Performance management of semiconductor substrate tools

Abstract

Active management of semiconductor substrate tools. A machine learning model is used to predict future performance characteristics of such tools. In some examples, the model may diagnose problems of environmental conditions of the tool or tool environment. In some examples, the model may recommend one or more remedial actions to maintain sufficient performance of the substrate tool.

Description

[0001] This application was filed as a PCT international patent application on May 24, 2023, and claims the benefit and priority of U.S. Provisional Patent Application Serial No. 63 / 346,358 filed on May 27, 2022, the entire disclosure of which is incorporated herein by reference in its entirety. Technical Field

[0002] The present disclosure relates to semiconductor component manufacturing and managing the performance of tools involved in the semiconductor component manufacturing process. Background Art

[0003] As part of the formation of a semiconductor chip or other type of integrated circuit (IC), a semiconductor substrate is constructed or manufactured. The components of the final IC can be incorporated into the substrate through a series of manufacturing steps. The manufacturing steps may include a deposition step in which a thin film layer is added to the substrate. The substrate can then be coated with a photoresist, and the circuit pattern of the mask can be projected onto the substrate using photolithography techniques. The etching process can then be performed using an etching tool.

[0004] In order for a finished substrate (e.g., a wafer or batch of wafers) to be usable, each tool involved in the substrate manufacturing process must perform within predefined acceptable operating tolerances for the aspects of the substrate for which the tool is responsible. If even a single tool in the manufacturing process performs outside of its tolerance, defects of sufficient magnitude in the substrate may result in the need to discard all wafers in that run or subsequent manufacturing runs. Over time and through use (e.g., with each successive manufacturing run), the operational performance of a particular manufacturing tool may degrade or drift such that the tool eventually requires recalibration in order to perform within specifications. Summary of the invention

[0005] In general, the present disclosure relates to active maintenance of semiconductor substrate tools (also referred to herein as tools). Substrate tools may include, for example, but not limited to, manufacturing tools (such as deposition tools, lithography tools, oxidation tools, epitaxial reactors, diffusion tools, ion implantation tools, etching tools, and chemical mechanical planarization (CMP) tools), inspection tools, cutting tools, grinding tools, polishing tools, metrology and other measurement tools. Although the present disclosure may relate to certain substrate manufacturing tools (such as deposition tools) in specific examples, the features of the present disclosure can be easily applied to other substrate tools that do not perform manufacturing functions.

[0006] More generally, the present disclosure relates to improving performance consistency among substrate tools of the same type.

[0007] Features of the present disclosure may be implemented as systems, computer-implemented methods, and as instructions stored on non-transitory computer-readable storage devices.

[0008] According to certain aspects, the present disclosure relates to using a machine learning model to predict future states of a substrate tool.

[0009] According to certain aspects, the present disclosure relates to using a machine learning model to predict poor performance of a substrate tool.

[0010] According to further aspects, the present disclosure relates to using a machine learning model to diagnose causes of poor performance or predicted poor performance of a substrate tool.

[0011] According to further aspects, the present disclosure relates to using a machine learning model to diagnose causes of performance differences or predicted performance differences between tools of the same type of substrate.

[0012] According to further aspects, the present disclosure relates to using a machine learning model to determine adjustments to a substrate tool to correct predicted underperformance of the substrate tool.

[0013] According to further aspects, the present disclosure relates to using a machine learning model to determine adjustments for a substrate tool to reduce a performance difference between the substrate tool and another substrate tool of the same type.

[0014] According to certain specific aspects, a computer-implemented method for predicting a future performance state of a semiconductor substrate tool after a future use of the tool (e.g., a future manufacturing run, a future inspection run, a future metrology run) includes: providing a current performance state of the manufacturing tool to a trained machine learning model; providing operating data of the tool to the trained machine learning model; and receiving the predicted future performance state and recommended recalibration of the tool, wherein the predicted future performance state and the recommended recalibration are determined by the trained machine learning model based on the current performance state and the operating data.

[0015] According to other specific aspects, a computer-implemented method for predicting a future performance state of a semiconductor substrate manufacturing tool after a future use of the tool (e.g., a future manufacturing run, a future inspection run, a future metrology run) includes: a trained machine learning model receiving a current performance state of the manufacturing tool; the trained machine learning model receiving operating data of the tool; and the machine learning model generating the predicted future performance state and recommended recalibration of the tool, wherein the predicted future performance state and the recommended recalibration are based on the current performance state and the operating data.

