produce the correct output when presented with inputs different from those present in the training dataset. In high-dimensional settings (such as large images), this generalization is achieved when sufficiently large and diverse training datasets are available. When the system is in a known good state, training can be performed using temperature measurements taken during substrate processing. When temperature measurements acquired during processing after the processing chamber deviate from the known good state, this deviation can be detected by a machine learning model that can suggest maintenance. In some embodiments, training can be performed using the processing chamber after the quartering process is complete. After training the machine learning model, temperature measurements can be acquired periodically or continuously and fed into the model during the quaternization process. The model can then output an indication of whether the processing chamber has reached the target quaternization state. Once trained, the temperature measurements can be fed into the model, which can output an indication of free radical flux drift. For example, a value of 0 can represent a known good state for free radical flux, and + / - values ​​deviating from 0 can indicate a positive or negative drift relative to the known good state, where the magnitude of the value indicates the amount of drift relative to the known good state.

[0069] In some embodiments, each of the trained machine learning models of the processing tool can be retrained periodically or continuously to enable continuous learning and improvement of the trained machine learning models. Each model can produce an output based on the input, perform an action based on the output, and measure the result of the action. In some cases, the result of the action is measured in fractions of a second (e.g., milliseconds), seconds, or minutes, and in others, a longer time is required to measure the result of the action. For example, one or more additional processes may be performed before the result of the action can be measured. The action and the result of the action can indicate whether the output is the correct output and / or what the output should be and what the output is. Thus, the action and the result of the action can be used to determine a target output, which can be used as a marker for the sensor measurement result. Once the result of the action is determined, the input (i.e., the sensor measurement result), the output of the trained machine learning model, and the target output of the machine learning model (or the action and the result of the action) can be used as new training data items. The new training data items can then be used to further train the trained machine learning model.

[0070] In one embodiment, a process manager (not shown) may be included, comprising one or more trained machine learning models trained to detect free radical flux, i.e., drift in free radical flux. As discussed above, such trained machine learning models for detecting free radical flux can be trained on a training dataset including temperature measurements (e.g., temperature sensor measurements) and labels indicating the free radical flux and / or the known state of the processing chamber. In one embodiment, the temperature measurements provide temperatures at various locations within the processing chamber, which may be correlated with values ​​representing the free radical flux at each location based on inputs to one or more trained machine learning models. Processing logic can identify differences between first and second measurements and whether such differences meet criteria. For example, the trained machine learning model of the process manager can use the temperature measurements to determine when chamber maintenance should be performed.

[0071] In one embodiment, training data items, including sensor measurements, predictions about whether the processing chamber is ready for reuse, and machine learning outputs about whether the processing chamber is actually ready for reuse (e.g., an indication that the processing chamber has passed or failed a re-identification test), are used to update the training of a trained machine learning model. The trained machine learning model can be retrained each time a processing chamber (or another processing chamber) is decommissioned for maintenance and then put back into service. Embodiments reduce the number of repetitions of the quartering process performed before the processing chamber is put back into service after maintenance. For example, a standard process for quartering an etched chamber might be 25 iterations of the quartering process running on the etched chamber, followed by a test process performed on the processing chamber. However, in embodiments, the processing logic can immediately determine when the processing chamber is ready to run the test process, rather than waiting until the full 25 iterations of the quartering process have been completed. In some embodiments, the test process is not run after the trained machine learning model has indicated that the processing chamber is ready for reuse.

[0072] In one embodiment, a maintenance manager (not shown) includes one or more trained machine learning models trained to detect when maintenance should be performed on a processing chamber. Such trained machine learning models, trained to detect when processing chambers should be maintained, can be trained on a training dataset that includes a number of different measurements generated by one or more processing chambers during a process performed on a product substrate (e.g., on a product wafer). The number of different measurements may include a first temperature at a first location of the processing chamber and / or a second temperature at a second location of the processing chamber, and a flag indicating whether the processing chamber should be maintained after the process of acquiring combined sensor measurements is complete. In an example, a first temperature sensor may be included at the inlet of the processing chamber through which plasma flows during a plasma-based process, and a second temperature sensor may be included at the exhaust line of the processing chamber. Furthermore, incidental testing processes can be performed using test substrates, blanket substrates (substrates with an unpatterned, uniform coating), bare substrates, sensor substrates (substrates on which multiple sensors are disposed), etc. Sensor measurements from the processing chamber (and, in some cases, from the sensor substrate) can be generated and these results are fed into a trained machine learning model to produce an output.

[0073] Different maintenance predictive machine learning models can be trained for each processing chamber and / or for each pair of processing chambers and the processes or sets of processes performed on each processing chamber. Once trained, the machine learning models are used, sensor measurements can be generated periodically or continuously by one or more sensors (and / or sensor substrates) in the processing chamber during product manufacturing and / or incidental testing processes. These measurements can be processed by the maintenance manager's trained machine learning model to determine when a drift in free radical flux occurs, and thus when the processing chamber requires maintenance and when it should be removed for maintenance. Examples of maintenance include cleaning the processing chamber, replacing one or more parts of the processing chamber, replacing the RPS unit, replacing one or more components of the RPS unit, changing one or more settings of the plasma source, etc. In an embodiment, the maintenance predictive machine learning model identifies the type of maintenance that should be performed on the processing chamber and / or plasma source based on sensor measurements. For example, a trained machine learning model can indicate that the processing chamber should be cleaned, the protective liner should be replaced, the processing sleeve ring should be replaced, the spray head should be replaced, the plasma source should be replaced, and so on.

