Chamber condition monitoring and simulation after preventive maintenance
The machine learning model predicts the chamber status, which solves the problem of inability to recover immediately after the chamber is cleaned, and realizes chamber health checks and abnormal detection, reduces equipment downtime and improves the production efficiency of semiconductor manufacturing.
Patent Information
- Application Number
- CN202280082123.7
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Priority Date
- 2021-12-14
- Filing Date
- 2022-12-07
- Publication Date
- 2024-11-08
AI Technical Summary
Prior art During semiconductor manufacturing, the chamber cannot be restored to the appropriate production state immediately after cleaning, resulting in production delays and equipment downtime, and traditional methods rely on test sample evaluation, are inefficient and do not provide in-situ chamber condition monitoring.
The machine learning model is used to predict the chamber status and conditions based on sensor data, and to predict whether the processing chamber meets threshold conditions through training data, chamber health checks and abnormal detection are realized, reducing downtime and improving productivity.
It realizes estimating chamber conditions without relying on test samples, providing quantitative health information, reducing chamber downtime, improving productivity of manufacturing equipment, and early detection of abnormal conditions through in-situ monitoring.
Smart Images

Figure CN118922790A_ABST
Abstract
Description
Technical Field
[0001] The present specification generally relates to chamber condition monitoring and simulation. More specifically, the present specification relates to monitoring and simulation of chamber conditions associated with chamber seasoning. Background Art
[0002] Substrate processing may include a series of processes to produce circuits in a semiconductor (e.g., a silicon wafer) according to a circuit design. These processes may be performed in a series of chambers. Successful operation of a modern semiconductor manufacturing facility may be directed toward facilitating a steady flow of wafers moving from one chamber to another during the process of forming circuits in the wafers. In the process of performing many substrate processes, the conditions of the processing chambers may change and may result in the processed substrates not meeting the desired conditions and results.
[0003] One such substrate processing may include plasma etching, which is a process of transferring a pattern in a mask material layer to another layer (such as a conductive or dielectric material layer) under the mask by removing layered materials from the wafer surface. Such processing inevitably produces different types of etching byproducts (such as silicon oxide and organic polymers), depending on the layered materials and the etching chemistry. Some byproducts are deposited on the inner surface of the chamber in which the plasma etching process is performed. The deposition of byproducts may affect the etching performance, such as by depositing particles (e.g., flakes) onto the substrate or by reacting with the plasma and affecting the processing results.
[0004] To mitigate the effects of etching byproducts, preventive maintenance (such as chamber cleaning) can be used to periodically remove deposits from the chamber walls. To do this, the chamber is shut down and a cleaning plasma (such as a CF4+O2 plasma used to clean silicon oxide deposited during silicon etching) is introduced into the chamber. This plasma reacts with the deposits, and the products of this reaction are pumped out of the chamber. However, after such chamber cleaning, it has been observed that the cleaned chamber walls make the chamber unsuitable for immediate production wafer etching. Chamber aging is the process of etching a series of substrates (e.g., blank silicon wafers) to restore chamber conditions suitable for production substrate processing. After chamber aging, a thin layer of silicon oxide covers the chamber walls. The chamber is then returned to production wafer etching until the next round of chamber cleaning and aging is required. Summary of the invention
[0005] A method and system for monitoring and simulation after preventive maintenance is performed. In some embodiments, a method performed by a processing device may include the following steps: receiving sensor data indicating an environmental state of a processing chamber, the processing chamber processing a series of substrates according to a chamber aging program. The series of substrates are processed after the preventive maintenance program is performed on the processing chamber. The processing device may determine a first set of values based on the sensor data. The first set of values may each be associated with a respective substrate in the series of substrates. The first set of values may indicate a likelihood that the associated substrate includes a first processing result that satisfies a threshold condition of the substrate processing program. The processing device may predict a first test result based on the first set of values. The first test result may indicate a likelihood that a first substrate processed after processing the series of substrates includes a second processing result that satisfies the threshold condition. The processing device may prepare the first test result for presentation on a graphical user interface (GUI). The processing device may change the operation of the processing chamber based on the first test result.
[0006] In some embodiments, a method for training a machine learning model to predict a condition of a processing chamber after a preventive maintenance procedure. The method includes the steps of generating training data for the machine learning model. The step of generating training data may include the steps of identifying a first training input having first sensor data. The first sensor data indicates a first state of the first processing chamber after performing the preventive maintenance procedure. The step of generating training data may further include the steps of identifying a first target output for the first training input. The first target output may indicate whether a substrate processed by the first processing chamber in the first state according to the first substrate processing procedure results in the first substrate having a first processing result that satisfies a first threshold condition associated with the first substrate processing procedure. The method may further include the steps of providing training data to train the machine learning model on (i) a set of training inputs including the first training input and (ii) a set of target outputs including the first target output. The trained machine learning model may receive a new input having new sensor data indicating a new state of the new processing chamber. The new output may indicate whether a new substrate processed by the new processing chamber in the new state according to the first substrate processing procedure results in the new substrate having a new processing result that satisfies the threshold condition.
[0007] In some embodiments, the above method may be stored on a non-transitory machine-readable storage medium including instructions that, when executed by a processing device, cause the processing device to perform operations that may include one or more steps described in the above method. BRIEF DESCRIPTION OF THE DRAWINGS
[0008] Aspects and embodiments of the present disclosure will become more fully understood from the detailed description and accompanying drawings given below, which are intended to illustrate the aspects and embodiments by way of example rather than limitation.
[0009] Figure 1 is a block diagram illustrating an exemplary system architecture in which embodiments of the present disclosure may operate.
[0010] FIG. 2A to FIG. 2B is a block diagram illustrating a chamber recovery process 200A-200B in which embodiments of the present disclosure may be performed.
[0011] Figure 3 is a block diagram illustrating a chamber monitoring and maintenance system 200 in which embodiments of the present disclosure may operate.
[0012] Figure 4 Depicts a graph of chamber recovery curves according to some embodiments of the present disclosure.
[0013] Figure 5 A flow chart depicting an exemplary method for monitoring and predicting conditions of a processing chamber according to some embodiments of the present disclosure.
[0014] Figure 6 is an exemplary illustration of the training phase of a machine learning system according to some embodiments of the present disclosure.
[0015] Figure 7 Illustrated are model training workflows and model application workflows for chamber monitoring and simulation according to aspects of the present disclosure.
[0016] Figure 8 A block diagram depicts an exemplary computing device operating in accordance with one or more aspects of the present disclosure. DETAILED DESCRIPTION
[0017] Substrate processing may include a series of processes to produce circuits in a semiconductor (e.g., a silicon wafer) according to a circuit design. These processes may be performed in a series of chambers. The successful operation of a modern semiconductor manufacturing facility may be aimed at facilitating a steady flow of wafers moving from one chamber to another during the process of forming circuits in the wafers. During the process of performing many substrate procedures, the conditions of the processing chambers may deteriorate and cause the processed substrates to fail to meet the desired conditions or processing results (e.g., critical dimensions, processing uniformity, thickness dimensions, etc.).
[0018] Preventive maintenance procedures (e.g., chamber cleaning) are often used as part of a chamber recovery process to return the state of the processing chamber to a state suitable for entering a production mode of substrate processing (e.g., batch processing of substrates). A recovery procedure is often used after a preventive maintenance procedure to prepare the chamber for production mode (e.g., "preheat" the chamber). A common recovery procedure that is often employed is to condition the processing chamber. Chamber conditioning is a procedure that includes processing a series of substrates (e.g., blank silicon wafers) to restore chamber conditions (e.g., coating chamber walls) suitable for production substrate processing (e.g., substrates processed in the chamber have processing results that meet a desired threshold standard). After chamber conditioning, the chamber may be operated in production mode for a period of time until another round of preventive maintenance and further chamber conditioning is required, or it is advisable to otherwise restore the state of the processing chamber.
[0019] As previously described, the chamber conditioning process includes processing a series of substrates to prepare the chamber for performing a substrate processing process. The substrate processing process may include setting conditions (e.g., temperature, pressure, vacuum conditions, plasma conditions, etc.) of the chamber to obtain substrates with target processing results (e.g., the processing results meet threshold criteria associated with the substrate processing recipe).
[0020] Traditionally, after a preventive maintenance procedure, a series of substrates are processed in a processing chamber. After processing this series of substrates, a test sample is processed and evaluated. The test sample may include a substrate that is processed in a processing chamber and evaluated (e.g., measuring the processing results) to determine whether the test sample contains a processing result that satisfies a threshold condition. If the condition (s) is not met, a further aged substrate may be processed and another test sample may be processed to determine whether the chamber is fully recovered. This process may be repeated until the chamber condition is met or a threshold amount of aged wafers have been used. For the conventional method of determining the chamber condition as described above, when a first number of aged wafers are used but a smaller number is actually required to restore the chamber to an operable state for substrate production, an unnecessary number of aged wafers (e.g., an "over-aged" chamber) may be required. In addition, the conventional method may require delaying the return of the chamber to production mode by waiting for the test sample measurement and evaluation to determine whether the chamber has been fully recovered. Because the tool or chamber is down (e.g., not running in production mode), the time spent in the recovery process is a loss of productivity. Additionally, conventional methods do not provide in situ chamber condition monitoring, but instead rely on testing samples and awaiting evaluation of the test samples during an aging process.
[0021] Aspects and embodiments of the present disclosure address these and other shortcomings of the prior art by providing, in various embodiments, methods and systems that are capable of estimating chamber conditions without relying on inspection results of test samples, providing quantitative information about chamber health and recovery after preventive maintenance procedures, and simulating chamber conditions with manual parameter adjustments. In some embodiments, a machine learning model may be utilized to determine chamber conditions and / or states during and / or after an aging procedure. In some embodiments, the present disclosure enables health checks and prevention during a chamber aging sequence, and early detection of abnormal chamber conditions prior to sample testing. One or more of these aspects and / or embodiments may reduce downtime of a processing tool (e.g., a processing chamber) and may enable higher productivity of manufacturing equipment.