[0016] According to another specific aspect, a computer-implemented method includes training a machine learning model to generate a predicted future performance state of a semiconductor substrate manufacturing tool and generate a recommended recalibration of the semiconductor substrate manufacturing tool based on a current performance state of the tool and operational data associated with the tool, wherein the operational data includes data generated by sensors associated with the tool or with the tool's surrounding environment and data generated by automatic tests performed by the tool.

[0017] According to another specific aspect, a computer-implemented method for predicting a predicted future performance state of a substrate tool after a future use of the substrate tool (e.g., a future manufacturing run, a future inspection run, a future metrology run) includes: providing a current performance state of the substrate tool to a trained machine learning model; providing operating data of the substrate tool to the trained machine learning model; and receiving the predicted future performance state and recommended recalibration of the substrate tool, the predicted future performance state and the recommended recalibration being determined by the trained machine learning model based on the current performance state and the operating data.

[0018] According to other specific aspects, a method for predicting a future performance state of a substrate tool after a future use of the substrate tool (e.g., a future manufacturing run, a future inspection run, a future metrology run) includes: a trained machine learning model receiving a current performance state of the substrate tool; the trained machine learning model receiving operating data of the substrate tool; and the trained machine learning model generating a predicted future performance state and a recommended recalibration of the substrate tool, the predicted future performance state and the recommended recalibration being based on the current performance state and the operating data.

[0019] According to another specific aspect, a computer-implemented method includes: training a machine learning model to generate a predicted future performance state of a substrate tool and generate a recommended recalibration of the substrate tool based on a current performance state of the substrate tool and operational data associated with the substrate tool, wherein the operational data includes data generated by sensors associated with the substrate tool or with the surrounding environment of the substrate tool and data generated by automatic tests performed by the substrate tool.

[0020] The present invention is provided to introduce a series of concepts in a simplified form, which will be further described in the following detailed description. The present invention is not intended to identify the key features or essential features of the claimed subject matter, nor is it intended to be used to limit the scope of the claimed subject matter. Additional aspects, features and / or advantages of the examples will be listed in part in the description below, and in part will be apparent from the description, or can be learned by practicing the present disclosure. BRIEF DESCRIPTION OF THE DRAWINGS

[0021] Non-limiting and non-exhaustive examples are described with reference to the following figures.

[0022] Figure 1 An example system for managing performance of a semiconductor substrate tool according to the present disclosure is schematically depicted.

[0023] Figure 2 Describes the basis Figure 1 Example embodiments of the system Figure 1 Details of a part of a system.

[0024] Figure 3 Describes the use of Figure 1 An example of a system using a machine learning model to manage the performance of a semiconductor substrate tool according to an example of the present disclosure.

[0025] Figure 4 Describes the basis Figure 1 Example embodiments of the system Figure 3 Details of additional data.

[0026] Figure 5 Describes the use of Figure 1 A system and method for managing performance of a semiconductor substrate tool according to examples of the present disclosure. DETAILED DESCRIPTION

[0027] Examples of the present disclosure describe systems, methods, and computer readable products for improving substrate manufacturing. More specifically, examples of the present disclosure describe systems, methods, and computer readable products for improving performance of a substrate tool and maintaining the improved performance of a substrate tool.

[0028] An example of such a substrate is a semiconductor wafer consisting of die. A given wafer has a yield, which may refer to the percentage of substrate die that will meet one or more defined operational, quality, or other acceptable standards. Defects in a wafer caused by manufacturing tools (e.g., deposition tools, photolithography tools, etching tools, CMP tools, and other tools described above) performing outside of specifications (also referred to herein as out of specification) may reduce the yield of the wafer. Depending on the yield and / or type, number and / or severity of defects in a wafer, wafers from a given manufacturing run or even entire batches of wafers may have to be discarded, which is expensive.

[0029] Substrate inspection and measurement tools (eg, metrology tools) that fall out of specification may cause similar problems, for example, by identifying defects in wafers that are not present and / or failing to identify defects in wafers that are present.

[0030] The present disclosure relates to a systematic approach for substrate tool maintenance. The systematic approach can be readily applied to many different types of substrate tools, such as manufacturing tools, inspection tools, metrology tools, etc. Advantageously, and regardless of the type of substrate tool, the systematic approach can reduce the amount of time a substrate tool is out of production and / or reduce the number of wafers that must be scrapped.

[0031] In addition, the disclosed systematic approach can improve performance consistency between different tools of the same type. For example, in the case of manufacturing tools, a given manufacturing facility may have multiple tools that perform the same manufacturing steps. Even if all of these tools are within specifications (also referred to herein as within specifications), there may be differences in performance between tools, which may adversely lead to inconsistencies between the finished substrates produced by the facility. Furthermore, such inconsistencies may be amplified by performance inconsistencies of other tools involved later in the manufacturing process.