[0074] Once a processing chamber has been marked as ready for use, a technician can determine whether the processing chamber should actually be put into use and / or the type of maintenance that should be performed on it. In one embodiment, training data items are used to update the training of a trained machine learning model. These training data items include sensor measurements, predictions about whether the processing chamber should be maintained (and / or the type of maintenance to be performed), and indications about whether maintenance is actually necessary. The trained machine learning model can be retrained each time maintenance is scheduled for a processing chamber (or other processing chamber). Alternatively or additionally, the machine learning model can be continuously or periodically retrained using data points associated with the substrates processed by the processing chamber, where the data points include sensor measurements, temperature-based differences between values ​​representing the free radical flux associated with plasma-based processes, whether maintenance should be performed, or when a quaternization process should be performed after a maintenance event. This embodiment reduces the number of substrates processed by processing chambers that need to be put into use and additionally ensures that processing chambers are not put into use more frequently than necessary.

[0075] In various embodiments, the server may be and / or include computing devices such as personal computers, server computers, programmable logic controllers (PLCs), microcontrollers, etc. The server may include (or be) one or more processing devices, which may be general-purpose processing devices such as microprocessors, central processing units, or the like. More specifically, the processing devices may be complex instruction set computing (CISC) microprocessors, reduced instruction set computing (RISC) microprocessors, very long instruction word (VLIW) microprocessors, or processors implementing other instruction sets or combinations of instruction sets. The processing devices may also be one or more special-purpose processing devices, such as application-specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs), digital signal processors (DSPs), network processors, or the like. The server may include data storage devices (e.g., one or more disk drives and / or solid-state drives), main memory, static memory, a network interface, and / or other components. In embodiments, the server's processing means may execute instructions to train a machine learning model and send the trained machine learning model to a platform controller and / or a controller of a standalone tool (e.g., a controller of a processing chamber). The instructions may be stored on a computer-readable storage medium, which may include main memory, static memory, secondary storage, and / or processing means (during instruction execution).

[0076] Figure 2This is a cross-sectional view of a processing tool 200 including a processing chamber 202 and a chamber controller 205 operatively connected to the processing chamber 202. The chamber controller 205 may be mounted to the processing chamber 202 or may be located near the processing chamber (e.g., connected to another component of a substrate processing system). The processing chamber 202 may be an etching processing chamber, a deposition chamber, an annealing chamber, or other types of processing chambers for processing substrates (e.g., wafers) such as semiconductor substrates via plasma-based processes. For example, the processing chamber 202 may be a chamber for a plasma etcher or plasma etching reactor, a plasma cleaner, a CVD or ALD reactor (e.g., a plasma-enhanced CVD or ALD reactor), an ion-assisted deposition (IAD) chamber, a physical vapor deposition (PVD) chamber, and so on.

[0077] In one embodiment, the processing chamber 202 includes a chamber body enclosing an internal volume 206 and a spray head 230. The spray head 230 may include a spray head base and a spray head gas distribution plate. Alternatively, the spray head 230 may be replaceable by a cover and nozzle in some embodiments, or by multiple disc-shaped spray head compartments and a plasma generation unit in other embodiments. The chamber body may be made of aluminum, stainless steel, or other suitable materials such as titanium (Ti). The chamber body typically includes sidewalls 208 and a bottom 210. A liner 216 may be disposed adjacent to the sidewalls 208 to protect the chamber body.

[0078] An exhaust port 226 may be defined within the chamber body and may couple the internal volume 206 to a pump system 228. The pump system 228 may include one or more pumps and throttle valves for evacuating and regulating the pressure of the internal volume 206 of the processing chamber 202.

[0079] A spray head 230 (or cover) may be supported on the sidewall 208 of the chamber body. The spray head 230 (or cover) may open to allow access to the internal volume 206 of the processing chamber 202, and when closed, provide a seal for the processing chamber 202. A remote plasma source 258 may be coupled to the processing chamber 202 to deliver processing and / or cleaning gases and / or remote plasma through the spray head 230 or cover and nozzles to the internal volume 206. The spray head 230 may be used in processing chambers for dielectric etching (etching dielectric materials). The spray head 230 may include a gas distribution plate (GDP) and may have multiple gas delivery orifices 232 throughout the GDP. The spray head 230 may include a GDP bonded to an aluminum substrate or anodized aluminum substrate. The GDP may be made of Si or SiC, or may utilize Y2O3, Al2O3, or Y3Al5O3.12 (YAG) and other coatings on ceramics.

[0080] Examples of processing gases that can be used to process the substrate in processing chamber 202 include halogen-containing gases such as C2F6, SF6, SiCl4, HBr, NF3, CF4, CHF3, CH2F3, F, NF3, Cl2, CCl4, BCl3, and SiF4, as well as other gases such as O2 or N2O. Examples of carrier gases include N2, He, Ar, and other gases inert to the processing gas (e.g., non-reactive gases).