[0022] In an exemplary embodiment, a method and system for monitoring and simulation after preventive maintenance are performed. In some embodiments, a method performed by a processing device may include the following steps: receiving sensor data indicating an environmental state of a processing chamber, the processing chamber processing a series of substrates according to a chamber aging program. The series of substrates are processed after a preventive maintenance program is performed on the processing chamber. The processing device may determine a first set of values based on the sensor data. Each of the first set of values may be associated with a respective substrate in the series of substrates. The first set of values may indicate the likelihood that the associated substrate includes a first processing result that satisfies a threshold condition of the substrate processing program. The processing device may predict a first test result based on the first set of values. The first test result may indicate the likelihood that a first substrate processed after processing the series of substrates includes a second processing result that satisfies the threshold condition. The processing device may prepare the first test result for presentation on a graphical user interface (GUI). The processing device may change the operation of the processing chamber based on the first test result.
[0023] In an exemplary embodiment, a method for training a machine learning model to predict a condition of a processing chamber after a preventive maintenance procedure. The method includes the steps of generating training data for the machine learning model. The step of generating training data may include the steps of identifying a first training input having first sensor data. The first sensor data may indicate a first state of the first processing chamber after performing the preventive maintenance procedure. The step of generating training data may further include the steps of identifying a first target output for the first training input. The first target output may indicate whether a substrate processed by the first processing chamber in the first state according to the first substrate processing procedure results in the first substrate having a first processing result that satisfies a first threshold condition associated with the first substrate processing procedure. The method may further include the steps of providing training data to train the machine learning model on (i) a set of training inputs including the first training input and (ii) a set of target outputs including the first target output. The trained machine learning model may receive a new input having new sensor data indicating a new state of a new processing chamber. The new output may indicate whether a new substrate processed by the new processing chamber in the new state according to the first substrate processing procedure results in the new substrate having a new processing result that satisfies the threshold condition.
[0024] In some embodiments, the above method may be stored on a non-transitory machine-readable storage medium including instructions that, when executed by a processing device, cause the processing device to perform operations that may include one or more steps described in the above method.
[0025] Figure 1 1 is a block diagram illustrating an exemplary system architecture 100 in which embodiments of the present disclosure may operate. Figure 1 As shown, the system architecture 100 includes a manufacturing system 102, a metrology system 110, a client device 150, a data storage 140, a server 120, and a machine learning system 170. The machine learning system 170 may be a part of the server 120. In some embodiments, one or more components of the machine learning system 170 may be fully or partially integrated into the client device 150. The manufacturing system 102, the metrology system 110, the client device 150, the data storage 140, the server 120, and the machine learning system 170 may each be hosted by one or more computing devices, including a server computer, a desktop computer, a laptop computer, a tablet computer, a notebook computer, a personal digital assistant (PDA), a mobile communication device, a cell phone, a handheld computer, or a similar computing device.
[0026] The manufacturing system 102, the metering system 110, the client device 150, the data storage 140, the server 120, and the machine learning system 170 can be coupled to each other via a network 160 (e.g., for performing the methods described herein). In some embodiments, the network 160 is a private network that provides each element of the system architecture 100 with access to each other and other privately available computing devices. The network 160 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, and / or any combination thereof. Alternatively or additionally, any elements of the system architecture 100 may be integrated together or otherwise coupled without the use of the network 160.
[0027] The client device 150 may be or include any personal computer (PC), laptop computer, cell phone, tablet computer, small notebook computer, network television ("smart TV"), networked media player (e.g., Blu-ray player), set-top box, over-the-top (OOT) streaming device and operation box, etc. The client device 150 may include a browser 152, an application 154, and / or other tools as described and executed by other systems of the system architecture 100. In some embodiments, the client device 150 may be processed at various stages of the system architecture 100 as described herein to enable access to the manufacturing system 102, the metrology system 110, the data storage 140, the server 120, and / or the machine learning system 170 and communicate (e.g., transmit and / or receive) sensor data, process data, data classification, chamber recovery data (e.g., recovery curves), process result data, and / or various processing tools (e.g., metrology tool 114, data preparation tool 116, classifier tool 124, chamber recovery tool 126, and / or chamber recovery components) The indication of input and output.
[0028] like Figure 1 As shown, the manufacturing system 102 includes a processing tool 104, a processing program 106, and a processing controller 108. The processing controller 108 can coordinate the operation of the processing tool 104 to execute one or more processing programs 106. For example, various processing tools can include specialized chambers such as etching chambers, deposition chambers (including chambers for atomic layer deposition, chemical vapor deposition, or plasma-enhanced versions thereof), annealing chambers, and / or the like. In another example, the machine can be combined with a sample transport system (e.g., a Selective Compliance Assembly Robot Arm (SCARA) robot, a transfer chamber, a front opening unit (FOUP), a side storage unit (SSP), and / or the like) to transport samples between machines and processing steps.
[0029] The process program 106 (or sometimes referred to as a process recipe or process steps) may include various specifications for performing operations by the process tool 104. For example, the process program 106 may include process specifications such as the start-up duration of the process operation, the process tool used for the operation, the temperature, flow rate and pressure of the machine (e.g., chamber), etc., and the deposition sequence, etc. In another example, the process program may include transfer instructions for transporting the sample to further processing steps or measuring the sample by the metrology system 110. One or more process programs 106 include preventive maintenance programs and aging programs.
[0030] The process controller 108 may include equipment designed to manage and coordinate the actions of the process tool 104. In some embodiments, the process controller 108 is associated with a process recipe or a series of process program 106 instructions that, when applied in a designed manner, produce a desired process result for substrate processing. For example, a process recipe may be associated with processing a substrate to produce a target process result (e.g., critical dimension, thickness and uniformity standards, etc.). The process controller 108 may coordinate the execution of preventive maintenance procedures (e.g., chamber cleaning) and chamber recovery procedures (e.g., chamber conditioning) to return a chamber to a production state.
[0031] like Figure 1 As shown, the metrology system 110 includes a metrology tool 114 and a data preparation tool 116. The metrology tool 114 may include various sensors to measure process results (e.g., critical dimensions, thickness, uniformity, etc.) within the manufacturing system 102. For example, wafers processed in one or more processing chambers can be used to measure critical dimensions. The metrology tool 114 may also include devices for measuring process results of substrates processed using the manufacturing system. For example, process results such as critical dimensions, thickness measurements (e.g., film layers from etching, deposition, etc.) of substrates processed according to a process recipe and / or actions performed by the process controller 108 can be evaluated. These measurements can also be used to measure the condition of the chamber during a recovery process, such as during a burn-in process that includes processing a series of wafers (e.g., blank wafers).
[0032] The data preparation tool 116 may include a process for extracting features and / or generating synthetic / engineered data associated with the data measured by the metrology tool 114. In some embodiments, the data preparation tool 116 may identify correlations, patterns, and / or anomalies in metrology or process performance data. For example, the data preparation tool 116 may perform feature extraction, wherein the data preparation tool 116 uses a combination of measurement data to determine whether a criterion is met. For example, the data preparation tool 116 may analyze multiple data points of relevant parameters (e.g., thickness, critical dimension, defectivity, plasma conditions, etc.) to determine whether rapid changes occur during iterations of a aging process within a processing chamber. In some embodiments, the data preparation tool 116 performs normalization on various sensor data associated with various processing chamber conditions. Normalization may include processing incoming sensor data to make it look similar between various chambers and various sensors used to acquire the data.
[0033] In some embodiments, the data preparation tool 116 can perform one or more of a process control analysis, a univariate limit violation analysis, or a multivariate limit violation analysis on the metrology data (e.g., obtained by the metrology tool 114). For example, the data preparation tool 116 can perform statistical process control (SPC) by employing a statistically based approach to monitor and control the process controller 108. For example, SPC can improve the efficiency and accuracy of preventive maintenance and / or wafer aging processes (e.g., by identifying data points that fall within and / or exceed control limits).
[0034] In some embodiments, the process chamber may be measured throughout the burn-in process. In some embodiments, an incremental amount of metrology data is acquired during a predetermined burn-in process. For example, additional sensors may be activated during or immediately after processing a wafer and / or currently activated sensors may acquire additional data. In some embodiments, the process controller 108 may trigger measurements by the metrology tool 114 based on operations to be performed by the process tool 104. For example, in response to one or more test wafers burning a wafer being processed during a transition period (during which the process chamber completes its burn-in process and is awaiting an incoming production wafer to be processed if the process result meets the criteria within the process program 106), the process controller 108 may trigger the activation of one or more process results (e.g., of the metrology tool 114).
[0035] In some embodiments, the extracted features, generated synthetic / engineered data, and statistical analysis can be used in association with the machine learning system 170 (e.g., to train, validate, and / or test the machine learning model 190). Additionally and / or alternatively, the data preparation tool 116 can output the data to the server 120 for use by any of the classifier tool 124 and / or the chamber recovery tool 126.
[0036] The data storage 140 can be a memory (e.g., random access memory), a drive (e.g., a hard disk and a flash drive), a database system, or another type of component or device capable of storing data. The data storage 140 can store one or more historical data 142, which includes old sensor data 144 of previous chamber conditions and / or old process result data 146 and process results of substrates processed under the associated chamber conditions. In some embodiments, the historical data 142 can be used to train, validate, and / or test the machine learning model 190 of the machine learning system 170 (see, e.g., for exemplary methods). Figure 6 ).
[0037] The server 120 may include one or more computing devices such as a rack server, a router computer, a server computer, a personal computer, a mainframe computer, a laptop computer, a tablet computer, a desktop computer, etc. The server 120 may include a classifier tool 124 , a chamber recovery tool 126 , and a data simulation tool 128 .
[0038] The classifier tool 124 receives metrology data (e.g., chamber sensor data) from the processing tool 104 and determines the state of the chamber throughout the chamber aging process. In some embodiments, the classifier tool 124 receives raw sensor data from the processing tool 104, and in other embodiments, the raw sensor data is combined with synthetic data processed from the data preparation tool 116. The classifier tool 124 can process the sensor data to determine a set of values that indicate the likelihood that a substrate processed under the relevant chamber conditions will result in a substrate having a processing result that meets the threshold criteria corresponding to the processing program 106. The classifier tool 124 can classify the state or condition of the processing chamber based on the received sensor data and the machine learning model trained in the machine learning system 170. For example, the state of the processing chamber during the aging process can be processed to determine whether the chamber meets the conditions for transitioning from a chamber recovery mode to a chamber production mode. In some embodiments, the classifier tool 124 includes a machine learning model that uses process data (e.g., sensor data (e.g., by the processing tool 104), synthetic and / or engineered data (e.g., from the processing tool 104), and general process parameter values corresponding to the processing program 106) and determines a set of values that indicate the likelihood that the chamber will meet a threshold condition during the aging process. As will be discussed later, the machine learning model may include a bootstrap aggregation model, a random forest decision tree model, or a partial least squares regression (PLS) model, among other models. The machine learning model may include ensemble modeling that includes multiple models and utilizes a higher confidence model to ultimately classify the received data.