[0032] During substrate manufacturing, the substrate undergoes many process steps performed by various tools. Such tools may include, for example, but not limited to, deposition tools, etching tools, lithography tools, oxidation tools, epitaxial reactors, diffusion tools, ion implantation tools, and coating tools. The calibration or adjustment of each tool is critical to how the tool performs. For example, a deposition tool within specification may deposit a 300 angstrom layer on a wafer within a tolerance of 10 angstroms. For each run of the tool (a run may correspond to the manufacture of a wafer batch or any other use iteration for a given type of tool), the calibration parameters from the tool may be offset due to any of several different factors. Inspection and measurement tools are used to inspect and measure various aspects of the wafer.

[0033] Furthermore, tools of the same type may all perform within specification but have variations. For example, a particular deposition tool may deposit a 300 angstrom layer within specification at 305 angstroms, while another deposition tool may deposit the same layer at 291 angstroms, resulting in inconsistencies in the final product produced by a facility or fabricator using the two tools.

[0034] According to current methods of tool management and maintenance, each tool is periodically taken out of production for testing and maintenance. In some cases, there is a periodic inspection of the wafers manufactured by the tool to see if the tool is performing within specifications, and if not, the tool is taken out of production for maintenance. These methods can be time consuming and expensive, and can be both over-inclusive and under-inclusive in identifying tools that need maintenance, because some tools are taken out of production prematurely, while other tools are taken out of production only after they have already performed out of specification, requiring wafers to be discarded. In addition, when a tool begins to perform out of specification, the cause must be manually determined, and therefore the remedy manually determined, which can extend the amount of time the tool is out of production.

[0035] According to the present disclosure, a trained machine learning model is used to predict, prior to use of a given tool (e.g., prior to a manufacturing run), whether the tool will perform within or outside of specifications for an upcoming (or future) use. Similarly, the model can predict when maintenance will be required for a given tool. The model can also determine the reasons for performance differences between tools. In this way, tool maintenance and management can be performed at a time that minimizes the cost of performing maintenance too frequently or too infrequently.

[0036] Additionally, the machine learning model is configured to determine tool parameters that have deviated or are beginning to deviate, thereby identifying and recommending tool parameters that will require recalibration or other adjustments to keep the tool within specifications.

[0037] In addition, the machine learning model is configured to perform predictive maintenance modeling for different tool types by considering input data specific to each tool type. Specifically, the model is trained to know how the input data is related to future tool performance and how the input data is related to the specific adjustment / recalibration needs of a specific tool.

[0038] Additionally, the machine learning model is configured to perform predictive maintenance modeling for different substrate processes (e.g., different deposition steps, different inspection steps) that may be performed by the same tool by considering input data that is specific to each tool type and specific to the functionality of each step (e.g., fabrication, inspection, measurement.) Specifically, the model is trained to know how the input data is related to future tool performance and how the input data is related to specific adjustment / recalibration needs for specific tools that perform specific fabrication steps and other substrate functions.

[0039] In addition to improved efficiency in tool maintenance, the present invention may also provide tool-to-tool matching. That is, for a given type of tool (e.g., a tool that performs a particular manufacturing step or other substrate function), use of the technology may ensure that all tools are and will continue to operate within a given predefined tolerance of each other, thereby improving manufacturing consistency between tools of the same type. Thus, a machine learning model may take into account current state performance data associated with one or more other tools to determine an appropriate maintenance plan for a given tool of the same type.

[0040] According to some examples, the machine learning model is a recurrent neural network (RNN) model. RNN models may be particularly suitable for the purposes of the present invention because RNN models are configured to take their previous output as part of their input along with additional inputs to generate the next output. As will be described in more detail herein, according to the present disclosure, the machine learning model is configured to provide a predicted future performance state of a given tool based on various inputs. These inputs include previously predicted future states of the tool generated by the model, as well as additional information that may not be available for previous model runs of the tool. Therefore, by using RNNs for tool management according to the present disclosure, for a given tool, each prediction may be based in part on one or more previous predictions of the tool taken before a previous run of the tool.

[0041] In some examples, the RNN model may be used in combination with one or more convolutional layers.

[0042] Figure 1 An example system 100 for managing performance of a semiconductor substrate tool according to the present disclosure is schematically depicted.

