Virtual measurement of conditions near a substrate using compressive sensing based on physical laws

Abstract

To provide a method and a storage medium for virtually measuring characteristics near a substrate in a manufacturing chamber using a compressed sensing machine learning model based on physical laws.SOLUTION: The method includes, during a process performed on a substrate 502 in a fabrication chamber 500, measuring a subset of values of a characteristic in the fabrication chamber, determining values of the characteristic in the fabrication chamber at locations excluded from the locations at which the measurements were taken, and performing a corrective action based on the determined characteristics.SELECTED DRAWING: Figure 5

Description

[Technical Field]

[0001] This specification relates to using physics-based compressed sensing machine learning models to virtually measure properties near a substrate in a fabrication chamber. [Background technology]

[0002] Chambers are used in many types of processing systems. Examples of chambers include etch chambers, deposition chambers, anneal chambers, and the like. Typically, a substrate, such as a semiconductor wafer, is placed in the chamber, and conditions within the chamber are established and maintained to process the substrate. To fully characterize and predict the behavior of processing conditions, it would be ideal to measure the property values ​​at the substrate. However, performing measurements at the substrate is often inconvenient or impossible. Many systems have less readily available, but more convenient, locations for measuring the values ​​of relevant operating properties, such as temperature, gas flow, and pressure. Summary of the Invention

[0003] The following is a simplified summary of the disclosure to provide a basic understanding of some aspects of the disclosure. This summary is not an extensive overview of the disclosure. It is not intended to identify key or critical elements of the disclosure, nor to delineate the scope or claims of particular embodiments of the disclosure. Its sole purpose is to present some concepts of the disclosure in a simplified form as a prelude to the more detailed description that is presented later.

[0004] In an aspect of the present disclosure, a method includes measuring values ​​of one or more properties at multiple locations within a fabrication chamber. The measurements are made by sensors located in the chamber during a process performed on a substrate. The method further includes providing the measured property values ​​as input to a trained machine learning model. The trained machine learning model can use the measurements at the locations where the sensors are located to infer property values ​​at other locations within the chamber, including the location of the substrate. The method further includes performing an action in consideration of the properties inferred by the trained machine learning model.

[0005] In another aspect of the present disclosure, a method includes training a machine learning model. The method further includes providing training input data to the machine learning model. The training input data includes measurements of property values ​​taken at locations within a fabrication chamber while a substrate is being processed, with at least a majority of the measurements being taken at locations not directly adjacent to the substrate. The method further includes providing target output data to the machine learning model. The target output includes a map of property values ​​within a region of the chamber that includes the location of the substrate. The method further includes providing the set of input and output data to the machine learning model to train the model to predict the map of property values ​​within a region of the chamber given measurements from multiple locations within the chamber.

[0006] In another aspect of the present disclosure, a system is disclosed that implements a method as discussed above. The system includes a computer-readable storage medium. The system further includes a processing device. The system is configured such that a computer-readable storage medium includes instructions that, when executed, cause the processing device to perform the method. The method performed by the processing device includes receiving measurements of a characteristic value during a manufacturing process in the manufacturing chamber from a sensor disposed in the manufacturing chamber. The method performed by the processing device further includes using the trained machine learning model to determine a characteristic value in a region of the manufacturing chamber that includes the location of the substrate. The method performed by the processing device further includes taking corrective action or sending an alert to a user device in consideration of the determined characteristic value.

[0007] The present disclosure is illustrated by way of example and not limitation in the figures of the accompanying drawings. [Brief explanation of the drawings]

[0008] [Figure 1] FIG. 1 is a block diagram illustrating an example system (example system architecture) according to some embodiments. [Figure 2] FIG. 1 is a block diagram illustrating an example dataset generator used to create a dataset for a machine learning model, according to some embodiments. [Figure 3] FIG. 1 is a block diagram illustrating a system for generating output data (e.g., compressed sensing data) according to some embodiments. [Figure 4A] 4A-4C are flow diagrams illustrating methods associated with compressive sensing predictive data to effect corrective action, according to some embodiments. [Figure 4B] 4A-4C are flow diagrams illustrating methods associated with compressive sensing predictive data to effect corrective action, according to some embodiments. [Figure 4C]4A-4C are flow diagrams illustrating methods associated with compressive sensing predictive data to effect corrective action, according to some embodiments. [Figure 5] 1 is a cross-sectional view illustrating a manufacturing chamber (e.g., a semiconductor wafer manufacturing chamber) according to some embodiments. [Figure 6] FIG. 1 is a block diagram illustrating a computer system according to some embodiments. DETAILED DESCRIPTION OF THE INVENTION

[0009] Described herein is a technology directed to the virtual measurement of property values ​​at the location of a substrate inside a manufacturing chamber using physics-based compressive sensing techniques. Manufacturing equipment is used to create substrates, such as semiconductor wafers. The properties of these substrates are controlled by the conditions under which the substrates are processed. Accurate knowledge of the properties inside an operating manufacturing chamber, particularly in the immediate vicinity of the substrate, can be used to predict the properties of the finished product, consistently produce substrates with the same properties, and adjust process parameters to account for variations in chamber design and construction.

[0010] In some cases, sensors can be placed at the substrate location, or sensors capable of measuring from a distance (e.g., IR temperature sensors) may be used. In some cases, such arrangements are impractical or impossible. When direct measurement of property values ​​near the substrate (e.g., on the substrate) is undesirable, empirical methods can be used to produce consistent substrates with selected properties. Such approaches involve experimentation, which can be costly in terms of time spent, materials used, energy and gas consumed, etc. Additionally, as fabrication chambers are used repeatedly, their properties tend to drift due to material accumulation, aging of parts, changes during cleaning procedures, etc. Any acquired knowledge for mapping input process parameters to output substrate properties deteriorates as chamber properties drift. A further problem with this conventional approach relates to chamber-to-chamber differences. Every new chamber design uses a new set of experiments to determine a new mapping of input parameters to finished product properties, but this mapping also degrades as the chamber ages. Additionally, manufacturing chambers are subject to tolerance ranges when constructed, meaning that each chamber is somewhat different, and even nominally identical chambers may perform differently. Not only are experiments performed to gain an accurate understanding of which input parameters lead to which products for each chamber type, but this process may be performed for each individual chamber, and the understanding of these characteristics may benefit from continuous updating. The problem is exacerbated when attempting to fabricate different types of substrates with characteristics outside the scope of previously gathered knowledge about the chamber's operation.

[0011] The disclosed methods and devices address at least some of these shortcomings of conventional approaches. The present disclosure utilizes physics-based compressed sensing to enable virtual measurement of properties at the substrate. In some embodiments, the disclosed methods include using a computer-aided, physics-based model. The physics-based model is configured to include all characteristics of the fabrication chamber of interest, including geometry, material properties, gas flows, etc. The physics-based model can then be run multiple times to account for a range of process parameters. For example, hardware parameters such as power supplied to the heater, gas mix used, gas flow rates and pressures, and substrate placement can all be varied. The output of these calculations using the physics-based model is a set of maps that indicate steady-state property values ​​within the fabrication chamber. The set of maps can span a significant portion of the parameter space relevant to substrate fabrication.

[0012] In some embodiments, the disclosed method further includes using the output of the physics-based model to construct a basis set that describes the possible combinations of property values ​​within the fabrication chamber. The conceptually simplest representation of properties within a chamber is generally to assign a value for each property to each point within the chamber. This can be thought of as a weighted combination of basis set elements, where each element of the basis set contains a value for the property at one particular point within the chamber (e.g., a temperature of 1 Kelvin) and assigns a null value for the property (e.g., zero Kelvin) to all other points within the chamber. Thus, the entire property value map of the chamber is constructed by adding together weighted combinations of each reference condition that assigns a non-zero temperature to all points within the chamber.

[0013] This conceptually simple representation is often not the most convenient. Instead, a basis set with more complex elements may be used, in which the characteristics within the chamber are represented as weighted additive combinations of reference condition elements. If the basis set is carefully chosen, the characteristic values ​​within the chamber may be adequately described using far fewer elements than would be required in a simple case in which all elements of the basis set contribute meaningfully. In some embodiments, the disclosed method includes constructing a basis set from the output of a physics-based model to adequately describe the characteristic values ​​within the chamber in a region of parameter space relevant to substrate fabrication using a weighted additive combination of fewer reference conditions (than in the simple representation). This reduced-dimensional basis set is referred to as a reduced model (ROM). In some embodiments, the ROM may be formed by performing proper orthogonal decomposition on the output of the physics-based model.

[0014] In some embodiments, the disclosed method further includes training a machine learning model using a limited subset of characteristic values ​​corresponding to characteristic values ​​at a plurality of locations within the fabrication chamber as input. The machine learning model is trained to produce as output a map of characteristic values ​​extending to locations beyond those used as input, including characteristic values ​​near the substrate. In some embodiments, the machine learning model is trained using a sparse regression model. Alternatively, other types of machine learning models, such as neural networks, support vector machines, etc., may be used.

[0015] In some embodiments, the disclosed method further includes providing measurements of a subset of property values ​​within the fabrication chamber during processing as input to a trained machine learning model. The method further includes receiving a prediction of the property near the substrate from the trained machine learning model as output. The disclosed method also provides for the use of a system including a processing device and a memory to implement a physics-based compressed sensing and virtual metrology method.