[0081] Heater assembly 248 may be disposed within internal volume 206 of processing chamber 202 beneath spray head 230 or cover. Heater assembly 248 includes support 250 for holding substrate 244 during processing. Support 250 is attached to an end of shaft 252 coupled to chamber body via a flange. Support 250, shaft 252, and flange may be made of, for example, a material containing AlN. Support 250 may further include a mesa (e.g., recess or protrusion). Support may additionally include wiring, such as tungsten wire (not shown), embedded in heater material of support 250. In one embodiment, support 250 may include a metal heater and sensor layer sandwiched between AlN ceramic layers. Such an assembly may be sintered in a high-temperature furnace to produce a monolithic assembly. Layers may include heater circuitry, sensor assemblies, ground planes, RF grids, and combinations of metal and ceramic flow channels.

[0082] Exemplary chamber components of the processing chamber 202 include, but are not limited to, electrostatic chucks, nozzles, gas distribution plates, spray heads (e.g., 230), electrostatic chuck components, chamber walls (e.g., 208), gaskets (e.g., 216), gasket sleeves, gas lines, chamber covers, nozzles, single rings, processing sleeve rings, edge rings, substrates, shielding, plasma screens, flow equalizers, cooling substrates, chamber viewing ports, bellows, any part of the heater assembly (including support 250, shaft 252, flanges), panels, baffles, and the like.

[0083] In this embodiment, the processing chamber includes a number of different sensors, including temperature sensors 235-236. Sensors may additionally or alternatively include optical sensors, such as optical emission spectrometers and / or reflectometers, pressure sensors, power sensors, other electrical sensors, flow rate sensors, and so on. Some sensors 235-236 may be located inside the processing chamber 202, while others may be located outside the processing chamber 202, measuring the flow and / or delivery of gases, power, etc., within the processing chamber 230. In one embodiment, sensor 235 may be located at the inlet of the processing chamber, and temperature sensor 236 may be located at the exhaust port of the processing chamber.

[0084] The chamber controller 205 may be configured to operate on one or more processing chambers (e.g., processing chamber 202) or on a platform containing multiple chambers (e.g., a tool cluster). For example, the chamber controller 205 may be configured to control an etching chamber of a cluster tool or an etching chamber performing a specific etching process. In an embodiment, the chamber controller 205 includes an autonomous tool engine 221, which may include a maintenance manager 223, a re-identification manager 225, and / or a process manager 227. For a single platform with multiple processing chambers attached thereto, each of the processing chambers may include its own dedicated chamber controller 205. Alternatively, some processing chambers attached to a cluster tool or host may share a common chamber controller. In one embodiment, instead of using the chamber controller 205, a platform controller is used to control all processing chambers attached to the cluster tool.

[0085] In some embodiments, the autonomous tool engine 221 uses sensor measurements (e.g., temperature sensor measurements) from one or more of sensors 235-236 to make decisions about the processing chamber 202. The chamber controller 205 can use the autonomous tool engine 221 to determine, for example, whether the processing chamber 202 should be maintained, the type of maintenance to be performed on the processing chamber 202, whether the processing chamber 202 is ready for reuse after maintenance and seasonalization, etc. In embodiments, the maintenance manager 223, process manager 227, and / or re-evaluation manager 225 may include one or more trained machine learning models trained to receive sensor measurements (e.g., temperature measurements) and make decisions about radical flux. Such decisions may be estimates of radical flux, estimates of the drift of radical flux relative to the radical flux of a processing chamber known to be in good condition, estimates of whether to perform maintenance, estimates of whether to stop the seasonalization process, etc.

[0086] Figures 3 to 4 This is a flowchart illustrating a method for training a machine learning model and / or using a trained machine learning model to make decisions for a processing chamber based on temperature sensor measurements, according to an embodiment. The method can be utilized with reference to Figure 1 to... Figure 2The components described are executed as will be apparent. For example, in an embodiment, the method may be executed by a chamber controller 205. At least some operations of the method may be executed by processing logic, which may include hardware (e.g., circuitry, dedicated logic, programmable logic, microcode, etc.), software (e.g., instructions running on a processing device to perform hardware emulation), or a combination thereof. Although illustrated in a particular sequence or order, the order of processes may be modified unless otherwise stated. Therefore, the illustrated embodiments should be understood as examples only, and the illustrated processes may be executed in different orders, and some processes may be executed in parallel. Furthermore, one or more processes may be omitted in various embodiments. Therefore, not all processes are executed in every embodiment. Other process flows are possible.

[0087] Figure 3 This is a flowchart of a method 300 for performing actions by a processing tool and / or substrate processing system according to an embodiment. At block 302 of method 300, processing logic causes a processing chamber to perform a process, such as an etching process, a deposition process, a plasma-based process, or a quaternization process. At block 305, the processing logic receives a first measurement from a temperature sensor at a first location associated with the processing chamber during and / or after the process (such as a plasma-based process). In some embodiments, the first location may include an inlet to the processing chamber through which plasma flows during a plasma-based process. In some embodiments, the temperature may be measured at a first time and a second time during a plasma-based process. In one embodiment, the plasma-based process may include a quaternization process for the processing chamber. In another embodiment, the process may include different plasma-based processes. A second value representing a second free radical flux based on a second temperature may be determined. The second value may be compared with a first value at the first location to identify the difference between the second and first values. The method may further include determining whether the difference meets a criterion and stopping the quartering process in response to the determination that the difference meets the criterion. In an embodiment, if the difference meets the criterion, maintenance can be scheduled.