[0043] System 100 includes computing device 202. Computing device 202 may be a server and / or other computing device that performs the operations discussed herein (such as the tool management operations described herein). Computing device 202 may include computing component 206. Computing component 206 includes at least one processor 208 and memory 204. Memory 204 may include non-transitory computer-readable media. Depending on the exact configuration, memory 204 (which stores, among other things, tool management instructions and other instructions for performing other operations disclosed herein) may be volatile (such as RAM), non-volatile (such as ROM, flash memory, etc.), or some combination of the two. Computing device 202 may include one or more graphics processing units (GPUs) configured to accelerate model training and / or model prediction. In addition, computing device 202 may also include storage devices (removable 210 and / or non-removable 212), including but not limited to solid-state devices, magnetic or optical disks, or tapes. In addition, the computing device 202 may also have an input device 216 such as a touch screen, keyboard, mouse, pen, voice input, etc., and / or an output device 214 such as a display, speaker, printer, etc. One or more communication connections 218 such as a local area network (LAN), a wide area network (WAN), point-to-point, Bluetooth, RF, etc. may also be incorporated into the computing device 202.

[0044] The system 100 may include one or more monitoring and / or inspection devices 102 in operative communication with a computing device 202, such as being linked to the computing device 202 via a network and communication connection 218. Non-limiting examples of such monitoring and / or inspection devices 102 are provided in Figure 2 Reference Figure 2 The monitoring and / or inspection equipment 102 may include, for example, a substrate inspection equipment 110 and a tool sensor 112 .

[0045] A non-limiting example of substrate inspection equipment 110 is a reflectometer or spectrometer that measures the intensity of light or other waves (e.g., sound waves) reflected from a substrate at different wavelengths and generates spectral data from which a property of the substrate, such as the thickness of a layer, can be determined. The inspection data can be fed as input to a machine learning model running on computing device 202, which can be used in part by the model to determine the future state of the tool responsible for the measured property.

[0046] Non-limiting examples of tool sensors 112 may include temperature sensors, vacuum sensors, light intensity sensors, vibration sensors, ambient light sensors, humidity sensors, optics temperature sensors, fan speed sensors, pressure sensors, flow sensors, current sensors, voltage sensors, and the like. Data from such sensors may relate to specific tool parameters, such as the heat or light intensity of a tool's lamp, or the quality of a vacuum chamber generated by the tool. In some examples, data from such sensors may relate to environmental conditions, such as ambient temperature, ambient humidity, ambient light, ambient noise, and the like. Tool sensor data may be fed as input to a machine learning model running on computing device 202, which may be used, in part, by the model to determine future states of the tool as affected by the sensed conditions.

[0047] The system 100 includes one or more substrate tools 104. The substrate tools 104 may include manufacturing tools that perform different substrate manufacturing steps, such as, for example, deposition, lithography, oxidation, diffusion, ion implantation, CMP, and etching. Additionally or alternatively, the substrate tools 104 may include inspection and / or measurement (or metrology) tools. The substrate tools 104 may be linked to the computing device 202 via a network and communication connection 218. In this way, the substrate tools 104 may feed, for example, tool calibration data as input to a machine learning model running on the computing device 202, which may be used in part by the model to determine the future state of the tool.

[0048] The substrate tool 104 itself may also provide data related to the predicted future performance. Alternatively or additionally, such data for the substrate tool 104 may be obtained by one or more or other tools. Such data may include tool automatic test data, calibration data, and runtime data. For example, the substrate equipment may monitor or periodically run monitoring tests on various calibrations and other aspects of the substrate equipment, such as alignment of the tool's manufacturing stage, swing of the tool's manufacturing stage, intensity of the tool's light, repeatability of the tool's aperture (e.g., material deposition aperture or lens aperture), video focus of the tool, and the like. In some examples, the substrate tool 104 itself may feed this automatic monitoring and / or automatic test data as input to a machine learning model running on the computing device 202, which may be used in part by the model to determine the future state of the tool affected by the sensed conditions.

[0049] Runtime data is captured for each run of the tool and can therefore help identify and track small changes in tool performance and the exact time they occur. Examples of runtime data include alignment data and autofocus data for the tool for each use of the tool (e.g., each manufacturing run). In contrast, automatic test data and calibration data may be captured periodically, such as as part of a tool health check process. In some examples, the automatic test data and calibration data may include types of performance data that are not monitored or do not exist each time the tool is run and therefore would not be included in the runtime data. Examples of calibration data include data related to calibrations performed by the tool.

[0050] In an alternative configuration, one or more components of the computing device 202 reside locally on the one or more monitoring and / or inspection devices 102 or the substrate tool 104. For example, the one or more monitoring and / or inspection devices 102 or the one or more substrate tools 104 may be configured to perform one or more of the machine learning model tool management operations described herein by themselves. That is, the machine learning model tool management instructions may be run directly on the one or more monitoring and / or inspection devices 102 or the one or more substrate tools 104.