[0016] Aspects of the present disclosure provide technical advantages over conventional methods. By arranging a measurement device to be disposed proximate to a substrate, predictions of property values ​​(e.g., values ​​such as temperature at a point on a substrate) near the substrate during processing can be accurately inferred without compromising quality. Additionally, using a basis set informed by a physics-based model of the chamber, temperature mapping and variations from chamber to chamber, or changes within a single chamber over time, can be accounted for using the methods of the present disclosure.

[0017] 1 is a block diagram illustrating an example system 100 (example system architecture) according to some embodiments. System 100 includes a client device 120, a manufacturing device 124, a sensor 126, a measurement device 128, a compressed sensing server 112, and a data store 140. Compressive sensing server 112 may be part of a compressive sensing system 110. Compressive sensing system 110 may further include server machines 170 and 180.

[0018] The sensors 126 can provide sensor data 142 associated with the manufacturing equipment 124 (e.g., associated with the production of a corresponding product, such as a wafer, by the manufacturing equipment 124). The sensor data 142 can be used, for example, for equipment health and / or product health (e.g., product quality). The manufacturing equipment 124 can manufacture products according to a recipe or operate over a period of time. In some embodiments, the sensor data 142 can include one or more values ​​of temperature (e.g., heater temperature), spacing (SP), pressure, high frequency radio frequency (HFRF), electrostatic chuck (ESC) voltage, current, flow (e.g., of one or more gases), power, voltage, etc. The sensor data 142 can include historical sensor data 144 and current sensor data 146. The manufacturing equipment 124 can be configured according to manufacturing data 150. The manufacturing data 150 may be associated with or indicative of manufacturing parameters, such as hardware parameters (e.g., settings or components (e.g., size, type, etc.) of the manufacturing equipment 124) and / or process parameters of the manufacturing equipment. The manufacturing data 150 may include historical manufacturing data 152 and / or current manufacturing data 154. The manufacturing data 150 may be indicative of input settings for the manufacturing devices (e.g., heater power, gas flow, etc.). The sensor data 142 and / or manufacturing data 150 may be provided while the manufacturing equipment 124 is performing a manufacturing process (e.g., equipment readings as the product is processed). The sensor data 142 may be different for each product (e.g., each wafer).

[0019] In some embodiments, the sensor data 142, the metrology data 169, and / or the manufacturing data 150 may be processed (e.g., by the client device 120 and / or by the compressed sensing server 112). The processing of the sensor data 142 may include generating features. In some embodiments, the features are patterns (e.g., slope, width, height, peaks, etc.) in the sensor data 142, the metrology data 169, and / or the manufacturing data 150, or combinations of values ​​(e.g., power derived from voltage and current, etc.) from the sensor data 142, the metrology data 169, and / or the manufacturing data 150. The sensor data 142 may include features, which may be used by the compressive sensing component 114 to perform signal processing and / or to obtain predicted data 168 and / or compressed sensing data 166, and possibly to perform corrective action. The compressed sensing data 166 may be any data associated with the compressive sensing system 110, such as predicted sensor data for a position of a substrate during a manufacturing process.

[0020] Each instance (e.g., set) of sensor data 142 may correspond to a product (e.g., a wafer), a set of manufacturing equipment, a type of substrate produced by the manufacturing equipment, combinations thereof, etc. Each instance of metrology data 169 and manufacturing data 150 may similarly correspond to a product, a set of manufacturing equipment, a type of substrate produced by the manufacturing equipment, combinations thereof, etc. The data store may further store information relating sets of different data types, e.g., information indicating that a set of sensor data, a set of metrology data, and / or a set of manufacturing data are all associated with the same product, manufacturing equipment, type of substrate, etc.

[0021] In some embodiments, the compressive sensing system 110 may use supervised machine learning to generate the compressive sensing data 166 (e.g., a target output including virtual sensor data provided in the compressive sensing system 110).

[0022] The client device 120, the manufacturing equipment 124, the sensor 126, the measurement equipment 128, the compressed sensing server 112, the data store 140, the server machine 170, and the server machine 180 may be coupled to each other via a network 130 to generate compressed sensing data 166 and / or predictive data 168, and optionally to implement corrective action.

[0023] In some embodiments, network 130 is a public network that gives client device 120 access to compressed sensing server 112, data store 140, and / or other publicly available computing devices. In some embodiments, network 130 is a private network that gives client device 120 access to manufacturing equipment 124, sensors 126, metrology equipment 128, data store 140, and / or other privately available computing devices. Network 130 may include one or more wide area networks (WANs), local area networks (LANs), wired networks (e.g., Ethernet networks), wireless networks (e.g., 802.11 networks or Wi-Fi networks), cellular networks (e.g., Long Term Evolution (LTE) networks), routers, hubs, switches, server computers, cloud computing networks, and / or combinations thereof.

[0024] The client device 120 may include a computing device such as a personal computer (PC), a laptop, a mobile phone, a smartphone, a tablet computer, a netbook computer, a network-connected television ("smart TV"), a network-connected media player (e.g., a Blu-ray player), a set-top box, an over-the-top (OTT) streaming device, an operator box, etc. The client device 120 may include a corrective action component 122. The corrective action component 122 may receive user input of an indication associated with the manufacturing equipment 124 (e.g., via a graphic user interface (GUI) displayed via the client device 120). In some embodiments, the corrective action component 122 sends the indication to the compressive sensing system 110, receives output (e.g., compressive sensing data 166) from the compressive sensing system 110, determines a corrective action based on the output, and performs the corrective action.

[0025] 1B , the system 110 may further include a prediction component 116. The prediction component 116 may generate predicted data 168 using data retrieved from the compressive sensing component 114. In some embodiments, the prediction component 116 provides the predicted data 168 to the client device 120, which takes corrective action in light of the predicted data 168 via the corrective action component 122. In some embodiments, the corrective action component 122 obtains sensor data 142 (e.g., current sensor data 146) associated with the manufacturing equipment 124 (e.g., from the data store 140, etc.) and provides the sensor data 142 (e.g., current sensor data 146) associated with the manufacturing equipment 124 to the compressive sensing system 110.

[0026] In some embodiments, the corrective action component 122 stores the sensor data 142 in the data store 140, and the compressed sensing server 112 retrieves the sensor data 142 from the data store 140. In some embodiments, the compressed sensing server 112 may store the output of the trained machine learning model 190 (e.g., the compressed sensing data 166) in the data store 140, and the client device 120 may retrieve the output from the data store 140. In some embodiments, the corrective action component 122 receives an indication of a corrective action from the compressed sensing system 110 and implements the corrective action. Each client device 120 may include an operating system that enables a user to one or more of create, view, or edit data (e.g., an indication associated with a manufacturing equipment 124, a corrective action associated with a manufacturing equipment 124, etc.).

[0027] In some embodiments, metrology data 192 corresponds to historical characteristic data of a product (e.g., generated using manufacturing parameters associated with historical sensor data 144 and historical manufacturing data 152), and predictive data 168 is associated with predicted characteristic data (e.g., of a product that will be or has been generated under conditions recorded by current sensor data 146 and / or current manufacturing data 154). In some embodiments, predictive data 168 is predicted metrology data (e.g., virtual metrology data) of a product that will be or has been generated according to conditions recorded as current sensor data 146 and / or current manufacturing data 154. In some embodiments, predictive data 168 is or includes an indication of an anomaly (e.g., an abnormal product, an abnormal component, an abnormal manufacturing equipment, an abnormal energy usage, etc.) and / or one or more causes of the anomaly. In some embodiments, predictive data 168 includes an indication of change or drift over time in some components, such as manufacturing equipment 124, sensors 126, or metrology equipment 128. In some embodiments, the forecast data 168 includes an indication of the end of life of components such as manufacturing equipment 124 , sensors 126 , metrology equipment 128 , and the like.

[0028] Performing a manufacturing process that results in a defective product is costly in terms of time, energy, product, components, manufacturing equipment 124, the cost of identifying the defects and discarding the defective product, etc. By inputting sensor data 142 (e.g., manufacturing parameters used to manufacture the product), receiving compressed sensing data 166 output, and performing corrective action based on the compressed sensing data 166, system 100 may have the technical advantage of avoiding the costs of creating, identifying, and discarding defective products.

[0029] Running a manufacturing process that results in a component failure of manufacturing equipment 124 can result in costly downtime, product damage, equipment damage, dispatch of replacement components, etc. By inputting sensor data 142 (e.g., manufacturing parameters currently being used or to be used to manufacture a product), receiving compressed sensing data 166 output, comparing compressed sensing data sets from the same chamber over time to diagnose drifting or failed components (recorded as predictive data 168), and performing corrective action (e.g., predicted operational maintenance such as component replacement, processing, cleaning, etc.) based on the predictive data 168, system 100 may have the technical advantage of avoiding the costs of one or more of unexpected component failures, unscheduled downtime, lost productivity, unexpected equipment malfunctions, product scrap, etc. Monitoring the performance of components, such as manufacturing equipment 124, sensors 126, metrology devices 128, etc., over time may provide an indication of deteriorating components.