[0088] At block 310, the first temperature measurement result is input into a trained machine learning model to determine and output a value representing the free radical flux. In one embodiment, the value indicates the amount of drift of the free radical flux from known favorable conditions. The magnitude of the value may indicate the amount of drift. Additionally or alternatively, the machine learning model may output instructions to perform one or more actions, such as maintenance actions, actions associated with the quaternization process (e.g., stopping or continuing the quaternization process), etc. At block 312, processing logic may determine whether the value representing the free radical flux meets a criterion. The criterion may be a threshold. If the value reaches or exceeds the threshold, actions such as maintenance actions may be suggested, scheduled, or initiated. In one embodiment, alternative to or in addition to outputting the value representing the free radical flux, the machine learning model may output an estimate of when the free radical flux drift will reach a threshold amount that should be addressed through maintenance. Thus, the machine learning model may output a recommendation to perform scheduled maintenance at a specified future time. In one embodiment, the machine learning model outputs a recommendation to currently perform one or more corrective actions (such as performing maintenance).

[0089] In one embodiment, at block 315, the processing logic receives a second measurement from a second temperature sensor at a second location within the processing chamber. The second location may include an exhaust line for the processing chamber. At block 320, the second temperature measurement is input to a trained machine learning model or a different trained machine learning model to determine and output a second value representing the mass flux and / or any of the other types of outputs discussed above. The outputs in blocks 310 and 320 (i.e., values ​​representing the free radical mass flux) may indicate whether the processing chamber should be scheduled for maintenance or whether the processing chamber should be put back into service. The trained machine learning model may have been trained as described above and may correspond to any of the trained machine learning models described above.

[0090] In one embodiment, at block 325, the processing logic determines that the output meets a criterion. The criterion may include a free radical flux drift threshold, a yes / no criterion, or some other criterion. In cases where a trained machine learning model is trained to detect values ​​representing free radical flux, the criterion may be a free radical flux drift threshold, and the criterion is met if the difference between a first value representation and a second value representation reaches or exceeds the free radical flux drift threshold. In one embodiment, the trained machine learning model outputs yes or no, where yes indicates that the free radical flux drift threshold has been reached, and no indicates that the free radical flux drift threshold has not been reached. In one embodiment, the trained machine learning model outputs yes or no, where yes indicates that maintenance should be performed on the processing chamber. Such operations may additionally or alternatively be performed at block 312.

[0091] The amount of drift can be correlated with the drift index. The drift index can be a value between 0 and 1. When there is no drift, the drift index is 0, and as drift increases, the drift index increases toward 1. In some embodiments, depending on the robustness of the given process, the temperature range can be correlated with a drift index of up to approximately 0.4, up to approximately 0.3, or up to approximately 0.2.

[0092] In one embodiment, the trained machine learning model outputs multiple maintenance categories, and for each maintenance category, the trained machine learning model provides "yes," indicating that the maintenance type associated with that maintenance category should be performed, or "no," indicating that the maintenance type associated with that maintenance category does not need to be performed. Examples of maintenance categories include scheduled cleaning, replacement of a first part, replacement of a second part, and so on. In one embodiment, the trained machine learning model outputs "yes" or "no," where "yes" indicates that further seasonalization of the processing chamber is unnecessary (and the processing chamber is ready for reuse), and "no" indicates that one or more seasonalization processes should still be performed on the processing chamber (and the processing chamber is not yet ready for reuse).

[0093] Figure 4 This is a flowchart of a method 400 for performing actions using a processing tool and / or a substrate processing system according to an embodiment. At block 402 of method 400, processing logic causes a processing chamber to perform a process, such as an etching process, a deposition process, a plasma-based process, or a quaternization process. At block 405, the processing logic receives a first measurement result from a temperature sensor at a first moment. During and / or after the process (such as a plasma-based process), the temperature sensor is placed in a location associated with the processing chamber. In some embodiments, the location may include an inlet to the processing chamber through which plasma flows during a plasma-based process. In some embodiments, the location may include an exhaust line to the processing chamber. In some embodiments, the plasma-based process may include a quaternization process for the processing chamber, or a process based on a different plasma.

[0094] At block 410, the first temperature measurement result is input into a trained machine learning model to determine and output a value representing the free radical flux. In one embodiment, the value indicates the amount of drift of the free radical flux from known favorable conditions. The magnitude of the value can indicate the amount of drift. Additionally or alternatively, the machine learning model can output instructions to perform one or more actions, such as maintenance actions, actions associated with the quaternization process (e.g., stopping or continuing the quaternization process), etc. At block 412, processing logic can determine whether the value representing the free radical flux meets a criterion. The criterion can be a threshold. If the value meets or exceeds the threshold, an action such as maintenance can be suggested, scheduled, or initiated. In one embodiment, alternatively or in addition to outputting the value representing the free radical flux, the machine learning model can output an estimate of when the free radical flux drift will reach a threshold amount that should be addressed through maintenance. Thus, the machine learning model can output a recommendation to perform scheduled maintenance at a specified future time. In one embodiment, the machine learning model outputs a recommendation to currently perform one or more corrective actions (such as performing maintenance).

[0095] In one embodiment, at block 415, the processing logic receives a second measurement result from a second temperature sensor in the processing chamber at a second time. At block 420, the second temperature measurement result is input into a trained machine learning model or a different trained machine learning model to determine and output a second value representing the mass flux and / or any of the other types of outputs discussed above. The outputs in blocks 410 and 420 (i.e., the values ​​representing the free radical mass flux) may indicate whether the processing chamber should be scheduled for maintenance or whether the processing chamber should be put back into use. The trained machine learning model may have been trained as described above and may correspond to any of the trained machine learning models described above.