[0039] The chamber recovery tool 126 may receive data from the classifier tool 124 and generate a chamber recovery curve associated with the recovery of a processing chamber associated with a preventive maintenance and / or aging program. The chamber recovery tool 126 may determine a preventive maintenance indicator that indicates a change in the likelihood that the chamber meets the conditions to move from the recovery mode of operation to the production mode of operation. For example, the chamber recovery tool may identify when the conditions of the chamber meet a threshold condition such that a substrate processed under the chamber conditions results in a substrate having a processing result that meets the threshold condition associated with one or more processing programs 106. In some embodiments, the chamber recovery tool 126 may use a machine learning model that receives input data from the classifier tool 124 and outputs a preventive maintenance indicator value associated with the likelihood that the chamber meets the threshold condition to transition to the production mode of operation.
[0040] The data simulation tool 128 coordinates with the classifier tool 124 and the chamber recovery tool 126 to identify changes in chamber status (e.g., ready to move to production mode, needing more aging, needing further preventive maintenance, etc.) based on manually adjusted components of sensor data acquired using the manufacturing system 102. The data simulation tool 128 may identify one or more sensor data parameters or parameter combinations that are more influential in the chamber recovery process. The data simulation tool may utilize the classifier tool 124 and the chamber recovery tool 126 to identify the state of the chamber under simulated conditions. The data simulation tool 128 may identify one or more process priorities associated with the chamber recovery procedure. For example, one or more simulation models may determine one or more weights associated with various process parameters (e.g., temperature, pressure, and vacuum conditions, etc.).
[0041] As previously described, some embodiments of the classifier tool 124, chamber recovery tool 126, and / or data simulation tool 128 may use machine learning models to perform the methods they describe. The associated machine learning model may be generated (e.g., trained, validated, and / or tested) using the machine learning system 170. An exemplary description of the machine learning system 170 will be described as follows in the context of using the machine learning system 170 to generate a machine learning model 190 associated with the classifier tool 124. However, it should be noted that this description is purely exemplary. As will be further discussed in conjunction with other embodiments, similar processing hierarchies and methods may be used to generate and execute machine learning models associated with the classifier tool 124, chamber recovery tool 126, and / or data simulation tool 128, alone and / or in combination with each other.
[0042] The machine learning system 170 may include one or more computing devices, such as a rack server, a router computer, a server computer, a personal computer, a mainframe computer, a laptop computer, a tablet computer, a desktop computer, and the like. The machine learning system 170 may include a chamber state component 194. In some embodiments, the chamber state component 194 may use the historical data 142 to determine whether the state of the processing chamber meets the conditions for transitioning to the production operation mode. In some embodiments, the chamber state component 194 may use the trained machine learning model 190 to determine the state of the chamber. The trained machine learning model 190 may use the historical data to determine the chamber state.
[0043] In some embodiments, the trained machine learning models 190 are stored on the server machine 192 and / or the server 120. The server machine 180 and / or the server machine 192 may store one or more recently trained machine learning models (e.g., to perform model performance management). For example, one or more models trained over a period of time (e.g., the past 6 months) may be stored in the server machine 192, and one or more final models for predicting chamber conditions may be stored on the server 120 (e.g., operable with the chamber recovery tool 126).
[0044] In some embodiments, the machine learning system 170 further includes a server machine 172 and a server machine 180. The server machines 172 and 180 can be one or more computing devices (e.g., rack servers, router computers, server computers, personal computers, mainframe computers, laptop computers, tablet computers, desktop computers, etc.), data storage (e.g., hard disk, memory database), network, software components, or hardware components.
[0045] The server machine 172 may include a data set generator 174 that can generate a data set (e.g., a set of data inputs and a set of target outputs) to train, validate, or test a machine learning model. The data set generator 174 may divide the historical data 142 into a training set (e.g., sixty percent of the historical data, or any other portion of the historical data), a validation set (e.g., twenty percent of the historical data, or some other portion of the historical data), and a test set (e.g., twenty percent of the historical data). In some embodiments, the chamber state component 194 generates multiple sets of training data. For example, one or more sets of training data may include each of the data sets (e.g., a training set, a validation set, and a test set).
[0046] The server machine 180 includes a training engine 182, a validation engine 184, and a testing engine 186. The training engine 182 may be capable of using one or more of the old sensor data 144 and the old process result data 146 of the historical data 142 (of the data storage 140) to train the machine learning model 190. In some embodiments, the machine learning model 190 may be trained using one or more outputs of the data preparation tool 116, the classifier tool 124, the chamber recovery tool 126, and / or the data simulation tool 128. For example, the machine learning 190 may be a hybrid machine learning model that uses sensor data and / or mechanical features (such as feature extraction, mechanical modeling, and / or statistical modeling (e.g., using the data preparation tool 116)). The training engine 182 may generate a plurality of trained machine learning models 190, wherein each trained machine learning model 190 corresponds to a different set of features for each training set.
[0047] The validation engine 184 may determine the accuracy of each trained machine learning model 190 based on the corresponding feature set of each training set. The validation engine 184 may discard trained machine learning models 190 with an accuracy that does not meet a threshold accuracy. The testing engine 186 may determine the trained machine learning model 190 with the highest accuracy of all trained machine learning models based on the test set (and optionally, the validation set).
[0048] In some embodiments, training data is provided to train the machine learning model 190, such that the trained machine learning model may receive a new input having new sensor data indicating a new state of a new processing chamber. A new output may indicate whether a new substrate processed by the new processing chamber in the new state according to the first substrate processing program results in the new substrate having a new processing result that satisfies a threshold condition.
[0049] The machine learning model 190 may refer to a model created by the training engine 182 using a training set that includes data inputs and corresponding target outputs (historical results of the process chamber under parameters associated with the target inputs). Patterns in the data set that map the data inputs to the target outputs (e.g., identifying connections between portions of sensor data and resulting chamber states) may be discovered, and mappings that capture these patterns are provided to the machine learning model 190. The machine learning model 190 may use one or more of logistic regression, syntactic parsing, decision trees, or support vector machines (SVMs). The machine learning may consist of a single-stage linear or nonlinear operation (e.g., an SVM) and / or may be a neural network.
[0050] The chamber state component 194 may provide current data (e.g., current sensor data associated with the state of the processing chamber during the burn-in procedure) as input to the trained machine learning model 190, and the chamber state component 194 may run the trained machine learning model 190 on the input to obtain one or more outputs, the one or more outputs including a set of values indicating the likelihood of satisfying a threshold condition at various stages of the burn-in process (e.g., to determine an optimal number of substrates to be processed during the burn-in process to prepare the chamber for production (e.g., without wasting unneeded burn-in wafers). The chamber state component 194 may be capable of identifying confidence data from the output indicating a confidence level that the state of the processing chamber at various stages of the burn-in process satisfies the threshold condition to move to production mode. In one non-limiting example, the confidence level is a real number between 0 and 1 (inclusive), where 0 indicates no confidence in one or more chamber states and 1 indicates absolute confidence in the chamber state.
[0051] For purposes of illustration and not limitation, aspects of the present disclosure describe the training of a machine learning model and the use of a trained learning model using information related to historical data 142. In other embodiments, a heuristic model or a rule-based model is used to determine chamber status.
[0052] In some embodiments, the functionality of client device 150, server 120, data storage 140, and machine learning system 170 may be provided by a comparison Figure 1 For example, in some embodiments, server machines 172 and 180 may be integrated into a single machine, while in some other embodiments, server machines 172, 180, and 192 may be integrated into a single machine. In some embodiments, machine learning system 170 may be provided in whole or in part by server 120.
[0053] In general, the functions described in one embodiment as being performed by client device 150, data storage 140, metrology system 110, manufacturing system 102, and machine learning system 170 may also be performed on server 120 in other embodiments, if appropriate. In addition, the functions attributed to a particular component may be performed by multiple components that are different or operate together.
[0054] In embodiments, a "user" may be represented as a single individual. However, other embodiments of the present disclosure encompass a "user" that is an entity controlled by multiple users and / or automation sources. For example, a group of individual users united as a group of administrators may be considered a "user."
[0055] FIG. 2A to FIG. 2B2 is a block diagram illustrating a chamber recovery process 200A-200B in which embodiments of the present disclosure may be run. The chamber recovery process 200A-200B may include performing preventive maintenance 202 on one or more processing tools. The preventive maintenance 202 may include any number of processes performed on one or more processing tools (e.g., processing chambers) to maintain the condition of the processing chambers over time. In some embodiments, the preventive maintenance 202 may employ a proactive preventive maintenance program to achieve one or more of the following goals: minimize downtime of processing equipment, keep maintenance costs low (e.g., maintain smaller repairs rather than prevent larger, less frequent repairs), extend the useful life of processing equipment, improve production quality (e.g., processing results of substrates), improve processing efficiency (e.g., keep tools in production mode longer and reduce downtime of processing equipment), and / or maintain the safety and integrity of equipment, etc.
[0056] In some embodiments, the preventive maintenance 202 includes cleaning (e.g., wet cleaning maintenance) one or more process chambers (e.g., removing buildup on chamber walls resulting from a process performed within the process chamber). The preventive maintenance 202 may restore the condition of one or more process tools to an unseasoned state (e.g., a clean chamber that has not undergone a seasoning process). The replacement and / or cleaning of one or more process tools may result in the process tools operating in a clean (e.g., pristine) state that is further fine-tuned by using the process tools to perform a substrate processing process. For example, after performing the preventive maintenance 202, the process chamber may be returned to an unseasoned state, such as a new state or a clean state.