[0030] The manufacturing parameters may be suboptimal for manufacturing a product, which may have costly consequences such as increased consumption of resources (e.g., energy, coolant, gas, etc.), increased time to manufacture the product, increased component failures, increased quantity of defective products, etc. By inputting the sensor data 142 into the trained machine learning model 190, receiving the output of the compressed sensing data 166, and performing corrective action (e.g., based on the compressed sensing data 166) to update the manufacturing parameters (e.g., set optimal manufacturing parameters), the system 100 may have the technical advantage of using optimal manufacturing parameters (e.g., hardware parameters, process parameters, optimal design) that avoid the costly consequences of suboptimal manufacturing parameters.

[0031] The corrective action may be associated with one or more of computational process control (CPC), statistical process control (SPC) (e.g., SPC for electronic components that determines the process being controlled, SPC that predicts the useful life of a component, SPC that compares to a 3 sigma graph, etc.), advanced process control (APC), model-based process control, preventative operational maintenance, design optimization, manufacturing parameter updates, manufacturing recipe updates, feedback control, machine learning corrections, etc.

[0032] In some embodiments, the corrective action includes providing an alert (e.g., an alarm that stops or prevents a manufacturing process from running if the compressed sensing data 166 or the predictive data 168 indicates a predicted anomaly, such as an anomaly in a product, component, or manufacturing equipment 124). In some embodiments, the corrective action includes providing feedback control (e.g., modifying a manufacturing parameter in response to the compressed sensing data 166 or the predictive data 168 indicating an anomaly). In some embodiments, the corrective action includes providing machine learning (e.g., modifying one or more manufacturing parameters based on the predictive data 168). In some embodiments, implementing the corrective action includes updating one or more manufacturing parameters.

[0033] The manufacturing parameters may include hardware parameters (e.g., replacing a component, using a specific component, replacing a processing chip, updating firmware, etc.) and / or process parameters (e.g., temperature, pressure, flow rate, speed, current, voltage, gas flow, lift speed, etc.). In some embodiments, the corrective action includes performing preventive operational maintenance (e.g., replacing, treating, cleaning, etc., of a manufacturing tool 124 component). In some embodiments, the corrective action includes performing design optimization (e.g., updating manufacturing parameters, manufacturing process, manufacturing tool 124, etc., for an optimized product). In some embodiments, the corrective action includes updating a recipe (e.g., that the manufacturing tool 124 is in idle mode, sleep mode, warm-up mode, etc.).

[0034] The compressed sensing server 112, the server machine 170, and the server machine 180 may each include one or more computing devices, such as a rack-mounted server, a router computer, a server computer, a personal computer, a mainframe computer, a laptop computer, a tablet computer, a desktop computer, a graphics processing unit (GPU), an accelerator application-specific integrated circuit (ASIC) (e.g., a tensor processing unit (TPU)), etc.

[0035] The compressive sensing server 112 may include a compressive sensing component 114. In some embodiments, the compressive sensing component 114 may receive (e.g., received from the client device 120, retrieved from the data store 140) the current sensor data 146 and / or the current manufacturing data 154 and generate an output (e.g., compressed sensing data 166) for implementing corrective actions associated with the manufacturing equipment 124 based on the current data. In some embodiments, the compressive sensing component 114 may use one or more trained machine learning models 190 to determine the output for implementing corrective actions based on the current data.

[0036] The compressed sensing server 112 may also include a prediction component 116. The prediction component 116 may be used to generate prediction data 168. In some embodiments, the prediction data 168 includes indications of aging, degradation, etc. of components of the manufacturing equipment 124. In some embodiments, the prediction component 116 may incorporate the compressed sensing data 166 and consider virtual measurements of conditions within the manufacturing chamber to determine predictions of wafer characteristics. In some embodiments, the prediction component 116 includes machine learning models.

[0037] In some embodiments, the sensor data 142 is provided to a trained machine learning model 190. The machine learning model 190 is trained to output data indicative of property values ​​near (e.g., at) the location of the substrate in the fabrication chamber based on the input sensor data 142. The model 190, in some embodiments, may utilize a sparse regression algorithm. In some embodiments, the machine learning model 190 takes as input temperature data from a number of sensors disposed at locations within the fabrication chamber. In some embodiments, at least a majority of the sensors are not at the location of the substrate. For example, temperature sensors may be disposed at various locations near a gas inlet, an exhaust conduit, a substrate support, a chamber wall, etc. The model 190 then produces as output temperature data at all points within the chamber. In some embodiments, the model 190 may produce as output property values ​​at areas of interest within the chamber, including at least locations near the substrate (excluding property values ​​at other areas not of interest). Producing data relating to only a subset of locations within the chamber provides the technical advantage of utilizing fewer sensors 126 to reliably virtually measure property values ​​on the substrate.

[0038] In some embodiments, model 190 produces a set of coefficients as output. These coefficients are used as weights to generate weighted additive combinations of the basis set elements. The weighted additive combinations of the basis sets represent a map of property values ​​within the fabrication chamber. The basis sets utilized by model 190, in embodiments, are generated by condensed model generator 174 on server machine 170. The basis sets may be stored as condensed basis data 164 in data store 140. The basis sets may be chosen as a way to represent a map of property values ​​within the fabrication chamber for different fabrication input parameters using a small number of basis set elements (e.g., a small number of coefficients output by model 190 are non-zero). The value of a parameter that determines the number of non-zero coefficients output by model 190 may be determined during training of model 190.

[0039] In some embodiments, the condensed model generator 174 generates a basis set by considering the output of the physics-based model 176. In some embodiments, the physics-based model 176 is used to solve equations that describe steady-state conditions in the fabrication chamber for various input parameters (such as power supplied to the heater, gas mix, and gas flow). The selected input parameters may span a region of parameter space deemed relevant to substrate fabrication. For each change in the input data, the physics-based model 176 generates a map of property values ​​corresponding to predicted property values ​​at steady state in the fabrication chamber given the input data. Data corresponding to the physics-based model may be stored in the data store 140 as physics-based model data 160, including storing the generated map of property values ​​as model property map data 162. The condensed model generator 174 takes as input a set of property value maps generated by the physics-based model 176 (possibly retrieved from the model property map data 162 in the data store 140). The condensed model generator 174 generates a basis set that describes the group of property value maps. The generated basis set is stored as reduced basis data 164 in data store 140. Reduced model generator 174 may generate the basis set using a proper orthogonal decomposition algorithm.

[0040] Historical sensor data 144 may be used in combination with current sensor data 146 and manufacturing data 150 to detect drift, changes, aging, etc., in components of manufacturing equipment 124. Predictive component 116 may use a combination and comparison of these data types to generate predictive data 168. In some embodiments, predictive data 168 includes data that predicts the lifespan of components such as manufacturing equipment 124, sensors 126, etc.

[0041] Combining sparse regression and physics-based modeling offers technical advantages over other techniques. Accurate physics-based models provide reliable maps of characteristic values ​​within a manufacturing chamber. However, it is not always possible to perfectly model a manufacturing chamber due to manufacturing tolerance ranges, aging components, and so on. For example, a heater may provide slightly less or more energy than expected, a gas flow regulator may not precisely allow the selected flow rate, contact between surfaces within the chamber may be somewhat less than ideal, and so on. Such variations may be unknown to the user and may not be captured by a purely physics-based model. These limitations can be overcome by implementing a machine learning sparse regression model that performs virtual measurements via compressed sensing. The machine learning model is trained for various input parameters spanning a region of parameter space. Variations in actual conditions within the chamber within the range of parameter space that the machine learning model is trained to operate on (due to model inaccuracies, as discussed above) can be accounted for by interpolation, and settings that lie somewhat outside the region of parameter space used to train the machine learning model can be accounted for by extrapolation. In this way, unexpected variations in the chamber operating conditions can be incorporated into the results from the compressive sensing system 110.

[0042] In some embodiments, the compressive sensing component 114 receives the current sensor data 146 and may perform preprocessing, such as extracting patterns in the data or combining the data into new composite data. The compressive sensing component 114 may then provide the data as input to a trained machine learning model 190. The compressive sensing component 114 may receive a map of characteristic values ​​from the trained machine learning model 190, including values ​​at the substrate's location. The compressive sensing component 114 may then take corrective action. The corrective action may include sending an alert to the client device 120. The corrective action may also include updating manufacturing parameters of the manufacturing tool 124. The corrective action may also include generating predictive data 168 indicative of chamber or equipment drift, aging, or failure.

[0043] Data store 140 may be memory (e.g., random access memory), a drive (e.g., a hard drive, a flash drive), a database system, or another type of component or device capable of storing data. Data store 140 may include multiple storage components (e.g., multiple drives or multiple databases) that may span multiple computing devices (e.g., multiple server computers). Data store 140 may store sensor data 142, manufacturing data 150, physics-based model data 160, compressed sensing data 166, prediction data 168, and metrology data 169. Sensor data 142 may include historical sensor data 144 and current sensor data 146. Sensor data may include time traces of sensor data over the duration of a manufacturing process, associations of data with physical sensors, preprocessed data such as averages and composite data, and data indicative of sensor performance over time (i.e., many manufacturing processes). Manufacturing data 150 and metrology data 169 may include similar characteristics. The historical sensor data 144 and the historical manufacturing data 152 may be historical data (e.g., at least a portion for training the machine learning model 190). The current sensor data 146 may be current data (e.g., at least a portion for inputting the trained machine learning model 190 following the historical data) for which the compressed sensing data 166 was generated (e.g., for implementing corrective action). The metrology data 169 may be metrology data of fabricated substrates, as well as sensor data, manufacturing data, and model data corresponding to those products. The metrology data 169 may be utilized to design processes for making additional substrates. The predictive data 168 may include predictions of which metrology data will result from a set of input parameters. The predictive data 168 may also include data indicative of aging and failure of components of the system 100.