[0051] Figure 3 Describes the use of Figure 1 An example of system 100 using a machine learning model to manage the performance of a semiconductor substrate tool according to an example of the present disclosure.

[0052] In some examples, Figure 3 One or more of the operations may be performed by Figure 1 The process may be performed by the computing device 202 and / or one or more of the monitoring and / or inspection device 102 and / or the substrate tool 104.

[0053] refer to Figure 3 In general, input 302 is provided to a machine learning model 308. In some examples, machine learning model 308 is an RNN model. Machine learning model 308 is trained such that machine learning model 308 is configured to analyze input 302 to generate output 310, which can be used to generate output 310. Figure 1 The computing device 202 is provided to the technician (e.g., displayed on a display device). The input 302 may be provided by any of a number of different sources. Examples of such sources may include the substrate tool 104 or the monitoring and / or inspection equipment 102 ( Figure 1 ). Input may also be provided from data stored on the computing device 202.

[0054] The inputs 302 may include tool current state data 304 and additional data 306. The tool current state data 304 may include a previous predicted state of a given tool predicted by a machine learning model 308. For example, for a given deposition tool, modeling may be performed to predict the future state of the tool after each successive use of the tool (e.g., a substrate manufacturing run). For each modeling, the output is the predicted state of the tool after the next use of the tool (e.g., the next manufacturing run). For each modeling, one of the inputs 302 to the machine learning model 308 is the previous predicted state of the tool predicted by the machine learning model 308 before the previous use of the tool (e.g., the previous manufacturing run). The input reflects the current state of the tool. That is, the input reflects the current performance level or quality of the tool (within specification, out of specification, within specification but deviates from another tool of the same type, etc.).

[0055] In the specific example of a layer deposition tool, for a particular substrate layer, the input may reflect the thickness of the deposited layer deposited by the tool during a previous use of the tool (e.g., a previous manufacturing run). In the specific example of an inspection tool, for a particular substrate, the input may reflect whether a defect was detected and / or the nature and severity of the defect. In the specific example of a metrology tool, for a particular feature such as a transistor or a thin film, the input may reflect whether a critical dimension is within specification. In the specific example of a lithography tool, for a given substrate, the input may reflect the alignment of two layers of the substrate relative to each other. For many different types of substrate tools, the input may reflect the critical dimensions of one or more structural or functional features of the substrate, etc. Regardless of the substrate tool or the performance aspect of the substrate tool in question, the input data is fed to the machine learning model 308 so that the machine learning model 308 can use the current state of the tool as a baseline data set from which to predict the next future state of the tool (e.g., the state of the tool after the next use of the tool, such as after the next manufacturing run, or after the next metrology run, etc.).

[0056] The tool current state data 304 may also include current state performance data of other tools, based on which the machine learning model 308 can perform tool-to-tool performance matching of tools of the same type by comparing the predicted performance of a given tool with the current or predicted performance of another tool of the same type.

[0057] The input 302 may also include additional data 306. The additional data may include operational data of the tool.

[0058] refer to Figure 4 , the additional data 306 may include data from the monitoring and / or inspection equipment 102 and / or the substrate tool 104 itself ( Figure 1) obtained from the tool. In some examples, the current data or some of the current data is obtained from the most recent manufacturing operation of the tool. For example, the additional data 306 may include sensor data 322, including both tool-specific sensor data and environmental condition sensor data. The additional data 306 may also include automatic test data 320 generated by the tool itself. The additional data 306 may also include runtime data 326 obtained from each use of the substrate tool 104 (e.g., each manufacturing run). The additional data 306 may also include calibration data 328 (e.g., data related to calibration performed by the substrate tool 104).

[0059] refer to Figure 4 , the additional data 306 may also include tool event data 324. The tool event data 324 may be provided to the machine learning model 308 by the tool itself or via another device or from another apparatus. The event data 324 may include, for example, anomalies that have occurred with the tool in question during use of the tool (e.g., during a manufacturing run), such as an error occurring during a run or the tool owner removing the tool during or after a run. For example, if a tool is removed from manufacturing and recalibrated, the relevance of the tool current state data 304 in predicting the future state of the tool may be discounted by the machine learning model 308 based on how the model has been trained. Another example of an event may be the replacement of a part or component of a tool, such as replacement of a tool light, a tool lens, a tool stage, etc.