[0057] As shown in FIG. 2 , the chamber recovery process 200A-200B includes chamber initialization. The chamber initialization 204 may include preheating the chamber or starting one or more systems associated with the processing chamber. For example, the chamber initialization 204 may be performed to prepare the processing chamber to receive and process the substrate. The chamber initialization may be combined with one or more processing routines (e.g., Figure 1 associated with the processing program 106).
[0058] As shown in FIG. 2 , chamber recovery processes 200A-200B include conditioning 206 of the processing chamber. Chamber conditioning 206 is a process of treating a series of substrates (e.g., blank silicon wafers) to restore chamber conditions suitable for production substrate processing. For example, in some embodiments, after chamber conditioning, a thin layer of silicon oxide covers the chamber walls. At conditioning 206, an initial number of aged wafers (e.g., 20 or more aged wafers) are treated. Chamber recovery process 200A utilizes test samples 208 to determine whether chamber conditions are satisfied to transition to production mode, while chamber recovery process 200B uses a preventive maintenance (PM) model 212 to determine whether chamber conditions are satisfied to transition to production mode.
[0059] In some embodiments, Figure 2A As shown, after conditioning the chamber with the first number of wafers, a sample test 208 is performed and evaluated. Testing the sample may include processing the substrates through the processing chamber under the post-conditioning processing conditions of the first number of substrates. Measurements are taken of the sample and various process result criteria are evaluated based on the sample test. For example, the sample test 208 may be evaluated to determine whether the substrates include process results that meet threshold criteria (such as critical dimensions, process uniformity thresholds, thickness distribution requirements, etc.). Processing and evaluating the sample test may involve the duration of downtime required to process the substrates and evaluate the processed substrates (e.g., measuring process results such as thickness and uniformity conditions across the substrate surface).
[0060] At block 210A, the chamber recovery process 200A may determine whether one or more threshold conditions of the processing chamber are satisfied based on the sample test 208. The threshold conditions may include criteria associated with one or more processing results of the sample test 208. As previously described, the sample test 208 may be evaluated to determine whether the substrate includes processing results that satisfy threshold criteria (e.g., critical dimensions, process uniformity thresholds, and thickness distribution requirements, etc.), any of which may be evaluation criteria for determining whether the chamber satisfies the conditions to transition to a production mode of operation. The processing logic may determine that the state of the chamber does not satisfy the threshold condition(s), and the processing logic may further perform additional seasoning 206 and another sample test 208 to determine whether the chamber is properly seasoned (e.g., satisfies the threshold conditions). The process of seasoning 206, performing the sample test 208, and evaluating whether the threshold conditions of the processing chamber (e.g., block 210) are satisfied based on the previous sample test may be repeated until the chamber satisfies the threshold conditions or until a threshold number of seasoned wafers or seasoning iterations have been performed. Upon determining that the threshold condition is met, the chamber recovery process 200A may continue by initializing a production mode of the processing chamber.
[0061] The chamber recovery process 200B includes performing preventive maintenance (PM), chamber initialization, and initial aging 206 as previously described. Figure 2B As shown, the chamber recovery process 200B continues by monitoring the chamber health and evaluating the state of the processing chamber using the PM model 212. The PM model 212 can receive sensor data indicating the state of the processing chamber during the aging 206. In some embodiments, the PM model includes one or more aspects and / or features of the classifier tool 124, the chamber recovery tool 126, and / or the data simulation tool 128. The PM model 212 can output a notification indicating the state of the processing chamber. In some embodiments, the notification can include an indication that the chamber has been recovered. For example, the PM model 212 can indicate authorization to perform the sample test 208 (e.g., to verify the results of the PM model 212).
[0062] At block 210, the chamber recovery process 200B receives output from the PM model 212 and determines further actions to be taken in response to the received output. In some embodiments, the notification may include an indication that the chamber has been recovered. For example, the PM model 212 may indicate authorization to perform a sample test 208 (e.g., to verify the results of the PM model 212). The PM model 212 may output a notification that the processing chamber has not been recovered and further conditioning of the chamber is required before entering production mode. The PM model 212 may output a notification indicating that the quality of the preventive maintenance does not meet the threshold condition and further preventive maintenance is required. In some embodiments, the chamber recovery process 200B may determine that the conditions of the processing chamber are met and the production mode can be initialized without using sample testing. In some embodiments, the chamber recovery process 200B may determine that the previously performed preventive maintenance is insufficient and the processing chamber may require further preventive maintenance to meet the threshold condition without performing multiple aging 206 iterations. For example, rather than directing processing logic to seasoning 206 when a threshold condition is not met, processing may be directed back to preventative maintenance 202 rather than performing seasoning that would fail to bring the chamber condition to meet the threshold condition (eg, wasting seasoned wafers).
[0063] In some embodiments, the chamber recovery process 200B uses less time (e.g., less downtime of the processing chamber) to return the chamber to production mode. For example, the chamber recovery process 200B may not require performing sample testing (e.g., processing and evaluation of test wafers) to determine that the processing chamber meets one or more threshold conditions. In another example, the chamber recovery process 200B may determine that further preventive maintenance may be preferable to further restore the state of the chamber to meet one or more threshold conditions, rather than performing more conditioning on the processing chamber.
[0064] As will be discussed further in other embodiments, the chamber recovery process 200A can be utilized to generate training data for the PM model 212. Sensor data of the processing chamber can be tracked during the aging process 206, and data associated with the sample tests 208 can be used to validate outputs from the PM model 212. For example, the PM model 212 can determine the state of the processing chamber. Evaluations of sample tests processed under the same chamber conditions can be compared to the predictions of the PM model 212 to validate and further refine the PM model 212.
[0065] Figure 3 is a block diagram illustrating a chamber monitoring and maintenance system 300 in which embodiments of the present disclosure may operate. The chamber monitoring and maintenance system 300 includes a substrate processing system 302 (e.g., Figure 1 The substrate processing system 302 may include a processing tool 304 and chamber data 306 (e.g., acquired using the processing tool 304). The processing tool 304 may include chamber sensors (e.g., for measuring radio frequency (RF), temperature, and gas, etc.) and additional sensors (e.g., optical emission spectroscopy (OES) sensors). The chamber data 306 may include data indicating the state of the processing chamber. For example, the processing tool 304 may measure and monitor parameters such as RF, temperature, gas, and spectral data, etc.
[0066] In some embodiments, the chamber data 306 is processed by a data preparation tool 308. The data preparation tool 308 may receive chamber data (e.g., raw sensor data) and generate synthetic data associated with various combinations, correlations, and / or artificial parameters. The data preparation tool 308 may reduce the raw sensor data into groups or features. For example, the data preparation tool 308 may generate features including rates of change of parameter values and / or sets of parameter values. In some embodiments, the data preparation tool 308 performs any of partial least squares analysis, principal component analysis, multi-factor dimensionality reduction, nonlinear dimensionality reduction, and / or any combination thereof. In some embodiments, the data preparation tool 308 is designed for edge detection of chamber data. For example, the data preparation tool 308 includes methods designed to identify data points that change sharply and / or have discontinuities. For example, the slope of one or more measured chamber parameters of a processing chamber.
[0067] In some embodiments, the data preparation tool 308 performs normalization on the various sensor data associated with the various processing chamber conditions. Normalization may include processing the incoming sensor data so that it appears similar between the various chambers and the various sensors used to acquire the data. In some embodiments, the data preparation tool 308 may include Figure 1 Aspects and / or features of the data preparation tool 116.
[0068] In some embodiments, the data preparation tool 308 is used to generate inputs for the classifier model 310. As further discussed in other embodiments, the classifier model 310 receives input data (e.g., raw sensor data, synthesized data, outputs of other models, etc.) to determine the state of the processing chamber (e.g., classify the processing chamber into one of a set of categories). In some embodiments, the classifier model receives sensor data and metrology data 307, which indicates the state of the processing chamber when processing a series of substrates (e.g., aged wafers). The classifier model can output a set of values indicating the likelihood that a substrate processed after an associated substrate in the series of substrates will result in a substrate having a processing result that satisfies a threshold condition (e.g., associated with a substrate processing procedure). For example, the classifier model 310 can output a series of probability values, each of which is associated with one or more substrates in the series of substrates (e.g., aged wafers). The probability values can indicate the likelihood that the associated substrate or a substrate processed after the associated substrate, when processed, will result in a substrate having a processing result that satisfies one or more threshold conditions associated with a substrate processing procedure (e.g., a substrate processing recipe). In some embodiments, as will be discussed further in other embodiments, the classifier model 310 can be incorporated into (e.g., using Figure 6 Trained and / or used Figure 5 implemented) in the machine learning model.
[0069] like Figure 3 As shown, the chamber monitoring and maintenance system 300 may include a chamber recovery tool 312. The chamber recovery tool 312 may receive the output from the classifier model 310. The chamber recovery tool 312 may approximate the chamber recovery curve by determining a series of PM indices associated with the output of the classifier model 310 (e.g., Figure 4 In some embodiments, a series of PM indices are based on current and previous data (e.g., Figure 1 The chamber recovery tool 312 may determine a threshold PM index that distinguishes whether the chamber condition is good enough to pass sample testing, whether more conditioning is needed, or whether further preventive maintenance is needed. PM index results for a series of PM wafers in conditioning processing may be obtained, which may enable in-situ chamber monitoring by the chamber monitoring and maintenance system 300.
[0070] In some embodiments, the chamber recovery tool 312 determines one or more PM indices associated with the state of the processing chamber. The chamber recovery tool 312 may perform statistical optimization calculations to determine threshold conditions and a mapping of the number of aged wafers and the PM index. An exemplary method for calculating the PM index may include the following:
[0071] PM Index th wafer=lnCUMSUM(Y est )*AVERAGE(Y est )
[0072] where Y est is the classifier model result associated with each aged wafer in a series of aged wafers in the ageing run.
[0073] In some embodiments, the chamber recovery tool 312 outputs instructions to the process controller 314 to change one or more operations associated with the substrate processing system 302. For example, the chamber recovery tool 312 may indicate the end of the seasoning process, and the process controller may change the operating parameters of the substrate processing system to enter a production operation mode. In another example, the chamber recovery tool 312 may indicate that the processing chamber is almost recovered and further seasoning is required to improve the condition of the processing chamber enough to enter a production operation mode. In another example, the chamber recovery tool 312 may indicate to the process controller that the preventive maintenance quality is insufficient (e.g., a threshold condition is not met) and instruct the process controller 314 to perform further preventive maintenance on the substrate processing system 302.