[0044] In some embodiments, the compressive sensing system 110 further includes a server machine 170 and a server machine 180. The server machine 170 includes a dataset generator 172 that can generate datasets (e.g., a set of data inputs and a set of target outputs) for training, validating, and / or testing the machine learning model 190. Some operations of the dataset generator 172 are described in detail below with respect to FIGS. 2 and 4A . In some embodiments, the dataset generator 172 may divide the historical data (e.g., historical sensor data 144, historical manufacturing data 152) and the physical model data (e.g., model characteristic map data 162, reduced basis data 164) into a training set (e.g., 6% of the data), a validation set (e.g., 20% of the data), and a test set (e.g., 20% of the data). In some embodiments, the compressive sensing system 110 generates multiple sets of features (e.g., via the compressive sensing component 114). For example, the first set of features may correspond to a first set of types of sensor data (e.g., from a first set of sensors, a first combination of values ​​from the first set of sensors, a first pattern in values ​​from the first set of sensors) corresponding to each of the datasets (e.g., a training set, a validation set, and a test set), and the second set of features may correspond to a second set of types of sensor data (e.g., from a first set of sensors different from the first set of sensors, a second combination of values ​​different from the first combination, a second pattern different from the first pattern) corresponding to each of the datasets.

[0045] Server machine 180 includes a training engine 182, a validation engine 184, a selection engine 185, and / or a test engine 186. The engines (e.g., training engine 182, validation engine 184, selection engine 185, and test engine 186) may refer to hardware (e.g., circuitry, dedicated logic, programmable logic, microcode, processing device, etc.), software (e.g., instructions executed on a processing device, general-purpose computer system, or dedicated machine), firmware, microcode, or a combination thereof. Training engine 182 can train machine learning model 190 using one or more sets of features associated with a training set from dataset generator 172. Training engine 182 can generate multiple trained machine learning models 190, where each trained machine learning model 190 corresponds to a distinct set of features of the training set (e.g., sensor data from a distinct set of sensors). For example, a first trained machine learning model may have been trained using all features (e.g., X1-X5), a second trained machine learning model may have been trained using a first subset of features (e.g., X1, X2, X4), and a third trained machine learning model may have been trained using a second subset of features (e.g., X1, X3, X4, and X5) that may partially overlap with the first subset of features. A dataset generator 172 may receive the trained machine learning model (e.g., 190) output, assemble the data into training, validation, and test datasets, and use the datasets to train a second machine learning model.

[0046] The validation engine 184 may validate the trained machine learning models 190 using a corresponding set of validation set features from the dataset generator 172. For example, a first trained machine learning model 190 trained using a first set of training set features may be validated using a first set of validation set features. The validation engine 184 may determine the accuracy of each of the trained machine learning models 190 based on the corresponding set of validation set features. The validation engine 184 may discard trained machine learning models 190 having an accuracy that does not meet a threshold accuracy. In some embodiments, the selection engine 185 may select one or more trained machine learning models 190 having an accuracy that meets the threshold accuracy. In some embodiments, the selection engine 185 may select the trained machine learning model 190 with the highest accuracy of the trained machine learning models 190.

[0047] The testing engine 186 may test the trained machine learning model 190 using a corresponding set of test set features from the dataset generator 172. For example, a first trained machine learning model 190 trained using a first set of training set features may be tested using a first set of test set features. The testing engine 186 may determine the trained machine learning model 190 with the highest accuracy from all of the trained machine learning models based on the test set.

[0048] The machine learning model 190 may refer to a model artifact created by the training engine 182 using a training set that includes data inputs and corresponding target outputs (correct answers for each training input). Patterns in the data set that map the data inputs to the target outputs (correct answers) can be found, and the machine learning model 190 is provided with a mapping that captures these patterns. In some embodiments, the machine learning model 190 utilizes a sparse regression algorithm to reconstruct property values ​​for substrates in a fabrication chamber from multiple values ​​provided by sensors disposed in the fabrication chamber, at least the majority of which are not located at the substrate's location. In some embodiments, the machine learning model 190 may use LASSO regression to generate property values ​​for the fabrication chamber.

[0049] The compressed sensing component 114 may provide the current sensor data 146 to the trained machine learning model 190 and may run the trained machine learning model 190 on the input to obtain one or more outputs. The compressed sensing component 114 may determine (e.g., extract) the compressed sensing data 166 from the output of the trained machine learning model 190 and may determine (e.g., extract) confidence data from the output indicating a level of confidence that the compressed sensing data 166 is an accurate predictor of a process associated with the input data for a product generated or to be generated using the manufacturing equipment 124 with the current sensor data 146. The prediction component 116 may determine prediction data 168, including predictions for finished substrate properties and predictions of the useful life of components of the manufacturing equipment 124, sensors 126, or metrology equipment 128, based on the output of the model 190. The prediction component 116 may be able to determine confidence data indicating a level of confidence that the prediction data 168 is an accurate predictor of a specified event. The compressive sensing component 114, the prediction component 116, or the corrective action component 122 may use the confidence data to determine whether to take corrective action associated with the manufacturing equipment 124 based on the compressive sensing data 166 and / or the prediction data 168.

[0050] The confidence data may include or indicate a level of confidence. As an example, the prediction data 168 may indicate characteristics of a finished wafer given a set of manufacturing inputs. The confidence data may indicate that the prediction data 168 is an accurate prediction of a product associated with at least a portion of the input data. In one example, the level of confidence is a real number between 0 and 1, inclusive, where 0 indicates no confidence that the prediction data 168 is an accurate prediction of a product processed according to the input data and 1 indicates absolute confidence that the prediction data 168 accurately predicts the characteristics of a product processed according to the input data. In response to the confidence data for a predetermined number of instances (e.g., a percentage of instances, a frequency of instances, a total number of instances, etc.) indicating a level of confidence below a threshold level, the prediction component 116 may retrain the trained machine learning model 190 (e.g., based on the current sensor data 146, the current manufacturing data 154, etc.).

[0051] For purposes of explanation and not limitation, aspects of the present disclosure describe using historical data (e.g., historical sensor data 144, historical manufacturing data 152) to train one or more machine learning models 190 and inputting current data (e.g., current sensor data 146, current manufacturing data 154) into the one or more trained machine learning models 190 to determine the compressed sensing data 166. In other implementations, the compressed sensing data is determined using a heuristic or rule-based model (e.g., without using a trained machine learning model, perhaps by specifying, rather than training, values ​​for parameters that control the sparsity of the fit). The compressed sensing component 114 and the prediction component 116 may monitor the historical sensor data 144, historical manufacturing data 152, and metrology data 169. Any of the information described with respect to the data input 210 in FIG. 2 may be monitored or otherwise used in the heuristic or rule-based models.

[0052] In some embodiments, the functionality of client device 120, compressive sensing server 112, server machine 170, and server machine 180 may be provided by fewer machines. For example, in some embodiments, server machines 170 and 180 may be combined into a single machine, and in other embodiments, server machine 170, server machine 180, and compressive sensing server 112 may be combined into a single machine. In some embodiments, client device 120 and compressive sensing server 112 may be combined into a single machine.

[0053] In general, functionality described in one embodiment as being performed by client device 120, compressive sensing server 112, server machine 170, and server machine 180 may also be performed by compressive sensing server 112 in other embodiments, where appropriate. In addition, functionality attributed to a particular component may be performed by different or multiple components operating together. For example, in some embodiments, compressive sensing server 112 may determine corrective actions based on predicted data 168. In another example, client device 120 may determine predicted data 168 based on output from a trained machine learning model.

[0054] Additionally, the functionality of a particular component may be performed by different or multiple components working together. One or more of the compressed sensing server 112, server machine 170, or server machine 180 may be accessed as a service offered to other systems or devices through an appropriate application programming interface (API).

[0055] In some embodiments, a "user" may be described as a single individual. However, other embodiments of the present disclosure encompass a "user" that is an entity controlled by multiple users and / or automated sources. For example, a collection of individual users working together as a group of administrators may be considered a "user."

[0056] Embodiments of the present disclosure may be applied to data quality assessment, feature enhancement, model evaluation, virtual metrology (VM), predictive maintenance (PdM), marginal optimization, and the like.

[0057] Although embodiments of the present disclosure are discussed in terms of generating compressed sensing data 166 and predictive data 168 to implement corrective actions at a manufacturing facility (e.g., a semiconductor manufacturing facility), embodiments may also be applied generally to improving data processing by utilizing physics-based virtual measurements and compressed sensing. Embodiments may also be applied generally to characterizing and monitoring based on different types of data.