[0096] In one embodiment, at block 425, the processing logic determines that the output meets a criterion. The criterion may include a free radical flux drift threshold, a yes / no criterion, or some other criterion. In cases where a trained machine learning model is trained to detect values ​​representing free radical flux, the criterion may be a free radical flux drift threshold, and the criterion is met if the difference between a first value representation and a second value representation reaches or exceeds the free radical flux drift threshold. In one embodiment, the trained machine learning model outputs yes or no, where yes indicates that the free radical flux drift threshold has been reached, and no indicates that the free radical flux drift threshold has not been reached. In one embodiment, the trained machine learning model outputs yes or no, where yes indicates that maintenance should be performed on the processing chamber. Such operations may additionally or alternatively be performed at block 412.

[0097] Figure 5This is a flowchart of a method 500 for training a machine learning model according to an embodiment. In an embodiment, the training method 500 can be performed to monitor a semiconductor processing tool, such as an RTP tool with RPS. At block 502 of method 500, a plurality of temperature measurements are collected to form a training dataset. In some embodiments, the training dataset includes a plurality of data items, wherein each data item may include a temperature measurement generated by a temperature sensor at a location associated with a processing chamber and a label indicating the state of the processing chamber (e.g., whether the processing chamber is in a known good condition). At block 405, processing logic receives the training dataset at the processing chamber.

[0098] At block 510, the machine learning model is trained using the training dataset of block 502 to produce a trained machine learning model. The trained machine learning model is trained to receive temperature measurements generated by a temperature sensor at that location and, based on temperature, determine a value representing the free radical mass flux associated with the plasma-based process. The machine learning model at block 510 can be further trained to predict, based on temperature, when to perform a quaternization process on the processing chamber after a maintenance event. In another embodiment, the machine learning model at block 510 can be further trained to predict, based on temperature, when to perform maintenance associated with the processing chamber.

[0099] Figure 6 This is a flowchart of a method 600 for automatically determining when to perform maintenance on a processing chamber according to an embodiment. At block 602 of method 600, processing logic initiates a process on a product substrate within the chamber. For example, the process may be an etching process, a deposition process, a plasma-based process, or some other process. The process may be performed on and / or can be performed on a product substrate having one or more films thereon to process the substrate thereon. At block 605, the processing logic receives one or more measurement results from a set of sensors in the processing chamber during and / or after the process. The measurement results may be, for example, temperature measurements acquired at locations where elevated free radical flux is known to occur. At block 610, the processing logic processes the measurement results using a trained machine learning model, which has been trained to determine whether maintenance should be performed on the processing chamber. The trained machine learning model may have been trained to produce outputs indicating values ​​representing free radical flux and / or outputs indicating whether maintenance is due and / or the type of maintenance to be performed.

[0100] At block 615, the processing logic determines whether the output of the trained machine learning model meets a criterion. In one embodiment, the processing logic compares the output value representing the free radical flux with a free radical flux drift threshold. If the value representing the free radical flux is higher than or equal to the free radical flux drift threshold, the processing logic determines that the output meets the criterion. If the value representing the free radical flux is lower than the free radical flux drift threshold, the criterion may not be met. In one embodiment, the output of the trained machine learning model is a yes / no indication of whether maintenance should be performed. If the output is yes, i.e., maintenance should be performed (or a specific type of maintenance should be performed), the criterion is met. If the output is no, i.e., maintenance should not be performed, the criterion is not met. If the criterion is not met, the method continues to block 620. If the criterion is met, the method proceeds to block 625.

[0101] At block 620, the processing logic initiates the process on the new substrate (after causing the robotic arm to remove the first substrate from the processing chamber and insert the new substrate into the processing chamber). The method then returns to block 605 and receives sensor measurements associated with the execution of the process on the new substrate. Furthermore, the method may proceed to block 635.

[0102] At block 625, the processing logic determines that the processing chamber should be maintained. At block 630, the processing logic may mark the processing chamber for maintenance (e.g., cleaning) and / or may proactively schedule cleaning of the processing chamber. At block 635, the processing logic may receive an instruction regarding whether maintenance should actually be performed on the processing chamber. At blocks 602 and / or 620, the processing logic may additionally or alternatively receive instructions regarding the status of the processing chamber and / or the critical dimension measurement results of the product substrate processed by the processing chamber. At block 640, the processing logic may update the training of the machine learning model based on at least one of the measurement results received at block 605, the output from block 610 indicating whether maintenance should be performed, and the instruction regarding whether maintenance should be performed and / or the difference between the measured critical dimension and the target critical dimension of the free radical flux at the location of the processing chamber. Thus, continuous learning can be performed to continuously update and improve the trained machine learning model. In one embodiment, retraining of the trained machine learning model can be performed on a tool on a controller, at which the trained machine learning model is deployed.

[0103] Figure 7This is a flowchart of a method 700 for automatically determining when to reactivate a processing chamber after maintenance has been performed, according to an embodiment. At block 702 of method 700, the processing logic initiates a quaternization process in the chamber. The quaternization process is a chamber conditioning process that causes the processing chamber to reach a known state. Appropriate quaternization or conditioning of the processing chamber after maintenance (e.g., after component replacement and / or after cleaning processes such as wet or dry cleaning processes) improves process repeatability between wafers. In one embodiment, the quaternization process causes the reactor surface to passivate with plasma-generated material, which can alter the reactive adhesion coefficient of free radicals. Chamber quaternization can be performed to ensure consistent reproduction of device critical dimensions by enabling the repetition of uniform plasmas with the same ion density, electron temperature, and flux between wafers. The process can be performed on blanket substrates, bare substrates, test substrates, etc.