[0074] In some embodiments, the chamber recovery tool 312 approximates the expected chamber conditions during the aging run. The chamber recovery tool 312 interweaves the probabilities received from the classifier model 310 and estimates the conditions of a chamber at a given point in time. These conditions can be expressed as a PM index. Figure 4 As further discussed, these PM indices can be defined by logarithmic functions. The chamber recovery tool 312 can determine the number of substrates to process during a burn-in procedure (e.g., to satisfy a threshold condition of the processing chamber). For example, the PM index can be used to determine an optimal burn-in time (e.g., an optimal number of wafers to use in a burn-in process).
[0075] like Figure 3 As shown, the process controller 314 receives input from the chamber recovery tool 312 and provides instructions to the substrate processing system 302 to perform further operations. For example, the process controller 314 can instruct the substrate processing system 302 to continue further conditioning of one or more processing chambers. In another example, the process controller 314 can instruct the substrate processing system to end conditioning of one or more processing chambers. In another example, the process controller 314 can instruct the substrate processing system to perform preventive maintenance on one or more processing chambers.
[0076] like Figure 3As shown, the chamber recovery tool 312 may output instructions for display on a graphical user interface (GUI). The graphical user interface may output a chamber state prediction on the graphical user interface. In some embodiments, the chamber recovery tool 312 outputs instructions associated with one or more operations performed on the substrate processing system for display on the GUI 316. For example, the GUI 316 may display a notification to continue further seasoning of one or more processes. In another example, the GUI 316 may display an instruction to end seasoning of one or more process chambers. In another example, the GUI 316 may display a notification to perform preventive maintenance on one or more process chambers.
[0077] Figure 4 A graph 400 illustrating a chamber recovery curve according to some embodiments of the present disclosure is depicted. Graph 400 depicts a chamber recovery curve (e.g., generated by a chamber recovery tool 312). The recovery curve may include a first data sequence 406 indicating each PM index associated with a separate aged wafer 404. The recovery curve may indicate a threshold 408 PM index, where the condition of the processing chamber indicates satisfaction of the chamber recovery process. For example, an intersection between the threshold 408 and the first data sequence 406 may indicate an optimal number of aged wafers employed in a seasoning run such that the processing chamber is ready for production mode. Specifically, the processing chamber may process subsequent substrates, and the substrates are likely to have processing results that satisfy the threshold conditions of the substrate processing program. In some embodiments, the first data sequence 402 may intersect the threshold 408 and indicate that the chamber is ready for execution in production mode. In other embodiments, data sequences in other recovery curves may not intersect the threshold and may indicate that the chamber is not ready for production mode and may require further preventive maintenance and / or chamber seasoning.
[0078] Figures 5 to 7 Flowcharts illustrating exemplary methods 500-700 associated with training and / or using a machine learning model associated with chamber data according to some embodiments of the present disclosure are depicted. To simplify explanation, methods 500-700 are depicted and described as a series of actions. However, actions according to the present disclosure may occur in various orders and / or concurrently with other actions not presented and described herein. In addition, not all of the actions shown need to be performed to implement methods 500-700 according to the disclosed subject matter. In addition, those skilled in the art will understand and appreciate that methods 500-700 may alternatively be represented as a series of interrelated states via state diagrams or events.
[0079] Figure 5A flowchart depicting an example method 500 for monitoring and predicting conditions in a process chamber according to some embodiments of the present disclosure. The method 500 is performed by processing logic, which may include hardware (e.g., circuitry and dedicated logic, etc.), software (such as running on a general purpose computer system or a dedicated machine), or any combination thereof. In one embodiment, the server 120 and the Figure 1 The method may be performed by a trained machine learning model 190, and in some other embodiments, may be performed by one or more other machines not depicted in the figure. Figure 5 One or more boxes.
[0080] The method 500 may include receiving sensor data (e.g., associated with a process chamber performing a burn-in procedure) and processing the sensor data using the trained machine learning model 190. The trained model may be configured to generate one or more outputs based on the sensor data, the one or more outputs indicating (i) a state of the process chamber and (ii) a confidence level that the state represents the actual state of the process chamber, respectively. The state of the process chamber may indicate whether the process chamber has been restored and is ready for a production mode of operation, whether the process chamber requires further burn-in, and / or whether the process chamber requires further preventive maintenance.
[0081] At block 502, sensor data indicative of an environmental state of a processing chamber is received. The state of the processing chamber is measured during a seasoning procedure (e.g., while processing a series of substrates). The sensor data may be raw data, or the sensor data may be processed using one or more of feature extraction, mechanical models, and / or statistical models to prepare the sensor for input into a machine learning model. The sensor data may indicate one or more parameters of the processing chamber (e.g., temperature, pressure, vacuum conditions, and spectral data, among others).
[0082] In some embodiments, the sensor data further includes synthetic data or data processed from raw sensor data. For example, as described in the previous embodiments, various engineering tools can perform feature extraction and / or create artificial and / or virtual parameter combinations. Feature extractors (e.g., Figure 1 The data preparation tool 116 of the embodiment of the present invention can create various features by performing variable analysis (such as process control analysis, univariate limit violation analysis and / or multivariate limit violation analysis) on the raw sensor data. In some embodiments, the sensor data in multiple process chambers and / or process recipes are normalized to create comparable data sets with a common basis.
[0083] At block 504, sensor data is provided as input to a trained machine learning model, and at block 506, one or more outputs are obtained from the trained machine learning model. At block 508, a first set of values is determined based on the output(s) from the machine learning model. The first set of values indicates a likelihood that the associated substrate includes a first process result that satisfies a threshold condition.
[0084] In some embodiments, the processing logic determines the first quantity based on the first set of values. The first value may indicate a threshold number of substrates that causes the processing chamber to process subsequent substrates when the threshold number of substrates are processed after the preventive maintenance procedure. The subsequent substrates include second processing results that meet threshold conditions associated with the substrate processing procedure (e.g., uniformity requirements, critical dimension standards, film thickness standards, etc.). The threshold number of substrates and the subsequent substrates are processed according to the substrate processing procedure. The first quantity indicates how many aged wafers are needed to restore the state of the processing chamber to transition to a production operation mode.
[0085] At block 510, the first set of values is used to predict a first test result. The first test result indicates a likelihood that a first substrate being processed includes a second process result that satisfies a threshold condition after processing a series of substrates. The first test result may be predicted based on a comparison between the first number and the second number (e.g., a difference between the first number and the second number). For example, a threshold number of substrates is compared to a current number of substrates to determine whether a sufficient number of process wafers have been processed to restore a state of a process chamber after performing preventive maintenance. In some embodiments, the difference between the identified numbers may be displayed on a GUI (e.g., to indicate over-aging or under-aging of one or more process chambers).
[0086] In some embodiments, processing logic determines a recovery curve corresponding to the processing chamber. The recovery curve may include a distribution of indicators indicating a change in the likelihood that a first substrate processed includes a second processing result that satisfies a threshold condition after processing a series of substrates. The recovery curve may be used to predict the first test result. The recovery curve may include a cumulative distribution function associated with a set of values received in one or more outputs of the machine learning model.
[0087] At block 512, the first test result is optionally prepared for presentation on a graphical user interface (GUI). For example, the first test result may include a notification that further conditioning of the process chamber is to be performed, conditioning is to be terminated, and / or that preventive maintenance of the process chamber is required. At block 514, the operation of the process chamber may be changed based on the first test result. For example, the processing logic may transmit instructions to one or more process controllers to change one or more operations of the processing device (e.g., to continue conditioning, to end conditioning, to start a preventive maintenance procedure, etc.).
[0088] Figure 6 is an exemplary illustration of the training phase of a machine learning system according to some embodiments of the present disclosure. According to embodiments of the present disclosure, the machine learning system 170 may use method 600 to perform at least one of training a machine learning model, validating a machine learning model, or testing a machine learning model. In some embodiments, the method 600 may be used as described in relation to Figure 1 One or more operations of method 600 are described as being performed by the data set generator 174 of the server machine 172. It may be noted that Figure 1 The components described can be used to illustrate Figure 6 aspects.
[0089] At block 602, processing logic initializes a training set. At block 604, processing logic identifies the training set including (e.g., Figures 1 to 5 The first data input of the first sensor (e.g., a first training input and a first verification input) may include data indicating an environmental condition of the processing chamber during the aging process. In some embodiments, the first data input may include any chamber environmental parameters (e.g., temperature, pressure, vacuum conditions, plasma conditions, and spectral conditions, etc.). The first target input may be identified from old sensor data 144 of historical data 142 (e.g., stored in data storage 140).
[0090] At block 606, processing logic identifies a first target output of one or more data inputs (e.g., a first data input). The first target output indicates whether a first substrate processed by a first processing chamber in a first state (e.g., a first data input) according to a first substrate processing procedure resulted in the first substrate having a first processing result that satisfies a first threshold condition associated with the first substrate processing procedure. The first target input may be identified from sensor data 144 of historical data 142 (of data storage 140).
[0091] At block 608, 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 data inputs described herein), target outputs of the data inputs (e.g., one or more data inputs described herein), target outputs of the data (e.g., where the target outputs identify associated chamber states), and associations between the data input(s) and the target outputs.
[0092] At block 610 , processing logic adds the mapping data generated at block 504 to data set T.
[0093] At block 612, processing logic branches based on whether the dataset T is sufficient to train, validate, or test at least one of the machine learning models 190. If so (the "yes" branch), execution proceeds to block 614; if not (the "no" branch), execution continues back to block 604. It should be noted that in some embodiments, the sufficiency of the dataset T may be determined simply based on the number of input / output mappings in the dataset, while in some other embodiments, in addition to or instead of the number of input / output mappings, the sufficiency of the dataset T may be determined based on one or more other criteria (e.g., a measure of diversity and accuracy of data examples, etc.).