[0058] 2 is a block diagram of an example dataset generator 272 (e.g., dataset generator 172 of FIG. 1 ) used to create a dataset for a machine learning model (e.g., model 190 of FIG. 1 ), according to some embodiments. Dataset generator 272 may be part of server machine 170 of FIG. 1 . In some embodiments, system 100 of FIG. 1 includes multiple machine learning models. In such cases, each model may have a separate dataset generator, or the models may share a dataset generator.

[0059] Referring to FIG. 2 , a system 200 including a dataset generator 272 (e.g., dataset generator 172 of FIG. 1 ) creates a dataset for a machine learning model (e.g., model 190 of FIG. 1 ). The dataset generator 272 may create the dataset using data retrieved as output from a computer-aided, physics-based model. In some embodiments, the dataset generator 272 creates training inputs by selecting a subset of points in the physics-based model (e.g., a subset of points in model characteristic map data 162 of FIG. 1 and the output from physics-based model 176 of FIG. 1 ). For example, the output of the physics-based model may be a map of temperatures at all locations within a chamber. The dataset generator 272 may form a training set by extracting temperature values ​​from a number of these locations. The locations used as training inputs correspond to the locations of sensors (e.g., sensor 126 of FIG. 1 ) within a fabrication chamber. The placement of these sensors may be selected by a user or may be output by a machine learning model trained to optimize the placement of sensors within a fabrication chamber for compressed sensing. In some embodiments, the training input 210 is simulated data representing data that may be taken by physical sensors, including a set of simulated sensor data 262A-Z. The dataset generator 272 also generates target outputs 220 for training the machine learning model. The target outputs include a map of characteristic values ​​230 within the fabrication chamber, including characteristic values ​​located away from the location of the sensors within the chamber. In some implementations, the target outputs 220 include a map of characteristics within the chamber near the location of the substrate. The map 230 of characteristic values ​​may include values ​​across the entire chamber, values ​​in an area of ​​interest, values ​​only near the substrate, etc. The input training data 210 and the target output data 220 are provided to a machine learning model (e.g., model 190 of FIG. 1 ).

[0060] It is within the scope of this disclosure that target output 220 may be represented in a variety of different ways. Map 230 of property values ​​may be virtual measurements of property values ​​away from virtual sensor locations, or may be coefficients indicating weights to be applied to members of the basis set of the reduced model to generate the virtual measurements.

[0061] 2 , in some embodiments, the dataset generator 272 generates a dataset (e.g., a training set, a validation set, a test set) that includes one or more data inputs 210 (e.g., training inputs, validation inputs, test inputs) and may include one or more target outputs 220 corresponding to the data inputs 210. The dataset may also include mapping data that maps the data inputs 210 to the target outputs 220. The data inputs 210 may also be referred to as “features,” “attributes,” or “information.” In some embodiments, the dataset generator 272 may provide a dataset to the training engine 182, the validation engine 184, or the test engine 186 of FIG. 1 , where the dataset is used to train, validate, or test the machine learning model 190 of FIG. 1 . Some embodiments of generating a training set may be further described with respect to FIG. 4A .

[0062] In some embodiments, the dataset generator 272 may generate a first data input corresponding to a first set of simulated sensor data 262A for training, validating, or testing a first machine learning model, and the dataset generator 272 may generate a second data input corresponding to a second set of simulated sensor data 262B for training, validating, or testing a second machine learning model.

[0063] In some embodiments, the dataset generator 272 may perform operations on one or more of the data inputs 210 and the target outputs 220. The dataset generator 272 may extract patterns from the data (slope, curvature, etc.), combine the data (average, feature generation, etc.), or separate simulated sensors into groups that train separate models.

[0064] The data inputs 210 and target outputs 220 for training, validating, or testing a machine learning model may include information about a particular manufacturing chamber (e.g., a particular semiconductor wafer manufacturing chamber). The data inputs 210 and target outputs 220 may include information about a particular manufacturing chamber design (e.g., used for all chambers of that design).

[0065] In some embodiments, the information used to train the machine learning model may be information from a particular type of manufacturing equipment at a manufacturing facility (e.g., manufacturing equipment 124 in FIG. 1 ) having particular characteristics, allowing the trained machine learning model to determine outcomes for a particular group of manufacturing equipment 124 based on inputs to current sensor data (e.g., current sensor data 146) associated with one or more components that share the characteristics of the particular group. In some embodiments, the information used to train the machine learning model may be information about components from more than one manufacturing facility, allowing the trained machine learning model to determine outcomes for components based on inputs from one manufacturing facility.

[0066] In some embodiments, following generation of the dataset and training, validation, or testing of the machine learning model using the dataset, the machine learning model may be further trained, validated, or tested, or tuned.

[0067] 3 is a block diagram illustrating a system 300 for generating output data (e.g., compressed sensing data 166 of FIG. 1) according to some embodiments. System 300 may be used to virtually measure values ​​of properties near a substrate during a manufacturing process using inputs from multiple sensors disposed away from the substrate and compressed sensing techniques.

[0068] 3, in block 310, system 300 (e.g., a component of compressive sensing system 110 of FIG. 1) performs computer-aided data partitioning (e.g., via dataset generator 172 of server machine 170 of FIG. 1) of the physics-based model data 360 (e.g., extracting simulated sensor data 362 from the full model dataset) to generate a training set 302, a validation set 304, and a test set 306. For example, the training set may be 60% of the simulated data, the validation set may be 20% of the simulated data, and the test set may be 20% of the simulated data.

[0069] At block 312, system 300A performs model training (e.g., via training engine 182 of FIG. 1 ) using training set 302. System 300 may train multiple models using multiple feature sets of training set 302 (e.g., a first feature set including a group of simulated sensors in training set 302, a second feature set including a different group of simulated sensors in training set 302, etc.). For example, system 300 may train machine learning models to generate a first trained machine learning model using a first set of features from the training set and generate a second trained machine learning model using a second set of features from the training set (e.g., data different from the data used to train the first machine learning model). In some embodiments, the first trained machine learning model and the second trained machine learning model may be combined to generate a third trained machine learning model (e.g., which may be a better predictor than the first or second trained machine learning models themselves). In some embodiments, the sets of features used to compare models may overlap (e.g., one model may be trained using simulated sensors 1-15, and a second model may be trained using simulated sensors 10-20). In some embodiments, hundreds of models may be generated, including models with various permutations of model features and combinations.

[0070] At block 314, the system 300 performs model validation (e.g., via validation engine 184 of FIG. 1 ) using the validation set 304. The system 300 may validate each trained model using a corresponding set of features from the validation set 304. For example, the validation set 304 may use a subset of simulated sensors that is the same as that used in the training set 302, but for different input conditions. In some embodiments, the system 300A may validate hundreds of models (e.g., models including various permutations of features, model combinations, etc.) generated at block 312. At block 314, the system 300 may determine the accuracy of each of the one or more trained models (e.g., through model validation) and determine whether one or more of the trained models have an accuracy that meets the threshold accuracy. In response to determining that none of the trained models have an accuracy that meets the threshold accuracy, flow returns to block 312, and the system 300 performs model training using a different set of features from the training set. In response to determining that one or more of the trained models have an accuracy that meets the threshold accuracy, flow continues to block 316. The system 300 may discard trained machine learning models that have an accuracy below the threshold accuracy (e.g., based on a validation set).

[0071] In block 316, the system 300 performs model selection (e.g., via the selection engine 185 of FIG. 1 ) to determine which of the one or more trained models that meet the threshold accuracy has the highest accuracy (e.g., selected model 308 based on validation of block 314). In response to determining that two or more of the trained models that meet the threshold accuracy have the same accuracy, flow may return to block 312, where the system 300 performs model training using a further refined training set corresponding to a further refined set of features to determine the trained model with the highest accuracy.

[0072] At block 318, the system 300 performs model testing (e.g., via the test engine 186 of FIG. 1 ) using the test set 306 to test the selected model 308. The system 300 may test the first trained machine learning model using a first set of features of the test set (e.g., simulated sensors 1-15) and determine that the first trained machine learning model meets a threshold accuracy (e.g., based on the first set of features of the test set 306). In response to the accuracy of the selected model 308 not meeting the threshold accuracy (e.g., the selected model 308 is overfitted to the training set 302 and / or the validation set 304 and is inapplicable to other datasets, such as the test set 306), flow continues to block 312, where the system 300 performs model training (e.g., retraining) using a different training set, possibly corresponding to a different set of features, or performs a reorganization of the substrate divided into training, validation, and test sets. In response to a determination that the selected model 308 has an accuracy that meets the threshold accuracy based on the test set 306, flow continues to block 320. At least in block 312, the model may learn patterns of simulated sensor data to make predictions, and in block 318, the system 300 may apply the model to the remaining data (e.g., the test set 306) to test the predictions.

[0073] In block 320, the system 300 receives current sensor data 346 (e.g., current sensor data 146 of FIG. 1 ) using the trained model (e.g., selected model 308), determines (e.g., extracts) compressed sensing data 366 (e.g., compressed sensing data 166 of FIG. 1 ) from the output of the trained model, and performs an action (e.g., performs a corrective action associated with manufacturing equipment 124 of FIG. 1 , provides an alert to client device 120 of FIG. 1 , etc.).