[0060] Machine learning model 308 may be a trained machine learning model that processes input 302 and generates output 310 based on input 302. Output 310 may be displayed via an interface of a computing device such as output device 214 ( Figure 1 ). For example, output 310 may be presented as a maintenance or management report for the tool indicating predicted maintenance of the tool. Output 310 may include one or more alerts or alarms. For example, an alert may be generated that the tool is predicted to fall out of specification at the next use of the tool, so that performance recalibration is necessary; or that the performance of the tool deviates from another tool of the same type by more than a predefined threshold size, so that tool-to-tool recalibration is necessary.

[0061] The output 310 may include a predicted future state of the tool 312. The predicted future state of the tool 312 may include a predicted performance metric for a given tool after the next manufacturing or other function of the tool. In the specific example of a layer deposition tool, for a specific substrate layer that the tool is responsible for manufacturing, the output may reflect a prediction of the thickness of the layer to be deposited by the tool when the tool is next used (e.g., at the next manufacturing run of the tool). Additionally or alternatively, the output may include an indication that the predicted performance of the tool for the next run will be within or outside of specification. Additionally or alternatively, the output may include an indication that the predicted performance of the tool for the next run will deviate from the performance of another tool of the same type by less than or more than a predefined maximum threshold.

[0062] The output 310 may include a predicted future performance state of the substrate tool. The output predicted future performance state may then be provided as an input to the trained machine learning model 308 as a subsequent performance state of the substrate tool. A subsequent predicted future performance state of the substrate tool after a subsequent future use of the substrate tool (e.g., a future manufacturing run) is determined by the trained machine learning model 308 based on the subsequent performance state of the substrate tool.

[0063] Output 310 may include tool diagnostics 313. Tool diagnostics 313 may indicate the reason why the predicted performance of the tool is out of specification or the predicted performance of the tool is greater than a minimum threshold deviation. Non-limiting example diagnostics may include a misaligned tool stage, too high tool optics temperature, a stage with too much wobble, too high or too low light intensity, too low fan speed, too high or too low ambient temperature, too high / too low variability in aperture repeatability, too high variability in video focus, etc. Thus, diagnostics 313 may relate to parameters of the tool itself, or to environmental conditions to which the tool is exposed.

[0064] Output 310 may include one or more remediation recommendations 314. For example, if it is determined that certain maintenance of the tool or the environmental conditions surrounding the tool is necessary based on the predicted tool future state 312 and tool diagnostics 313, machine learning model 308 is configured to generate a recommendation for that maintenance. For example, recommendation 314 may include a recommendation to lower or raise the ambient temperature or ambient light, replace a fan, adjust the alignment of the tool's stage by a specific amount, adjust the intensity of the tool's light by a specific amount, replace the tool's optics, tighten the tool's stage by a specific amount, etc.

[0065] Advantageously, because the machine learning model 308 determines diagnoses and diagnostic-specific remedial actions, the indicated maintenance is targeted and, therefore, can be performed more quickly and with less production disruption than, for example, periodic maintenance, in which many more aspects of the tool are checked and tested to see if adjustments or other recalibration are needed for each aspect.

[0066] Additionally, the machine learning model 308 is configured to identify when performance deviations or poor performance are not a result of tool parameters or calibration, but rather a result of the surrounding environment. For example, a remedial action determined by the machine learning model 308 may be to reduce the ambient temperature of the environment surrounding the tool by a specified number of degrees, which may be an easy and straightforward solution that does not require pulling the tool out of production for any length of time.

[0067] Figure 5 Describes the use of Figure 1 The system 100 of the present disclosure provides a method 400 for managing the performance of a semiconductor substrate tool according to an example of the present disclosure. It should be understood that different embodiments of the present disclosure may include different combinations of subsets of the steps of the method 400, non-limiting examples of which are described herein.

[0068] In step 402 of method 400, a machine learning model (e.g., machine learning model 308 ( Figure 3 )). For a given tool or tool type, performance data from continuous use (e.g., continuous manufacturing runs) and automated test data, sensor data, and event data associated with each manufacturing are processed. Based on this information, the model 308 learns how different factors (such as tool-specific factors, environmental factors, etc.) affect the performance of a given tool or tool type individually and collectively over time, resulting in a shift in performance over the continuous use of the tool (e.g., over continuous manufacturing runs). For example, the machine learning model 308 may learn that for a particular type of layer deposition tool, a particular amount of lamp intensity may cause the performance of the tool to change in a particular way (e.g., the deposited layer of the tool becomes thicker) depending on other tool and environmental factors.