[0094] At block 610, processing logic provides a data set T to train, validate, or test the machine learning model 190. In some embodiments, the data set T is a training set, which is provided to the training engine 182 of the server machine 180 to perform training. In some embodiments, the data set T is a validation set, which is provided to the validation engine 184 of the server machine 180 to perform validation. In some embodiments, the data set T is a test set, which is provided to the testing engine 186 of the server machine 180 to perform testing. For example, in the case of a neural network, the input values (e.g., the numerical values associated with the data inputs) of a given input / output mapping are input to the neural network, and the output values (e.g., the numerical values associated with the target outputs) of the input / output mapping are stored in the output nodes of the neural network. The connection weights in the neural network are then adjusted according to a learning algorithm (e.g., back propagation, etc.), and this process is repeated for other input / output mappings in the data set T. After block 614, the machine learning model (e.g., machine learning model 190) may 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 chamber state component 194 (of the server machine 192) to identify a chamber state of the chamber (e.g., a likelihood that a substrate processed in the chamber has a processing result that satisfies a threshold condition) based on the received sensor data.
[0095] In some embodiments, the generated training data set is used to train a machine learning model and / or a physical model. The model can be trained to receive chamber sensor data as input. The model can output a state of the chamber (e.g., a likelihood that a substrate processed in the chamber has a processing result that satisfies a threshold condition) based on the received sensor data. In embodiments, the model is agnostic to the individual processing chambers used in production mode, the preventive maintenance performed, and / or the processing recipes. Thus, a model can be generated based on training data items (the training data items are generated based on a aging process run on a first processing chamber) and then used for a second processing chamber without having to perform transfer learning to adjust the model for the second processing chamber.
[0096] In one embodiment, the trained machine learning model is a regression model trained using regression. Examples of regression models are regression models trained using linear regression or Gaussian regression. The regression model predicts the value of Y given the known value of the X variable. The regression model can be trained using regression analysis, which can include interpolation and / or extrapolation. In one embodiment, the least squares method is used to estimate the parameters of the regression model. Alternatively, Bayesian linear regression, partial least squares regression, percentage regression, minimum absolute deviation, nonparametric regression, contextual optimization, and / or distance metric learning can be performed to train the regression model. In one embodiment, the trained machine learning model includes a support vector machine or other type of machine learning model.
[0097] In one embodiment, training a machine learning model employs ensemble modeling. Ensemble modeling may include training multiple models (e.g., two or more related but different analytical models) and synthesizing the results of each model into a single score or distribution (e.g., improving accuracy and reducing from any one machine learning model). For example, a trained machine learning model may include a bootstrap aggregation ("bagging") model. Bagging may include an aggregation phase and a bootstrap phase for each individual model. Bootstrapping includes sampling or creating a data set (e.g., using method 600), where each sample (e.g., sensor data associated with performing an aging procedure on a processing chamber) is selected and replaced (e.g., randomly) within a broader set of samples. As previously described, for each model, multiple models may be generated based on different subsets of a broader total training data set (e.g., historical data 142). Multiple models may output synthesized (e.g., aggregated) results to form a total score (e.g., a total confidence level) associated with the identified output of the model set (e.g., a predicted chamber state). In another example, a machine learning model may include a random forest model. The machine learning model utilizes multiple decision trees or multiple types of analytical models designed to predict outputs based on different variables and rules. The random forest model can mix multiple decision trees that analyze different sample data, evaluate different factors, or weight common variables by different amounts. The results of multiple decision trees can be converted into a composite result (e.g., a simple average or through further weighted aggregation) to predict the state of the chamber based on the input sensor data.
[0098] In one embodiment, the trained machine learning model is an artificial neural network (also referred to as a neural network for short). The artificial neural network can be, for example, a convolutional neural network (CNN) or a deep neural network. In one embodiment, the processing logic performs supervised machine learning to train the neural network.
[0099] Artificial neural networks generally include a classifier or regression layer that maps features to a target output space. For example, a convolutional neural network (CNN) hosts multiple layers of convolution filters. Pooling is performed at the lower layer and nonlinear problems can be solved. A multi-layer perceptron is usually attached to the top of the lower layer, which maps the top features extracted by the convolution layer to a decision (e.g., a classification output). A neural network can be a deep network with multiple hidden layers or a shallow network with zero or several (e.g., 1 to 2) hidden layers. Deep learning is a class of machine learning algorithms that uses a series of multi-layer nonlinear processing units for feature extraction and conversion. Each consecutive layer uses the output of the previous layer as input. Neural networks can learn in a supervised (e.g., classification) and / or unsupervised (e.g., pattern analysis) manner. Some neural networks (e.g., such as deep neural networks) include a multi-layer structure in which different layers learn representations of different levels corresponding to different levels of abstraction. In deep learning, each level learns to convert its input data into a slightly more abstract and complex representation.
[0100] Training of neural networks can be accomplished using supervised learning, which involves feeding a training dataset consisting of labeled inputs through the network, observing its output, defining an error (by measuring the difference between the output and the labeled value), and using techniques such as deep gradient descent and backpropagation to adjust the weights of all layers and nodes of the network to minimize the error. In many applications, repeating this process for many labeled inputs in the training dataset results in a network that produces the correct output when the input is different from the input present in the training dataset.
[0101] The trained machine learning model can be periodically or continuously retrained to achieve continuous learning and improvement of the trained machine learning model. The model can generate output based on the input, can perform actions based on the output, and can measure the results of the actions. In some cases, the results of the actions are measured within seconds or minutes, while in some cases, it takes longer to measure the results of the actions. For example, one or more additional processes can be performed before the results of the actions can be measured. The actions and the results of the actions can indicate whether the output is the correct output and / or the difference between what the output should be and what the output is. Therefore, the actions and the results of the actions can be used to determine the target output that can be used as a label for sensor measurement. Once the results of the actions are determined, the inputs (e.g., sensor data), the outputs of the trained machine learning model (e.g., chamber states), and the target results (e.g., target processing results of substrates) and actual measurement results (e.g., measured processing results of substrates) can be used to generate new training data items. The new training data items can then be used to further train the trained machine learning model. In some embodiments, this retraining process can be performed on a tool on a manufacturing system (e.g., (multiple) processing chambers).
[0102] Figure 7 The model training workflow 705 and model application workflow 717 for chamber monitoring and simulation according to aspects of the present disclosure are illustrated. In embodiments, the model training workflow 705 may be executed on a server that may or may not include a chamber classifier and / or chamber recovery predictor application, and the trained model is provided to the chamber classifier and / or chamber recovery predictor application that may execute the model application workflow 717. The model training workflow 705 may be executed on a server that may or may not include a chamber classifier and / or chamber recovery predictor application. Figure 1 The processing logic executed by the processor of the server 120) performs the model training workflow 705 and the model application workflow 717. For example, a processing device implemented by one or more machine learning modules and / or other software and / or firmware executed on the processing device can be provided to implement one or more of these workflows 705, 717.
[0103] The model training workflow 705 is used to train one or more machine learning models (e.g., deep learning models) to perform one or more determination, prediction, and modification tasks associated with chamber classification and chamber recovery prediction (e.g., determining whether a processing chamber is properly aged and ready for a production mode of operation). The model application workflow 717 is used to apply one or more trained machine learning models to perform determination and / or adjustment tasks on chamber data (e.g., raw sensor data and synthetic data indicating a processing chamber state). The one or more machine learning models may receive process result data (e.g., chamber state and / or recovery curves).
[0104] Various machine learning outputs are described herein. A specific number and arrangement of machine learning models are described and shown. However, it should be understood that the number and type of machine learning models used and the arrangement of such machine learning models may be modified to achieve the same or similar end results. Therefore, the arrangement of machine learning models described and shown is merely an example and should not be construed as limiting.
[0105] In an embodiment, one or more machine learning models are trained to perform one or more of the following tasks. Each task may be performed by a separate machine learning model. Alternatively, a single machine learning model may perform each task or a subset of tasks. Additionally or alternatively, different machine learning models may be trained to perform different combinations of tasks. In an example, one or several machine learning models may be trained, where the trained machine learning (ML) model is a single shared neural network with multiple shared layers and multiple higher-level different output layers, where each output layer outputs different predictions, classifications, and recognitions, etc. One or more trained machine learning models may be trained to perform tasks as follows:
[0106] a. Chamber classifier. As previously described, the chamber classifier may receive and process various input data, such as sensor data, pre-processed data, and synthesized data indicating the state of a processing chamber during a seasoning process (e.g., processing a series of seasoned wafers). The chamber classifier may output various values corresponding to various seasoned wafers, which indicate the likelihood that processing of the substrate under the corresponding chamber conditions will result in a substrate having a processing result that satisfies a threshold condition associated with the substrate process (e.g., substrate and / or manufacturing recipe).
[0107] b. Chamber Recovery Predictor. The chamber recovery predictor may receive data from the chamber classifier and generate a chamber recovery curve associated with recovery of a processing chamber associated with a preventive maintenance and / or aging treatment. The chamber recovery predictor may determine a preventive maintenance indicator that indicates a change in the likelihood that the chamber will meet conditions to move from a recovery operating mode to a production operating mode. For example, the chamber recovery predictor may identify when a condition of the chamber meets a threshold condition such that a substrate processed under the chamber condition results in the substrate having a processing result that meets the threshold condition associated with one or more processing procedures. In some embodiments, the chamber recovery predictor outputs a preventive maintenance indicator value associated with the likelihood that the chamber will meet the threshold condition to transition to a production operating mode.