[0104] At block 705, the processing logic receives one or more measurements from a set of sensors in the processing chamber during and / or after processing. At block 710, the processing logic processes the measurements using a trained machine learning model, which has been trained to determine whether quaternization is complete and / or whether the processing chamber is ready for reuse. The trained machine learning model may have been trained to produce outputs indicating estimates of free radical flux and / or indicating whether quaternization is complete (and the processing chamber is ready for reuse).

[0105] At block 715, the processing logic determines whether the output of the trained machine learning model meets a criterion. In one embodiment, the processing logic compares the output estimate representing the free radical flux with a threshold. If the estimated free radical flux is at or above the free radical threshold, the processing logic determines that the output meets the criterion. If the estimated free radical flux is below the threshold r, the criterion may not be met. In one embodiment, the output of the trained machine learning model is a yes / no indication of whether quartering is complete. If the output is no, i.e., quartering is not complete, the criterion is not met. If the output is yes, i.e., quartering is complete, the criterion is met. If the criterion is not met, the method continues to block 720. If the criterion is met, the method proceeds to block 725.

[0106] At block 720, the processing logic initiates another iteration of the quartering process and / or continues the quartering process, as appropriate, on a new substrate (after causing the robotic arm to remove the first substrate from the processing chamber and insert the new substrate into the processing chamber). The method then returns to block 705 and receives sensor measurements associated with the execution of the process on the new substrate.

[0107] At block 725, the processing logic determines that the processing chamber is ready for requalification and / or ready for reuse (to be used on a product substrate). At block 730, the processing logic may mark the processing chamber for qualification and / or schedule a requalification process. At block 735, the processing logic may receive an indication of whether the processing chamber has passed the requalification test. The indication may include one or more measurement results of one or more test substrates treated with a test formulation or test process. In one embodiment, a blanket wafer etching process is performed on a blanket wafer, a patterned wafer etching process is performed on a patterned wafer, and / or a particle testing process is performed on a particle wafer (e.g., which may be a blank wafer or a blanket wafer). According to the blanket wafer etching process, the average blanket wafer etching rate and blanket wafer etching uniformity can be measured. According to the patterned wafer etching process, the average patterned wafer etching rate and patterned wafer etching uniformity can be measured. After particle testing, the particles on the particle wafer can be counted. Measurement results may include, for example, particle count on the wafer, metal contamination, film thickness, film composition, blanket wafer etch rate, blanket wafer etch uniformity, patterned wafer etch rate, patterned wafer etch uniformity, etc. The processing logic may additionally or alternatively receive indications of the state of the processing chamber. Actual values ​​representing the free radical flux can be determined for the processing chamber based on the measurement results.

[0108] At block 740, the processing logic can update the training of the machine learning model based on the measurement results received at block 605, the indication from block 610 regarding whether maintenance output should be performed, and the indication and / or the result of the re-identification test regarding whether the processing chamber has passed the re-identification test. Thus, continuous learning can be performed to continuously update and improve the trained machine learning model. In an embodiment, retraining of the trained machine learning model can be performed on a tool on a controller, where the trained machine learning model is deployed.

[0109] Figure 8A graphical representation of a machine in the example form of a computing device 1000 is shown, in which a set of instructions can be executed to cause the machine to perform any one or more methodologies discussed herein. In alternative embodiments, the machine may be connected (e.g., network-connected) to a local area network (LAN), an intranet network, an extranet network, or other machines on the Internet. The machine may operate in a client-server network environment within the capacity of a server or client machine, or as a peer machine in a peer-to-peer (or distributed) network environment. The machine may be a personal computer (PC), tablet computer, set-top box (STB), personal digital assistant (PDA), cellular phone, network device, server, network router, switch, or bridge, or any machine capable of executing a set of instructions (continuously or otherwise) that specifies actions to be taken by that machine. Furthermore, although only a single machine is shown, the term "machine" should also be considered to include any collection of machines (e.g., computers) that independently or jointly execute instruction sets (or multiple instruction sets) to perform any one or more methodologies discussed herein.

[0110] Example computing device 1000 includes processing device 1002, main memory 1004 (e.g., read-only memory (ROM), flash memory, dynamic random access memory (DRAM) such as synchronous DRAM (SDRAM) or Rambus DRAM (RDRAM), etc.), static memory 1006 (e.g., flash memory, static random access memory (SRAM), hard disk (magnetic storage) etc.), and auxiliary memory (e.g., data storage device 1018), which communicate with each other via bus 1030.

[0111] Processing device 1002 represents one or more general-purpose processors, such as microprocessors, central processing units, or the like. More specifically, processing device 1002 may be a Complex Instruction Set Computing (CISC) microprocessor, a Reduced Instruction Set Computing (RISC) microprocessor, a Very Long Instruction Word (VLIW) microprocessor, a processor implementing other instruction sets, or a processor implementing a combination of instruction sets. Processing device 1002 may also be one or more special-purpose processing devices, such as application-specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs), digital signal processors (DSPs), network processors, or the like. Processing device 1002 is configured to execute processing logic (instructions 1022) for performing the operations and steps discussed herein.