[0069] In step 402, by processing the training data together with known tool performance data from multiple runs of the tool, the machine learning model 308 learns how to predict the future state of the tool, diagnose problems with the tool, and recommend remedial actions to resolve the problem. For example, the model learns to align the relevant data of the target tool with the corresponding data of the training tool with known results.

[0070] At step 404 of method 400, the machine learning model 308 obtains the current performance state of the tool. In some examples, the current performance state of the tool is the predicted future state 312 of the tool from a previous run of the model of the tool. Figure 3 ). In other examples, for example, if the tool is new or has been recalibrated since the last model run, a substrate inspection device (such as a metrology device) may be utilized to determine the current performance state of the tool. For example, for a layer deposition tool, the current performance state of the tool may be determined by a spectrometer that measures the thickness of a substrate deposited layer from the most recent (or first) run of the tool. However, the current performance state of the tool is determined, the data is fed to the model 308 at step 408.

[0071] In step 406 of method 400, additional data is obtained. The additional data may include the additional data 306 ( Figure 4 ). This data can be obtained from the substrate tool 104 and the monitoring and inspection equipment 102 ( Figure 1 ) is obtained and, in step 408, is fed to the machine learning model 308, as described above.

[0072] At step 410 of method 400, the machine learning model 308 generates a model output. The model output may include the output 310 ( Figure 3 As described above, the model output may be provided via any suitably configured output device.

[0073] Even after the initial training phase, the machine learning model 308 ( Figure 3 ) provide known performance data. For example, the output predictions of the model after running the method 400 can be empirically tested (e.g., using substrate inspection equipment), and the results of the testing can be input into the model as additional training data to improve the model's ability to predict the need for recalibration for a given tool.

[0074] As mentioned, the machine learning model 308 may be trained on a tool type by tool type basis. Calibration parameters, environmental factors, and event occurrences may affect the performance of different types of tools in different ways and to different degrees. Therefore, in step 402 ( Figure 5 ), machine learning model 308( Figure 3 ) can be trained individually for each tool. In addition, tool current status data 304 ( Figure 3 ) may include the type of tool. The machine learning model 308 uses the type of tool to select an appropriate modeling path to generate output 310 for the tool.

[0075] Additionally, the machine learning model 308 may be trained separately for each manufacturing step or other tool function. Calibration parameters, environmental factors, and event occurrences may affect the performance of different types of manufacturing steps and other tool functions in different ways and to different degrees. Thus, in step 402 ( Figure 5 ), the machine learning model 308 may be trained separately for each type of manufacturing step or other tool function ( Figure 3 ). In addition, tool current status data 304 ( Figure 3 ) may include a manufacturing step or tool function type. The machine learning model 308 uses the type of manufacturing step or tool function to select an appropriate modeling path to generate output 310 for the tool.

[0076] The embodiments described herein can be implemented and executed using software, hardware, or a combination of software and hardware. Although specific devices are described as performing specific functions throughout this disclosure, it will be appreciated by those skilled in the art that these devices are provided for illustrative purposes, and other devices can be used to perform the functions disclosed herein without departing from the scope of this disclosure. In addition, some aspects of this disclosure are described above with reference to the block diagram and / or operational diagrams of the systems and methods of various aspects of this disclosure. The functions, operations, and / or actions mentioned in the frame may not occur in the order shown in any corresponding flow chart. For example, according to the functions involved and the specific implementation, the two frames shown in succession can actually be implemented or executed substantially simultaneously or in reverse order.

[0077] The present disclosure describes some embodiments of the present technology with reference to the accompanying drawings, wherein only some embodiments in possible embodiments are shown. However, other aspects can be embodied in a variety of different forms, and should not be construed as being limited to the embodiments described herein. On the contrary, these embodiments are provided to make the present disclosure thorough and complete, and the scope of possible embodiments is fully conveyed to those skilled in the art. In addition, as used herein and in the claims, the phrase "at least one of element A, element B, or element C" is intended to convey any one of the following: element A, element B, element C, element A and B, element A and C, element B and C, and element A, B and C. In addition, those skilled in the art will understand that terms such as "about" or "substantially" convey the degree according to the measurement technology used herein. In the case where those skilled in the art may not be able to clearly define or understand such terms, the term "about" should mean plus or minus ten percent.

[0078] Although specific embodiments are described herein, the scope of the present technology is not limited to those specific embodiments. In addition, although different examples and embodiments can be described separately, such embodiments and examples can be combined with each other when implementing the technology described herein. Those skilled in the art will recognize other embodiments or improvements within the scope and essence of the present technology. Therefore, the specific structures, actions or media disclosed are only exemplary embodiments. The scope of the present technology is limited by the following claims and any equivalents thereof.