[0108] One type of machine learning model that can be used to perform some or all of the above tasks is an artificial neural network, such as a deep neural network. An artificial neural network generally includes a feature representation component with a classifier or regression layer that maps features to a desired output space. For example, a convolutional neural network (CNN) hosts multiple layers of convolutional filters. Pooling is performed at lower layers and nonlinear problems can be solved, and a multi-layer perceptron is usually attached on top of the lower layers, which maps the top features extracted by the convolutional layer to decisions (e.g., classification output). Deep learning is a class of machine learning algorithms that uses a series of multi-layer nonlinear processing units for feature extraction and transformation. Each successive layer uses the output of the previous layer as input. Deep neural networks can be learned in a supervised (e.g., classification) and / or unsupervised (e.g., pattern analysis) manner. A deep neural network includes a multi-layer structure in which different layers learn different levels of representation corresponding to different levels of abstraction. In deep learning, each layer learns to transform its input data into a slightly more abstract and complex representation. For example, in plasma processing tuning, the raw input may be chamber data (e.g., raw sensor chamber measurement parameters such as temperature, pressure, and vacuum conditions within the chamber, etc.); the second layer may consist of feature data associated with parameters associated with a substrate processing program or process recipe; and the third layer may include historical thresholds and chamber states under similar conditions. It is worth noting that the deep learning process can learn by itself which features are best placed at which level. The "depth" in "deep learning" refers to the number of layers that the data passes through. More precisely, deep learning systems have a considerable credit assignment path (CAP) depth. CAP is a chain of transformations from input to output. CAP describes the potential cause-effect relationship between input and output. For a feedforward neural network, the depth of CAP can be the depth of the network and can also be the number of hidden layers plus one. For a recurrent neural network where a signal may propagate through a layer multiple times, the CAP depth may be infinite.
[0109] In one embodiment, one or more machine learning models are recurrent neural networks (RNNs). RNNs are a class of neural networks that include memory that enables the neural network to capture time dependencies. RNNs are able to learn input-output mappings that depend on current inputs and past inputs. RNNs will resolve past and future sensor and / or process result measurements and make predictions based on such continuous sensor information (e.g., sensor data indicating the changing conditions of a processing chamber). RNNs can be trained using a training data set to generate a fixed number of outputs (e.g., to determine a set of substrate processing rates, and to determine chamber conditions, production mode requirements, and condition thresholds). One class of RNNs that can be used is a long short-term memory (LSTM) neural network.
[0110] Training of neural networks can be accomplished using supervised learning, which involves feeding a training dataset consisting of labeled inputs through the network, observing its output, defining an error (by measuring the difference between the output and the labeled value), and using techniques such as deep gradient descent and backpropagation to adjust the weights of all layers and nodes of the network to minimize the error. In many applications, repeating this process for many labeled inputs in the training dataset results in a network that produces the correct output when the input is different from the input present in the training dataset.
[0111] For the model training workflow 705, a training data set containing hundreds, thousands, tens of thousands, hundreds of thousands, or more chamber data 710 (e.g., sensor data, synthetic data indicating the state of the associated processing chamber) should be used to form the training data set. In an embodiment, the training data set may also include associated processing result data 712 (e.g., measured parameters of the substrate, such as critical dimensions, uniformity requirements, and film thickness results, etc.) used to form the training data set, wherein each data point may include various labels or classifications of one or more types of useful information. For example, each case may include data indicating that one or more processing chambers are undergoing a seasoning process, and associated processing results of the substrates evaluated during and / or after the seasoning process. This data may be processed to generate one or more training data sets 736 for training one or more machine learning models. The machine learning model may be trained, for example, to automatically classify a chamber (e.g., a chamber meets the conditions to operate in a production mode) and / or predict a threshold number of substrates when the chamber meets the threshold conditions to operate in a production mode.
[0112] In one embodiment, generating one or more training data sets 736 includes performing a chamber recovery process 200A by performing preventive maintenance 202 and aging 206 and evaluating sample tests 208 to determine whether the chamber meets a threshold condition (e.g., box 210). One or more markers may be used in various iterations of the chamber recovery process 200A to identify the sample tests 208 and the measured chamber conditions as meeting the threshold conditions, thereby initiating a production mode, further aging, and / or further preventive maintenance corresponding to the processing chamber. The markers used may depend on what the particular machine learning model is being trained to do. In some embodiments, as described in other embodiments, the chamber data, process results, and / or sample test results may be represented as vectors, and the process rates may be represented as one or more matrices.
[0113] To implement training, processing logic inputs the training data set(s) 736 into one or more untrained machine learning models. The machine learning model may be initialized before the first input is input into the machine learning model. Processing logic trains the untrained machine learning model(s) based on the training data set(s) to generate one or more trained machine learning models that perform the various operations described above.
[0114] Training may be performed by inputting one or more of the chamber data 710 and the process result data 712 into the machine learning model one at a time.
[0115] Machine learning models process inputs to generate outputs. Artificial neural networks include an input layer consisting of values in data points. The next layer is called a hidden layer, and the nodes at the hidden layer each receive one or more input values. Each node contains parameters (e.g., weights) applied to the input values. Therefore, each node essentially inputs the input values to a multivariate function (e.g., a nonlinear mathematical transformation) to generate an output value. The next layer may be another hidden layer or an output layer. In either case, the nodes at the next layer receive output values from the nodes at the previous layer, and each node applies weights to these values and then generates its own output value. This operation can be performed at each layer. The last layer is the output layer, in which there is a node for each class, prediction, and / or output that the machine learning model can produce.
[0116] Thus, the output may include one or more predictions or inferences. For example, the output predictions or inferences may include one or more modifications to the plasma processing data (e.g., modifications to one or more plasma exposure periods). The processing logic may process the substrate using an updated recipe (e.g., with the identified modifications) and receive an updated thickness distribution. The processing logic may compare the updated thickness distribution with the target thickness distribution and determine whether a threshold criterion is met (e.g., the thickness value measured across the wafer surface falls within a target threshold window). The processing logic determines an error (i.e., a classification error or a prediction error) based on the difference between the identified chamber state or recovery prediction and the evaluated processing result (e.g., a sample test result). The processing logic adjusts the weights of one or more nodes in the machine learning model based on the error. An error term or increment may be determined for each node in the artificial neural network. Based on this error, the artificial neural network adjusts one or more parameters of one or more of its nodes (the weights of one or more inputs of the node). The parameters may be updated in a back-propagation manner so that the nodes of the highest layer are updated first, and then the nodes of the next layer are updated, and so on. The artificial neural network includes multiple layers of "neurons", each of which receives as input values from the neurons of the previous layer. The parameters of each neuron include weights associated with the values received from each neuron of the previous layer. Thus, adjusting the parameters may include adjusting the weights assigned to each input of one or more neurons at one or more layers in the artificial neural network.
[0117] Once the model parameters have been optimized, model validation may be performed to determine whether the model has been improved and to determine the current accuracy of the deep learning model. After one or more rounds of training, the processing logic may determine whether a stopping criterion has been met. The stopping criterion may be a target level of accuracy, a target number of processed images from a training data set, a target amount of change in a parameter over one or more previous data points, a combination of the above, and / or other criteria. In one embodiment, the stopping criterion is met when at least a minimum number of data points have been processed and at least a threshold accuracy has been reached. The threshold accuracy may be, for example, 70%, 80%, or 90% accuracy. In one embodiment, the stopping criterion is met if the accuracy of the machine learning model has stopped improving. If the stopping criterion is not met, further training is performed. If the stopping criterion has been met, the training may be complete. Once the machine learning model is trained, a retained portion of the training data set may be used to test the model.
[0118] Once the one or more trained machine learning models 738 are generated, the one or more trained machine learning models 738 may be stored in a model storage 745 and may be added to a substrate processing rate determination and / or process adjustment application. The substrate processing rate determination and / or process adjustment application may then use the one or more trained ML models 738 and additional processing logic to implement an automatic mode in which manual user input is minimized or, in some cases, even eliminated.
[0119] For the model application workflow 717, according to one embodiment, input data 762 may be input into a process rate determiner 767, which may include a trained machine learning model. Based on the input data 762, the process rate determiner 767 outputs information indicating a likelihood that the associated substrate includes a first process result that satisfies a threshold condition of a substrate processing procedure.
[0120] According to one embodiment, the input data 762 may be input to a chamber recovery predictor 764, which may include a trained machine learning model. Based on the input data 762, the chamber recovery predictor 764 outputs a chamber recovery curve, a threshold number of aged wafers required for a chamber to meet a threshold condition to operate in a production operating mode, and / or a first test result indicating a likelihood that a first substrate processed after processing a series of substrates includes a processing result that meets a threshold condition (e.g., which is associated with a substrate processing recipe).
[0121] Figure 8 800 is a block diagram of an exemplary computing device 800 that operates in accordance with one or more aspects of the present disclosure. In various illustrative examples, various components of computing device 800 may represent Figure 1 Various components of client device 150 , metrology system 110 , server 120 , data storage 140 , manufacturing system 102 , and machine learning system 170 are shown.
[0122] The exemplary computing device 800 can be connected to other computer devices in a LAN, an intranet, an extranet, and / or an internet. The computing device 800 can operate as a server in a client-server network environment. The computing device 800 can be a personal computer (PC), a set-top box (STB), a server, a network router, a switch or a bridge, or any device capable of (sequentially or otherwise) executing a set of instructions (which specifically specifies the actions to be taken by this device). In addition, although only a single exemplary computing device is described, the term "computer" should also be deemed to include any collection of computers that individually or jointly execute a set (or multiple sets) of instructions to perform any one or more methods discussed herein.
[0123] The exemplary computing device 800 may include a processing device 802 (also referred to as a processor or CPU), a main memory 804 (e.g., read-only memory (ROM), flash memory, dynamic random access memory (DRAM), such as synchronous DRAM (SDRAM), etc.), a static memory 806 (e.g., flash memory and static random access memory (SRAM), etc.), and an auxiliary memory (e.g., a data storage device 818), which may communicate with each other via a bus 830.
[0124] The processing device 802 represents one or more general-purpose processing devices, such as a microprocessor and a central processing unit, etc. More specifically, the processing device 802 can be a complex instruction set computing (CISC) microprocessor, a reduced instruction set computing (RISC) microprocessor, a very long instruction word (VLIW) microprocessor, a processor implementing other instruction sets, or a processor implementing a combination of instruction sets. The processing device 802 can also be one or more special-purpose processing devices, such as an integrated circuit (ASIC), a field programmable gate array (FPGA), a digital signal processor (DSP), and a network processor or the like. According to one or more aspects of the present disclosure, the processing device 802 can be configured to perform the implementation Figure 5 and Figure 6 Instructions of method 500 and method 600 shown in .
[0125] The exemplary computing device 800 may further include a network interface device 708, which may be communicatively coupled to a network 820. The exemplary computing device 800 may further include a video display 810 (e.g., a liquid crystal display (LCD), a touch screen, or a cathode ray tube (CRT)), an alphanumeric input device 812 (e.g., a keyboard), a cursor control device 814 (e.g., a mouse), and a sound signal generating device 816 (e.g., a speaker).