[0112] The computing device 1000 may further include a network interface device 1008. The computing device 1000 may also include a video display unit 1010 (e.g., a liquid crystal display (LCD) or a cathode ray tube (CRT)), an alphanumeric input device 1012 (e.g., a keyboard), a cursor control device 1014 (e.g., a mouse), and a signal generation device 1016 (e.g., a speaker).

[0113] Data storage device 1018 may include machine-readable storage medium (or more specifically, computer-readable storage medium) 1028, thereon storing one or more sets of instructions 1022 that contain any of the methodologies or functions described herein. Instructions 1022 may also be wholly or at least partially residing in main memory 1004 and / or processing device 1002 during execution by computer device 1000, main memory 1004 which also constitutes computer-readable storage medium, and processing device 1002.

[0114] Computer-readable storage medium 1028 may also be used to store autonomous tool engine 121, and / or software link libraries containing methods that call autonomous tool engine 121. Although computer-readable storage medium 1028 is illustrated as a single medium in the example embodiment, the term "computer-readable storage medium" should be considered to include a single medium or multiple media (e.g., a centralized or distributed database, and / or associated caches and servers) that store one or more instruction sets. The term "computer-readable storage medium" should also be considered to include any medium capable of storing or encoding instruction sets for execution by a machine and causing the machine to execute any one or more methodologies disclosed herein. The term "computer-readable storage medium" should therefore be considered to include, but is not limited to, non-transitory computer-readable media, such as solid-state memory, and optical and magnetic media.

[0115] The modules, components, and other features described in this document (e.g., regarding...) Figure 1A , Figure 1B and Figure 2 Modules can be implemented as discrete hardware components or integrated into the functionality of hardware components such as ASICs, FPGAs, DSPs, or similar devices. Furthermore, modules can be implemented as functional circuit systems within firmware or hardware devices. Additionally, modules can be implemented in any combination of hardware devices and software components, or solely in software.

[0116] Some parts of the implementation have been presented with regard to algorithms and notational representations of operations on data bits within computer memory. These algorithmic descriptions and representations are means by which those skilled in the art of data processing most effectively communicate the essence of their work to others skilled in the art. Algorithms herein and generally are considered as a self-consistent sequence of steps leading to a target result. The steps are those that require the manipulation of physical quantities. Typically, although not essential, these quantities take the form of electrical or magnetic signals capable of being stored, transmitted, combined, compared, and otherwise manipulated. Primarily for general reasons, referring to these signals as bits, values, elements, symbols, characters, terms, numbers, or the like sometimes proves convenient.

[0117] However, it should be remembered that all such and similar terms will be associated with appropriate physical quantities and are merely convenient notations for use with such quantities. Unless otherwise specifically stated, it will be understood from the following discussion that throughout the description, the use of terms such as “receive,” “identify,” “determine,” “select,” “provide,” “store,” or similar expressions refers to the actions and processes of a computer system or similar electronic computing device that manipulate and transform data representing physical (electronic) quantities in the caches and memory of a computer system into other data representing physical quantities similarly represented in the memory or caches or other such information storage, transmission, or display devices of a computer system.

[0118] Embodiments of the present invention also relate to apparatus for performing the operations described herein. This apparatus may be specifically configured for the purposes discussed, or may comprise a general-purpose computer system selectively programmed by computer programs stored in a computer system. Such computer programs may be stored in computer-readable storage media, such as, but not limited to, any type of disk, including floppy disks, optical disks, CD-ROMs, and magneto-optical disks, read-only memory (ROM), random access memory (RAM), EPROM, EEPROM, disk storage media, optical storage media, flash memory devices, other types of machine-accessible storage media, or any type of media suitable for storing electronic instructions, each coupled to a computer system bus.

[0119] The preceding description has set forth several specific details, such as examples of specific systems, components, methods, etc., to provide a good understanding of several embodiments of this disclosure. However, it will be apparent to those skilled in the art that at least some embodiments of this disclosure can be practiced without such specific details. In other instances, well-known components or methods have not been described in detail and are presented in a simple block diagram format to avoid unnecessarily obscuring this disclosure. Therefore, the specific details set forth are merely exemplary. Specific implementations may vary from these exemplary details and are still contemplated within the scope of this disclosure.

[0120] Throughout this specification, references to "an embodiment" or "an embodiment" mean that a particular feature, structure, or characteristic described in connection with the embodiment is included in at least one embodiment. Therefore, the phrase "in an embodiment" appearing in various places throughout this specification does not necessarily refer to the same embodiment. Furthermore, the term "or" is intended to mean inclusive rather than exclusionary. When the terms "about" or "approximately" are used herein, it is intended to mean that the presented nominal values ​​are accurate within ±10%.

[0121] Although the operations of the methods described herein are illustrated and depicted in a specific order, the order of operations for each method may be changed, such that some operations may be performed in reverse order, or that some operations may be performed at least partially concurrently with other operations. In another embodiment, instructions or sub-operations of different operations may be performed intermittently and / or alternately.

[0122] It will be understood that the above description is intended to be illustrative and not restrictive. Numerous other embodiments will become apparent to those skilled in the art upon reading and understanding the above description. Therefore, the scope of this disclosure should be determined by referring to the entire scope of the appended claims and their equivalents.