Claims

1. A computer-implemented method for determining a predicted future performance state of a substrate tool, comprising: providing a current performance status of the substrate tool to a trained machine learning model; providing the trained machine learning model with operational data of the substrate tool; as well as The predicted future performance state of the substrate tool is output, the predicted future performance state being determined by the trained machine learning model based on the current performance state and the operational data.

2. The computer-implemented method of claim 1 , further comprising outputting a recommended recalibration of the substrate tool determined by the trained machine learning model based on the current performance state.

3. The computer-implemented method of any one of claims 1 to 2, wherein the operational data comprises data generated by a sensor associated with the substrate tool or associated with a surrounding environment of the substrate tool.

4. The computer-implemented method of any one of claims 1 to 3, wherein the operational data comprises data generated by automated testing performed by the substrate tool.

5. The computer-implemented method of any one of claims 1 to 4, wherein the operational data comprises data generated by the occurrence of an error associated with the substrate tool.

6. The computer-implemented method of any one of claims 1 to 5, wherein the current performance status comprises a type of the substrate tool.

7. The computer-implemented method of any one of claims 1 to 6, wherein the current performance state comprises a type of manufacturing step or a type of another substrate function performed by the substrate tool.

8. A computer-implemented method according to any one of claims 1 to 7, wherein the trained machine learning model comprises a recurrent neural network.

9. The computer-implemented method of any one of claims 1 to 8, further comprising: outputting the predicted future performance state as an output predicted future performance state; providing the output predicted future performance state as input to the trained machine learning model as a subsequent performance state of the substrate tool; as well as A subsequent predicted future performance state of the substrate tool following a subsequent future use of the substrate tool is received, the subsequent predicted future performance state being determined by the trained machine learning model based on the subsequent performance state of the substrate tool.

10. A computer-implemented method according to any one of claims 1 to 9, wherein the current performance state includes a thickness of a substrate layer; and Wherein the predicted future performance state includes a predicted thickness of a substrate layer, the thickness being different from the predicted thickness.

11. The computer-implemented method of any one of claims 1 to 10, wherein the current performance state is determined by a substrate inspection tool.

12. A computer-implemented method according to any one of claims 1 to 11, Wherein the recommended recalibration includes recalibrating the identified parameters of the substrate tool prior to future use of the substrate tool.

13. The computer-implemented method of claim 12, wherein the identified parameter is light intensity.

14. The computer-implemented method of any one of claims 1 to 13, further comprising: A recommendation is received to adjust a condition of an ambient environment around the substrate tool, the recommendation generated by the trained machine learning model based on the current performance state and the operational data.

15. The computer-implemented method of any one of claims 1 to 14, wherein the predicted future performance state comprises an indication that performance of the substrate tool will be outside of a predefined performance specification for a future use of the substrate tool.

16. A computer implemented method according to any one of claims 1 to 15, wherein the predicted future performance state comprises an indication that performance of the substrate tool will deviate from performance of another tool in future uses of the tool by more than a predefined maximum deviation.

17. A method for predicting future performance states of a substrate tool, comprising: means for receiving, from a trained machine learning model, a current performance status of the substrate tool; means for receiving, by the trained machine learning model, operational data of the substrate tool; as well as Means for generating, by the trained machine learning model, a predicted future performance state of the substrate tool, the predicted future performance state being determined based on the current performance state and the operational data.

18. The method according to claim 17, further comprising: Means for generating, by the trained machine learning model, a recommended recalibration of the substrate tool based on the current performance state and the operational data.

19. A method according to any one of claims 17 to 18, wherein the operational data comprises data generated by one or more of: a sensor associated with the substrate tool or with the surrounding environment of the substrate tool; automated testing performed by said substrate tool; runtime data for the substrate tool, the runtime data comprising data associated with alignment or autofocus of the substrate tool; calibration performed by said substrate tool; and An event comprising replacement of a component of the substrate tool.

20. The method of any one of claims 17 to 19, wherein the operational data comprises data generated by the occurrence of an error associated with the substrate tool.

21. A system for determining a predicted future performance state of a substrate tool, comprising: one or more processors; and a non-transitory computer-readable storage device having stored thereon instructions that, when executed by the one or more processors, cause the system to: providing a current performance status of the substrate tool to a trained machine learning model; providing the trained machine learning model with operational data of the substrate tool; as well as The predicted future performance state of the substrate tool is output, the predicted future performance state being determined by the trained machine learning model based on the current performance state and the operational data.

22. The system of claim 21, wherein the operational data comprises data generated by the occurrence of an error associated with the substrate tool.

Images