[0126] The data storage device 818 may include a machine-readable storage medium (or, more specifically, a non-transitory machine-readable storage medium) 828 on which one or more sets of executable instructions 822 are stored. According to one or more aspects of the present disclosure, the executable instructions 822 may include Figure 5 and Figure 6 Executable instructions associated with the execution methods 500 and 600 shown in FIG.
[0127] The executable instructions 822 may also reside completely or at least partially within the main memory 804 and / or the processing device 802 during execution by the exemplary computing device 800, the main memory 804 and the processing device 802 also constituting computer-readable storage media. The executable instructions 822 may further be transmitted or received over a network via the network interface device 808.
[0128] Although Figure 8 828 as a single medium, but the term "computer-readable storage medium" should be considered to include a single medium or multiple media (e.g., centralized or distributed databases, and / or associated caches and servers) that store one or more sets of operating instructions. The term "computer-readable storage medium" should also be considered to include any medium that can store or encode a set of instructions for machine execution, which causes the machine to perform any one or more of the methods described herein. Therefore, the term "computer-readable storage medium" should be understood to include, but is not limited to, solid-state memory, optical and magnetic media.
[0129] Some portions of the foregoing detailed description are presented in terms of algorithms and symbolic representations of operations on data bits within a computer memory. These algorithmic descriptions and representations are the means used by those skilled in the art of data processing to most effectively convey the content of their work to others skilled in the art. An algorithm is generally considered herein to be a self-consistent sequence of steps leading to a desired result. The steps are those that require physical manipulation of physical quantities. Typically, but not necessarily, these quantities take the form of electrical or magnetic signals that can be stored, transferred, combined, compared, and otherwise manipulated. At times, it has proven convenient, primarily for reasons of common usage, to refer to these signals as bits, values, elements, symbols, characters, terms, numbers, or the like.
[0130] It should be remembered, however, that all of these and similar terms are to be associated with the appropriate physical quantities and are merely convenient labels applied to these quantities. Unless otherwise specifically stated, as will be apparent from the following discussion, it will be understood that discussions throughout this specification using terms such as "identify," "determine," "store," "adjust," "result," "return," "compare," "create," "stop," "load," "copy," "throw," "replace," and "execute" or the like refer to the actions and processes of a computer system or similar electronic computing device to process and convert data represented as physical (electronic) quantities in the computer system's registers and memories to other data similarly represented as physical quantities in the computer system's registers or memories or other such information storage, transmission, or display devices.
[0131] Examples of the present disclosure also relate to an apparatus for performing the methods described herein. The apparatus may be specially constructed for the desired purpose, or the apparatus may be a general-purpose computer system selectively programmed by a computer program stored in a computer system. Such a computer program may be stored in a computer-readable storage medium, such as, but not limited to, any type of disk including optical disks, compact disk read-only memory (CD-ROM) and magneto-optical disks, read-only memory (ROM), random access memory (RAM), erasable programmable read-only memory (EPROM), electronically erasable programmable read-only memory (EEPROM), magnetic disk storage media, optical storage media, flash memory devices, other types of machine-accessible storage media, or any type of medium suitable for storing electronic instructions, each coupled to a computer system bus.
[0132] The methods and displays presented herein are not inherently associated with any particular computer or other device. Various general purpose systems may be used in conjunction with the programs taught herein, or it may prove convenient to construct more specialized equipment to perform the required method steps. The structures required for various of these systems will appear in the description below. In addition, the scope of the present disclosure is not limited to any particular programming language. It will be appreciated that a variety of programming languages may be used to implement the teachings of the present disclosure.
[0133] It should be understood that the above description is intended to be illustrative, not restrictive. After reading and understanding the above description, many other implementations will be apparent to those skilled in the art. Although the present disclosure describes specific examples, it will be appreciated that the systems and methods of the present disclosure are not limited to the examples described herein, but may be modified within the scope of the claims to implement the systems and methods of the present disclosure. Therefore, the description and drawings should be considered illustrative, not restrictive. Therefore, the scope of the present disclosure should be determined with reference to the appended claims and the full scope of equivalents to which the claims are entitled.
Claims
1. A method comprising the following steps: receiving, by a processing device, sensor data indicative of a state of an environment of a processing chamber, the processing chamber processing a series of substrates according to a substrate processing routine, the series of substrates being processed following a preventative maintenance routine performed on the processing chamber; determining, by the processing device based on the sensor data, a first set of values, each of the first set of values being associated with one of the series of substrates, and the first set of values indicating a likelihood that the associated substrate includes a first processing result that satisfies a threshold condition of the substrate processing procedure; predicting, by the processing device based on the first set of values, a first test result indicating a likelihood that a first substrate processed after processing the series of substrates includes a second processing result that satisfies the threshold condition; and At least one of: a) preparing the first test result for presentation on a graphical user interface (GUI) or b) changing the operation of the processing chamber based on the first test result is performed by the processing device.
2. The method according to claim 1, further comprising the steps of: A first number is determined based on the first set of values, the first number indicating a threshold number of substrates that, when the threshold number of substrates is processed after the preventive maintenance procedure, causes the processing chamber to process subsequent substrates having a second processing result that satisfies the threshold condition, wherein the threshold number of substrates and the subsequent substrates are processed according to the substrate processing procedure.
3. The method of claim 2, wherein the series of substrates comprises a second number of substrates, the method further comprising the steps of: The first number is compared to the second number, wherein the first test result is predicted based on the comparison between the first number and the second number.
4. The method according to claim 3, further comprising the steps of: determining a first difference between the first amount and the second amount; and A notification indicating the first difference is prepared to be presented on the GUI.
5. The method according to claim 1, further comprising the steps of: The processing chamber is determined to meet one or more conditions for a production mode of operation based on the first test result.
6. The method according to claim 1, further comprising the steps of: A recovery curve corresponding to the processing chamber is determined, the recovery curve comprising an indicator distribution indicating a change in the likelihood that the first substrate processed includes the second processing result satisfying the threshold condition after processing the series of substrates, wherein the first test result is further predicted using the indicator distribution.
7. The method of claim 6, wherein the recovery curve further comprises a cumulative distribution function associated with the first set of values.
8. The method according to claim 1, further comprising the steps of: using the sensor data as input to a machine learning model; and One or more outputs of the machine learning model are obtained, the one or more outputs indicating the first set of values.
9. The method of claim 1, wherein the machine learning model comprises at least one of a bootstrap aggregation model, a random forest model, a decision tree model, or a partial least squares regression (PLS) model.
10. A method for training a machine learning model to predict a condition of a processing chamber, the method comprising the steps of: Generating training data for the machine learning model, wherein the step of generating the training data comprises the following steps: identifying a first training input having first sensor data indicating a first state of a first processing chamber after performing a preventive maintenance procedure; and identifying a first target output for the first training input, wherein the first target output indicates whether a first substrate processed by the first processing chamber in the first state according to a first substrate processing procedure results in the first substrate having a first processing result that satisfies a first threshold condition, the first threshold condition being associated with the first substrate processing procedure; and The training data is provided to train the machine learning model on (i) a set of training inputs including the first training inputs and (ii) a set of target outputs including the first target outputs, wherein the trained machine learning model will receive new inputs having new sensor data, wherein the new sensor data indicates a new state of a new processing chamber, and the new output indicates whether a new substrate processed by the new processing chamber in the new state according to the first substrate processing procedure results in the new substrate having a new processing result that satisfies the threshold condition.
11. The method of claim 10, wherein the first sensor data comprises one or more measurements acquired after a first series of substrates have been processed in the first processing chamber.
12. The method of claim 10, wherein the first sensor data comprises one or more measurements acquired before the first processing chamber processed a first series of substrates.
13. The method of claim 10, wherein each training input in the set of training inputs is mapped to a target output in the set of target outputs.
14. A non-transitory machine-readable storage medium comprising instructions which, when executed by a processing device, cause the processing device to perform operations comprising: receiving sensor data indicative of a state of an environment of a processing chamber, the processing chamber processing a series of substrates according to a substrate processing procedure, the series of substrates being processed following a preventative maintenance procedure performed on the processing chamber; determining a first set of values based on the sensor data, the first set of values each being associated with one of the series of substrates and indicating a likelihood that the associated substrate includes a first processing result that satisfies a threshold condition of the substrate processing procedure; predicting a first test result based on the first set of values, the first test result indicating a likelihood that a first substrate processed after processing the series of substrates includes a second processing result that satisfies the threshold condition; and At least one of: a) preparing the first test result for presentation on a graphical user interface (GUI), or b) changing the operation of the processing chamber based on the first test result.
15. The non-transitory machine-readable storage medium of claim 14, the operations further comprising: A first number is determined based on the first set of values, the first number indicating a threshold number of substrates that, when the threshold number of substrates is processed after the preventive maintenance procedure, causes the processing chamber to process subsequent substrates having a second processing result that satisfies the threshold condition, wherein the threshold number of substrates and the subsequent substrates are processed according to the substrate processing procedure.
16. The non-transitory machine-readable storage medium of claim 15, wherein the series of substrates includes a second number of substrates, wherein the operations further comprise: The first number is compared to the second number, wherein the first test result is predicted based on the comparison between the first number and the second number.
17. The non-transitory machine-readable storage medium of claim 16, the operations further comprising: determining a first difference between the first amount and the second amount; and A notification indicating the first difference is prepared to be presented on the GUI.
18. The non-transitory machine-readable storage medium of claim 14, the operations further comprising: A recovery curve corresponding to the processing chamber is determined, the recovery curve comprising an indicator distribution indicating a change in the likelihood that the first substrate processed includes the second processing result satisfying the threshold condition after processing the series of substrates, wherein the first test result is further predicted using the indicator distribution.
19. The non-transitory machine-readable storage medium of claim 14, wherein the operations further comprise the following operations: using the sensor data as input to a machine learning model; and One or more outputs of the machine learning model are obtained, the one or more outputs indicating the first set of values.
20. The non-transitory machine-readable storage medium of claim 19, wherein the machine learning model comprises at least one of a bootstrap aggregation model, a random forest model, a decision tree model, or a partial least squares regression (PLS) model.