Claims

1. A method comprising the following steps: During the plasma-based process, a first temperature is measured at a first location associated with the processing chamber; and The value representing the first free radical flux associated with the plasma-based process is determined based on this first temperature.

2. The method of claim 1, wherein the first location includes an inlet to the processing chamber through which plasma flows during the plasma-based process.

3. The method of claim 1, wherein the step of determining the value representing the free radical flux comprises the following steps: The temperature is input into a trained machine learning model, which outputs a value representing the flux of the free radical.

4. The method of claim 1, further comprising the following steps: During the plasma-based process, a second temperature is measured at a second location associated with the processing chamber; and A second value representing the flux of the second free radical material associated with the plasma-based process is determined based on this second temperature.

5. The method of claim 1, wherein the second location includes an exhaust line for the processing chamber.

6. The method of claim 1, further comprising the following steps: The determination indicates whether the value of the free radical flux meets the standard; and The schedule is maintained in response to the determination that the value of the free radical flux meets the standard.

7. The method of claim 1, wherein the temperature measurement is performed at a first instant, the method further comprising the following steps: During the plasma-based process, a second temperature at the first location is measured at a second time, wherein the plasma-based process includes a quaternization process for the processing chamber. The second value representing the flux of the second free radical substance is determined based on this second temperature; The second value is compared with the first value to identify the difference between the second value and the first value; Determine whether the difference between the second value and the first value meets the criteria; and The seasonalization process is stopped in response to the determination that the difference meets the standard.

8. The method of claim 1, wherein the temperature measurement is performed at a first instant, the method further comprising the following steps: During the second plasma-based process, a second temperature at the first location is measured at a second time, wherein the second plasma-based process is the same as the plasma-based process. The second value representing the flux of the second free radical substance is determined based on this second temperature; The second value is compared with the first value to identify the difference between the second value and the first value; Determine whether the difference between the second value and the first value meets the criteria; and The schedule is maintained in response to the determination that the difference meets the criterion.

9. The method of claim 8, wherein the criterion includes a free radical flux drift threshold, and wherein the criterion is satisfied in response to the difference reaching or exceeding the free radical flux drift threshold.

10. A method comprising the following steps: Receive a training dataset associated with a plasma-based process performed in a processing chamber, the training dataset containing a plurality of data items, wherein each of the training data items includes temperature measurements generated by a temperature sensor at a location associated with the processing chamber and a marker indicating the state of the processing chamber; and The training dataset is used to train a machine learning model to produce a trained machine learning model that is trained to receive temperature measurements generated by the temperature sensor at that location and to determine a value representing the free radical flux associated with the plasma-based process based on that temperature.

11. The method of claim 10, wherein the machine learning model is further trained to predict, based on the temperature, when to perform a seasonal process on the processing chamber after a maintenance event is completed.

12. The method of claim 10, wherein the machine learning model is further trained to predict when maintenance associated with the processing chamber will be performed based on the temperature.

13. A system comprising: The processing chamber is configured to perform plasma-based processes; A temperature sensor, located at a first position associated with the processing chamber, is used to generate one or more temperature measurements during the plasma-based process. as well as A computing device, wherein the computing device is configured to receive a first temperature measurement result generated by the temperature sensor during the plasma-based process; as well as The value representing the first free radical flux associated with the plasma-based process is determined based on this first temperature.

14. The system of claim 13, wherein the first location includes an inlet to the processing chamber through which plasma flows during the plasma-based process.

15. The system of claim 13, wherein, in order to determine the value representing the free radical flux, the computing device is used to input the temperature into a trained machine learning model, the model outputting the value representing the free radical flux.

16. The system of claim 13, further comprising: A second temperature sensor, located at a second position associated with the processing chamber, is used to generate one or more additional temperature measurements during the plasma-based process. The computing device is further configured as follows: Receive the second temperature measurement result; and A second value representing the flux of the second free radical material associated with the plasma-based process is determined based on the second temperature measurement result.

17. The system of claim 16, wherein the second location includes an exhaust line for the processing chamber.

18. The system of claim 13, wherein the computing device is further configured to: The determination indicates whether the value of the free radical flux meets the standard; and The schedule is maintained in response to the determination that the value of the free radical flux meets the standard.

19. The system of claim 13, wherein the measurement of the temperature is performed at a first moment, and the computing device is further configured to: Receive a second temperature measurement result generated by the temperature sensor at a second time, wherein the plasma-based process includes a quaternization process for the processing chamber; The second value representing the flux of the second free radical substance is determined based on this second temperature; The second value is compared with the first value to identify the difference between the second value and the first value; Determine whether the difference between the second value and the first value meets the criteria; and In response to the determination that the difference meets the criterion, the seasonalization process is stopped.

20. The system of claim 13, wherein the measurement of the temperature is performed at a first moment, and the computing device is further configured to: Receive a second temperature measurement result generated by the temperature sensor at a second time, wherein the second plasma-based process is the same process as the plasma-based process; The second value representing the flux of the second free radical substance is determined based on this second temperature; The second value is compared with the first value to identify the difference between the second value and the first value; Determine whether the difference between the second value and the first value meets the criteria; and The schedule is maintained in response to the determination that the difference meets the criterion.

21. A non-transitory computer-readable medium containing instructions that, when executed by a processor, cause the processor to perform operations including: During the plasma-based process, a first temperature is measured at a first location associated with the processing chamber; and The value representing the first free radical flux associated with the plasma-based process is determined based on this first temperature.

Images