[0074] In some embodiments, retraining of a machine learning model is accomplished by supplying additional data to further train the model. Current sensor data 352 may be provided in block 312. Additional physics-based model data 361 may be provided as well. This data may differ from the data originally used to train the model by incorporating input parameters that were not part of the original training, combinations of input parameters outside the parameter space covered by the original training, or may be updated to reflect chamber-specific knowledge (e.g., differences from an ideal chamber due to manufacturing tolerance ranges, component aging, etc.). The selected model 308 may be retrained based on this data.

[0075] In some embodiments, one or more of steps 310-320 may be performed in various orders and / or may involve other steps not presented or described herein. In some embodiments, one or more of steps 310-320 may not be performed. For example, in some embodiments, one or more of data partitioning of block 310, model validation of block 314, model selection of block 316, or model testing of block 318 may not be performed.

[0076] 4A-4C are flow diagrams illustrating methods 400A-C associated with generating compressed sensing data for effecting corrective action, according to some embodiments. Methods 400A-C may be implemented by processing logic, which may include hardware (e.g., circuitry, dedicated logic, programmable logic, microcode, a processing device, etc.), software (e.g., instructions executed on a processing device, a general-purpose computer system, or a dedicated machine), firmware, microcode, or a combination thereof. In some embodiments, methods 400A-C may be implemented in part by compressive sensing system 110. Method 400A may be implemented in part by compressive sensing system 110 (e.g., server machine 170 and dataset generator 172 in FIG. 1 , dataset generator 272 in FIG. 2 ). Compressive sensing system 110 may use method 400A to generate a dataset for at least one of training, validating, or testing a machine learning model, according to embodiments of the present disclosure. Method 400B may be performed by server machine 180 (e.g., training engine 182, etc.). Method 400C may be performed by compressive sensing server 112 (e.g., compressive sensing component 114). In some embodiments, a non-transitory storage medium stores instructions that, when executed by a processing device (e.g., of compressive sensing system 110, of server machine 180, of compressive sensing server 112, etc.), cause the processing device to perform one or more of methods 400A-C.

[0077] For ease of explanation, methods A-C are shown and described as a series of steps. However, steps according to the present disclosure may be performed in various orders and / or simultaneously, and with other steps not presented or described herein. Moreover, not all illustrated steps may be performed to implement methods 400A-C in accordance with the disclosed subject matter. In addition, those skilled in the art will understand and appreciate that methods 400A-C may alternatively be represented as a series of interrelated states, such as by a state diagram or events.

[0078] FIG. 4A is a flow diagram of a method 400A for generating a dataset for a machine learning model that generates compressed sensing data (e.g., compressed sensing data 166 of FIG. 1) according to some embodiments.

[0079] Referring to FIG. 4A, in some embodiments, at block 401, processing logic performing method 400A initializes a training set T to an empty set.

[0080] At block 402, processing logic generates a first data input (e.g., a first training input, a first validation input), which may include sensor data (e.g., simulated sensor data 262 of FIG. 2), etc. In some embodiments, the first data input may include a first set of features for a type of data, and the second data input may include a second set of features for a type of data (e.g., as described with respect to FIG. 3).

[0081] At block 403, processing logic generates a first target output for one or more of the data inputs (e.g., the first data input). In some embodiments, the first target output is a property value received as an output of the physics-based model. In some embodiments, the target output may be a coefficient indicating a weighted additive combination of basis set members in the reduced model.

[0082] At block 404, processing logic optionally generates mapping data indicating an input / output mapping. The input / output mapping (or mapping data) may refer to data inputs (e.g., one or more of the data inputs described herein), target outputs for the data inputs, and associations between the data inputs and the target outputs.

[0083] At block 405, processing logic, in some embodiments, adds the mapping data generated at block 404 to dataset T.

[0084] At block 406, processing logic branches based on whether dataset T is sufficient for at least one of training, validation, and / or testing of machine learning model 190 of FIG. 1. If so, execution proceeds to block 407; if not, execution returns to block 402. It should be noted that in some embodiments, the sufficiency of dataset T may be determined solely based on the number of inputs in the dataset, which in some embodiments are mapped to outputs; in other embodiments, the sufficiency of dataset T may be determined based on one or more other criteria in addition to or instead of the number of inputs (e.g., a measure of diversity of data examples, accuracy, etc.).

[0085] At block 407, processing logic provides dataset T (e.g., to server machine 180 of FIG. 1 ) for training, validating, and / or testing machine learning model 190. In some embodiments, dataset T is a training set and is provided to training engine 182 of server machine 180 to perform training. In some embodiments, dataset T is a validation set and is provided to validation engine 184 of server machine 180 to perform validation. In some embodiments, dataset T is a test set and is provided to test engine 186 of server machine 180 to perform testing. For a sparse regression machine learning model, input values ​​of a given input / output mapping (e.g., numerical values ​​associated with data input 210 of FIG. 2 ) are inputs to the sparse regression model, and output values ​​of the input / output mapping (e.g., numerical values ​​associated with target output 220 of FIG. 2 ) are stored as target outputs. Additionally, a basis set representing a reduced model is provided to the sparse regression model. The machine learning model learns to fit the coefficients of a reduced model to the input sensor data in a sparse manner, i.e., many of the coefficients are zero. This regression is performed by penalizing both the inaccuracy of the fit and the number of non-zero coefficients.

[0086] By way of example and not limitation, the mechanics of sparse regression are briefly considered below. A sparse regression model is for recovering a list of property values ​​at various locations in a chamber, which may be represented as a vector x. The basis set may be represented as a matrix P, and the problem reduces to finding a vector a such that x = Pa. With sensors located at those locations, a vector representing known property values ​​in the chamber, called y, is considered. One can say y = Cx, where C is a sparse matrix of 1s and 0s representing which data points x of the chamber property value are known. The problem can then be expressed as CPa = y, which is an overdetermined system (there are more sensors y than there are non-zero basis set element coefficients a). Sparse regression is a method for finding solutions to such systems.

[0087] Over many training sets, the machine learning model chooses an appropriate value for a parameter that controls the number of non-zero coefficients in the reduced model. After block 407, the machine learning model (e.g., machine learning model 190) can be at least one of trained using the training engine 182 of the server machine 180, validated using the validation engine 184 of the server machine 180, or tested using the testing engine 186 of the server machine 180. The trained machine learning model may be implemented by the compressive sensing component 114 (of the compressive sensing server 112) to generate compressed sensing data 166 for performing signal processing or for implementing corrective actions associated with the manufacturing equipment 124.

[0088] FIG. 4B is a method 400B for training a machine learning model (eg, model 190 of FIG. 1) to determine compressed sensing data and perform corrective action.

[0089] Referring to FIG. 4B , at block 410 of method 400B, processing logic receives data from a physics-based model. The data indicates values ​​of properties in a steady-state fabrication chamber for a particular configuration of input settings (process parameters, hardware settings, etc.). The physics-based model data may be represented as coefficients of a basis set in a reduced model. The model data may be retrieved by the processing logic from memory (e.g., data store 140 of FIG. 1 ) rather than directly from the physics-based model. In some cases, more than one machine learning model may be part of a single composite machine learning model. In this case, training one component of this composite model may involve receiving an output from another component of the model as a training input for the component of the model being trained.

[0090] At block 412, processing logic generates a simulated sensor data set. This data may be a subset of data points from the physics-based model data received at block 410. The specific data points to be used as the simulated sensor may be determined by determining which data points correspond to locations within the fabrication chamber where the sensor can be located. This data is mapped to the physics-based model data set of which it is a subset.

[0091] At block 414, processing logic provides the set of simulated sensor data as training inputs to a machine learning model. Processing logic provides the physics-based model data as target outputs to the machine learning model. The machine learning model is then trained to provide virtual sensing of property values ​​within the fabrication chamber by performing sparse regression on the sensor data to build a map of property values ​​within the chamber at specific property values ​​near the substrate.

[0092] At block 416, the machine learning model may be retrained using additional data from the physics-based model. In one embodiment, the training of the machine learning model is continuously updated to account for or predict drift in manufacturing equipment, sensors, metrology devices, etc., to predict equipment failures, reflect procedures or recipes, etc. Processing logic provides the additional data from a physics-based model different from the original as additional training input data. Processing logic provides simulated sensor data from the physics-based model data as target output data.

[0093] FIG. 4C is a method 400C for using a trained machine learning model (eg, model 190 of FIG. 1) to determine compressive sensing data.

[0094] 4C, in block 420 of method 400C, processing logic provides current sensor data associated with the substrate manufacturing process to the trained machine learning model. The sensor data types may correspond to the sensor data types of block 412 of method 400B for training the machine learning model.

[0095] At block 422, processing logic obtains one or more outputs from the trained machine learning model indicative of compressed sensing data. In some embodiments, the compressed sensing data may be temperature data, including data representative of a location near the substrate within the fabrication chamber.

[0096] At block 424, processing logic causes corrective action to be taken. In some embodiments, the corrective action may be taken considering the output of the machine learning model after the output data has been further processed, for example, after coefficients of the reduced model have been applied as weights to additional combinations of basis set elements to restore the characteristic value in the chamber. The types of corrective actions consistent with the present disclosure may be wide-ranging. In some embodiments, taking the corrective action may include one or more of: providing an alert to a user; interrupting the functionality of the manufacturing equipment; updating manufacturing parameters, including process parameters and hardware parameters; scheduling replacement of a component of the manufacturing equipment; placing one or more components in a sleep or idle mode at a specific time during the production of the product to reduce energy usage; replacing one or more components to reduce energy usage; performing preventative maintenance; modifying a component (e.g., tightening mounting screws, replacing bindings, etc.); correcting sensor drift of a sensor associated with the manufacturing equipment; correcting chamber drift; updating a process recipe; etc. The predictive data and / or corrective action may indicate a combination (e.g., a combination of components, a combination of manufacturing parameters) that is causing the anomaly (e.g., when only one of the items in the combination would not, by itself, cause the anomaly).

[0097] 5 illustrates a cross-sectional view of a fabrication chamber 500 (e.g., a semiconductor wafer fabrication chamber) according to some embodiments of the present disclosure. The fabrication chamber 500 may be one or more of an etch chamber, a deposition chamber (including atomic layer deposition, chemical vapor deposition, physical vapor deposition, or plasma-enhanced versions thereof), an anneal chamber, etc. For example, the fabrication chamber 500 may be a chamber for a plasma etcher, a plasma cleaner, etc. Example chamber components may include a substrate support assembly 504, an electrostatic chuck, a ring (e.g., a process kit ring), chamber walls, a base, a showerhead 506, a gas distribution plate, a liner, a liner kit, a shield, a plasma screen, a flow balancer, a cooling base, a chamber viewport, a chamber lid, a nozzle, etc.

[0098] In one embodiment, the fabrication chamber 500 includes a chamber body 508 and a showerhead 506 that encloses an interior volume 510. In some chambers, the showerhead 506 may be replaced with a lid and nozzle. The chamber body 508 may be constructed from aluminum, stainless steel, or other suitable materials. The chamber body 508 generally includes sidewalls 512 and a bottom 514.

[0099] An exhaust port 516 may be defined in the chamber body 508 and may couple the interior volume 510 to a pumping system 518. The pumping system 518 may include one or more pumps and valves used to evacuate and adjust the pressure in the interior volume 510 of the fabrication chamber 500.

[0100] The showerhead 506 may be supported by a sidewall 512 of the chamber body 508 or by a top portion of the chamber body. The showerhead 506 (or, in some embodiments, a lid) may be open to provide access to the interior volume 510 of the fabrication chamber 500 and may provide a seal for the fabrication chamber 500 when closed. A gas panel 520 may be coupled to the fabrication chamber 500 to provide process or cleaning gases to the interior volume 510 through the showerhead 506 (or lid and nozzle). The showerhead 506 may include multiple gas delivery holes throughout. Examples of process gases that may be used to process substrates in the fabrication chamber 500 include halogen-containing gases such as C2F6, SF6, SiCl4, HBr, NF3, CF4, CHF3, F2, Cl2, CCl4, BCl3, and SiF4, as well as other gases such as O2 or N2O, among others. Examples of carrier gases include N2, He, Ar, and other gases inert to the process (eg, non-reactive gases).

[0101] A substrate support assembly 504 is disposed within the interior volume 510 of the fabrication chamber 500 below the showerhead 506. In some embodiments, the substrate support assembly 504 includes a susceptor 522 and a shaft 524. The substrate support assembly 504 supports a substrate during processing. In some embodiments, one or more heaters 526 and reflectors 528 are also disposed within the fabrication chamber 500.

[0102] Understanding the value of a property in a fabrication chamber near the substrate is useful for controlling process results. In many situations, it is impractical to monitor conditions at the substrate location. In other situations, it is possible to monitor conditions in some portions of the substrate, but not over the entire area. The disclosed methods and systems provide a way to infer the value of a property at a substrate location without placing a sensor at that location.

[0103] In some embodiments, virtual measurements of property values ​​near the substrate are obtained by performing compressive sensing (e.g., using the compressive sensing system 110 of FIG. 1). If it is inconvenient to place a sensor to monitor conditions within the fabrication chamber 500 near the substrate 502, there are many other areas within the fabrication chamber where sensors can be more easily located. Such locations may include the susceptor 522, the shaft 524, the reflector 528, the exhaust port 516, the chamber body 508, etc.

[0104] Using a computer-aided, physics-based model (e.g., physics-based model 176 of FIG. 1 ), measurements of property values ​​at these accessible locations enable virtual measurements of conditions at the substrate's location. To utilize the physics-based model, a set of input parameters is defined, such as heater power, gas composition, gas flow, chamber pressure, hardware parameters, etc. These different input conditions define a region of parameter space, which can be understood as the entire spectrum of possible input parameters for a manufacturing process. The set of input parameters may be selected to be tightly grouped in the parameter space around conditions useful for a particular process, or may be selected using another metric to span substantially the entire region of parameter space deemed likely to be useful for substrate processing.

[0105] A representation of the fabrication chamber, along with process parameters representing one possible set of conditions, is received by a physics-based model. The model then uses computational techniques to solve equations describing fluid flow, thermodynamics, heat transfer, etc., to recover a map of property values ​​within the fabrication chamber. The map may be a representation of steady-state property values. This process is repeated for selected combinations of process parameters. The group of property value maps resulting from this procedure spans a region of property value space within the fabrication chamber that is associated with the region of parameter space defined by the selected input parameters.

[0106] In some embodiments, the information output by the physics-based model can be represented in a form more convenient for virtual measurement. A basis set is generated whose elements, when combined using weighted addition, can recreate substantially any portion of the property value space covered by the property value map output by the physics-based model. This basis set is generated as a reduced model (e.g., by reduced model generator 174 of FIG. 1). The reduced model is constructed so that a fairly small number of basis set elements can reconstruct most of the physical properties in the chamber, i.e., reconstruct property values ​​at points in the chamber that are substantially equal to those output by the physics-based model. The basis set may be constructed using a proper orthogonal decomposition algorithm. The specific range of what "substantially equal" means may vary based on the application but may be related to experimental error, measurement error, acceptable manufacturing tolerances, etc.

[0107] In some embodiments, the selection of points output by the physics-based model is of interest. For example, by limiting virtual measurements to the boundaries between components of the fabrication chamber 500 (e.g., ignoring property values ​​inside the components), it is possible to make useful predictions of property values ​​with fewer sensors than would be used to virtually measure property values ​​at every point in the chamber. Regions of interest may be locations where the change in property value exceeds a certain threshold in the output from the physics-based model with different process parameter inputs.

[0108] Once data is obtained from the physics-based model, sensors are placed within the chamber. The possible locations where sensors can be placed may be limited by geometry, engineering concerns, etc. Information can be extracted from locations where property values ​​vary with changes in input values ​​to the physics-based model or input parameters to the manufacturing process. Information can be extracted if there is no correlation between property values ​​at the locations where sensors are placed. Sensor locations are chosen taking these and any other constraints into consideration. In some embodiments, machine learning tools may be used to optimize sensor locations.

[0109] Without being bound by theory, it is believed that the number of sensors used to reconstruct the map of characteristic values ​​within a fabrication chamber is TIFF2026004298000002.tif12170, where n s where k is the number of sensors, k is the number of basis set elements used to reconstruct the map of property values, and N is related to the number of locations in the chamber for which property values ​​are predicted. The area of ​​the chamber that is mapped and the number of basis set elements used are adjusted in some implementations to allow compressed sensing using a feasible or convenient number of sensors.

[0110] During the manufacturing process, conditions within the manufacturing chamber 500 are monitored by sensors. Data indicative of measurements made by the sensors are provided as input to a trained machine learning model (e.g., model 190 in FIG. 1 ). The trained machine learning model is trained to reconstruct values ​​of properties in areas of the chamber not measured by the sensors (e.g., at the substrate location). The trained machine learning model utilizes a sparse regression algorithm to determine weights for multiple elements in the basis set generated by the condensed model generator 174. In this manner, conditions near the substrate can be virtually measured without placing a sensor at the substrate location. Physics-based virtual measurement and compressed sensing offer advantages over other techniques. Chamber-to-chamber differences due to part variations, drift, manufacturing tolerance ranges, etc. are accounted for by using sensor data projected into property value space. Additionally, conditions outside the space encompassed by the output of the physics-based model can also be understood using these methods.

[0111] In some embodiments, the compressed sensing data is then used to take corrective action. The corrective action may include sending an alert to a client device. The corrective action may include adjusting a process recipe by making an adjustment to at least one process parameter. The corrective action may be facilitated by feeding the compressed sensing data to another model. The second model may be used to monitor component health, drift, chamber aging, etc. The second model may take corrective action (e.g., recommend maintenance, recommend component replacement, etc.) in light of the compressed sensing data.

[0112] 6 is a block diagram illustrating a computer system 600, according to some embodiments. In some embodiments, computer system 600 may be connected to other computer systems (e.g., via a network such as a local area network (LAN), an intranet, an extranet, or the Internet). Computer system 600 may operate in the capacity of a server or a client computer in a client-server environment, or as a peer computer in a peer-to-peer or distributed network environment. Computer system 600 may be provided by a personal computer (PC), a tablet PC, a set-top box (STB), a personal digital assistant (PDA), a cellular phone, a web appliance, a server, a network router, switch, or bridge, or any device capable of executing a set of instructions (sequential or otherwise) that specify actions the device is to take. Furthermore, the term “computer” is intended to include any collection of computers that, individually or jointly, execute a set (or sets) of instructions to perform one or more of the methodologies described herein.

[0113] In a further aspect, the computer system 600 may include a processing device 602, a volatile memory 604 (e.g., random access memory (RAM)), a non-volatile memory 606 (e.g., read-only memory (ROM) or electrically erasable programmable ROM (EEPROM)), and a data storage device 618, which may communicate with each other via a bus 608.

[0114] The processing device 602 may be provided by one or more processors, such as a general-purpose processor (e.g., a complex instruction set computing (CISC) microprocessor, a reduced instruction set computing (RISC) microprocessor, a very long instruction word (VLIW) microprocessor, a microprocessor implementing another type of instruction set, or a combination of instruction set types), or a special-purpose processor (e.g., an application-specific integrated circuit (ASIC), a field-programmable gate array (FPGA), a digital signal processor (DSP), or a network processor).

[0115] Computer system 600 may further include a network interface device 622 (e.g., coupled to a network 674). Computer system 600 may also include a video display unit 610 (e.g., an LCD), an alphanumeric input device 612 (e.g., a keyboard), a cursor control device 614 (e.g., a mouse), and a signal generating device 620.

[0116] In some embodiments, the data storage device 618 may include a non-transitory computer-readable storage medium 624 (e.g., a non-transitory machine-readable storage medium) for storing instructions 626 encoding any one or more of the methods or functions described herein, including instructions encoding the components of FIG. 1 (e.g., compressive sensing component 114, model 190, etc.), and for performing the methods described herein.

[0117] The instructions 626 may also reside, completely or partially, within the volatile memory 604 and / or within the processing device 602 during execution by the computer system 600; thus, the volatile memory 604 and the processing device 602 may also constitute machine-readable storage media.

[0118] Although computer-readable storage medium 624 is shown as a single medium in the illustrative example, the term "computer-readable storage medium" is intended to include single or multiple media (e.g., centralized or distributed databases and / or associated caches and servers) that store one or more sets of executable instructions. The term "computer-readable storage medium" also includes any tangible medium that can store or encode a set of instructions for execution by a computer, such a set of instructions causing a computer to perform one or more of the methodologies described herein. The term "computer-readable storage medium" is intended to include, but is not limited to, solid-state memory, optical media, and magnetic media.

[0119] The methods, components, and features described herein may be implemented by discrete hardware components or integrated into the functionality of other hardware components, such as an ASIC, FPGA, DSP, or similar device. In addition, the methods, components, and features may be implemented by firmware modules or functional circuitry within a hardware device. Furthermore, the methods, components, and features may be implemented as any combination of hardware devices and computer program components, or as a computer program.

[0120] Unless specifically indicated otherwise, terms such as "receiving," "performing," "providing," "obtaining," "causing," "accessing," "determining," "adding," "using," "training," and the like refer to actions and processes performed or implemented by a computer system that manipulate and transform data represented as physical (electronic) quantities in the computer system's registers and memory into other data similarly represented as physical quantities in the computer system's memory or registers, or in other such information storage, transmission, or display devices. Also, terms such as "first," "second," "third," "fourth," and the like, as used herein, are intended to be labels for distinguishing between different elements and may not have any sequential meaning according to their numerical designations.

[0121] The examples described herein also relate to apparatus for performing the methods described herein. This apparatus may be specially constructed to perform the methods described herein or may comprise a general-purpose computer system selectively programmed by a computer program stored on the computer system. Such a computer program may be stored on a computer-readable tangible storage medium.

[0122] The example methods and diagrams described herein are not inherently related to any particular computer or other apparatus. Various general-purpose systems may be used in accordance with the teachings described herein, or it may prove convenient to construct more specialized apparatus to perform the methods described herein and / or each of their individual functions, routines, subroutines, or operations. Examples of structures for these various systems are set forth above.

[0123] The above description is intended to be illustrative, not limiting. While the present disclosure has been described with reference to certain illustrative examples and embodiments, it will be understood that the present disclosure is not limited to the described examples and embodiments. The scope of the present disclosure should be determined with reference to the following claims, along with the full scope of equivalents to which such claims are entitled.

Claims

1. measuring a subset of characteristic values ​​within the fabrication chamber using a plurality of sensors disposed within the fabrication chamber during a process performed on a substrate within the fabrication chamber; determining a map of property values ​​proximate the substrate within the fabrication chamber based on inputting measurements of the subset of property values ​​within the fabrication chamber into a trained machine learning model; taking an action in consideration of said map of property values; A method comprising:

2. The method of claim 1 , wherein the action comprises taking a corrective action or sending an alert to a client device, the corrective action comprising making an adjustment to at least one processing parameter.

3. 10. The method of claim 1, wherein the trained machine learning model is configured to represent property values ​​within a manufacturing chamber according to a reduced model, and the map of property values ​​is determined as a weighted additive combination of elements of the reduced model.

4. The method of claim 3 , wherein the trained machine learning model comprises a sparse regression model, and the map of feature values ​​is determined using a sparse regression algorithm of the regression model.

5. 4. The method of claim 3, wherein the members selected by a trained machine learning model to determine the map of characteristic values ​​are a subset of the basis set, and the basis set is constructed such that a weighted additive combination of basis sets describes the characteristic values ​​of objects within the manufacturing chamber.

6. The method of claim 5 , wherein the basis set is determined by considering the results of a physics-based computer-aided model that predicts characteristic values ​​throughout a fabrication chamber for a plurality of different process parameter configurations.

7. The method of claim 1 , wherein the map of property values ​​comprises temperatures at multiple locations on the substrate.

8. 8. The method of claim 7, wherein said measuring the subset of characteristic values ​​within the fabrication chamber comprises taking temperature measurements at a plurality of points within the fabrication chamber, at least most of which are not on the substrate.

9. generating training data for a machine learning model, identifying a first training input including data indicative of a first characteristic value associated with a limited subset of the characteristic values ​​in the fabrication chamber; generating training data, the training data including identifying a first target output for the first training input, the first target output including a map of characteristic values ​​at a plurality of locations within the fabrication chamber, the plurality of locations including more locations than the subset of characteristic values ​​used as the first training input; providing the training data to train the machine learning model for (i) a set of training inputs comprising the first training inputs and (ii) a set of target outputs comprising the first target outputs, wherein the trained machine learning model receives new inputs comprising data indicative of second characteristic values ​​associated with a limited subset of characteristic values ​​in the fabrication chamber, the second characteristic values ​​being different from the first characteristic values, and generating new outputs based on the new inputs comprising a map of characteristic values ​​near substrates within the fabrication chamber; A method comprising:

10. The method of claim 9 , wherein each training input in the set of training inputs is mapped to a target output in the set of target outputs.

11. 10. The method of claim 9, wherein the training input comprises data indicative of a plurality of measurements taken of property values ​​within the chamber at locations not adjacent to the substrate, and the target output comprises a prediction of the property value at the location of the substrate.

12. 10. The method of claim 9, wherein the target output comprises data indicative of results of a computer-aided, physics-based model that describes characteristic values ​​within the fabrication chamber for a plurality of process parameters.

13. 13. The method of claim 12, wherein a subset of the results of the computer-aided physics-based model that correspond to locations within the chamber where the change in the value of the property of interest exceeds a threshold when comparing steady states with different manufacturing parameters is used.

14. 10. The method of claim 9, wherein the output of the machine learning model comprises weights applied to basis sets that form additive combinations in a reduced model.

15. The method of claim 9 , wherein the machine learning model comprises a sparse regression model.

16. The method of claim 9 , wherein the subset of characteristic values ​​comprises measurements of conditions at a limited number of locations within the chamber.

17. 17. The method of claim 16, wherein the limited number of locations are selected by inputting the results of a computer-aided, physics-based model into a trained machine learning model, and the output of the trained machine learning model comprises optimized locations for taking measurements.

18. 1. A non-transitory machine-readable storage medium storing instructions that, when executed by a processing device, cause the processing device to perform a process, the process comprising: receiving data indicative of measurements of a subset of the characteristic values ​​within a fabrication chamber containing the substrate; determining a map of property values ​​within the fabrication chamber using a machine learning model, the map including the property values ​​proximate the substrate; (i) adjusting a manufacturing parameter associated with the manufacturing chamber in view of the map of characteristic values; or (ii) sending an alert to a user device; 1. A non-transitory machine-readable storage medium, comprising:

19. 20. The non-transitory machine-readable storage medium of claim 18, wherein the processing further comprises producing an output describing a map of characteristic values ​​within the manufacturing chamber as coefficients for a weighted additive combination of multiple elements of a basis set, the basis set being constructed taking into account results of a computer-aided, physics-based model of the chamber evaluated at multiple process parameters.

20. 20. The non-transitory machine-readable storage medium of claim 18, wherein the measurements of a subset of characteristic values ​​within the fabrication chamber include temperature measurements taken at a plurality of points within the fabrication chamber, at least most of which are not on the substrate.

21. 20. The non-transitory machine-readable storage medium of claim 18, wherein the map of property values ​​comprises temperatures at a plurality of points in the fabrication chamber, including at least a point on the substrate.

Images