Laser maintenance systems and processes

Abstract

A process or system of classifying the status of a laser system (302) in one or more categories includes a processor (370) that runs a sequence of operations: It is collecting data (361) in a data storage system (362). The high-dimensional data (363) runs through a transformer (364) to produce lower-dimensional data (365). This is used by a trained linear classifier (366) to generate classifications (367) that are used by a decision and communication system (368) to either initiate a correction or optimisation of the laser system (302), or to notify a technician. A preferred use is a forecast of a gas exchange in the laser system (302).

Description

LASER MAINTENANCE SYSTEMS AND PROCESSESCROSS-REFERENCE TO RELATED APPLICATIONS

[0001] This application claims priority to US Application No. 63 / 751,491, filed January 30, 2025, titled LASER MAINTENANCE SYSTEMS AND PROCESSES, which is incorporated herein by reference in its entirety.TECHNICAL FIELD

[0002] The disclosed subject matter relates to an apparatus for and methods of maintenance of laser systems and more specifically to systems and processes for automated identification and / or prediction of laser system health conditions such as the condition of requiring a discharge chamber gas refill.BACKGROUND

[0003] Laser systems are used, for example, as sources of radiation (light) in facilities that fabricate semiconductor devices, i.e., “fabs.” Laser systems, or laser-based light source systems, can be used to produce radiation for exposure of patterns in or on resists layers in photolithography processes, as well as for inspection and metrology, such as inspection and metrology of semiconductor substrates, patterns thereon, reticle properties, and any anomalies relating to these that may impact yield. Laser systems typically require maintenance actions from time to time in order to operate in accordance with specifications and avoid generating manufacturing process defects in lithography or metrology or inspection deficiencies in those applications. These maintenance actions include service gas refills in which the gas in a discharge chamber is replaced.

[0004] Sufficient notice of a requirement for gas replacement allows scheduling the gas refill for minimal disruption to production or other operations. Sufficient notice also allows time to prepare for efficient completion of the refill, minimizing total downtime. Providing notice of a requirement for gas refill sufficiently shortly before a failure or fault condition would have occurred avoids unnecessarily increasing the average downtime and wasting gas by refilling too often. Avoiding unexpected failures also protects equipment from damage or excessive wear and helps ensure workers’ safety. Accordingly, improvements in predicting a need or requirement for gas refill are desirable, as well as improvements in assessing other categories of the health of a laser or an apparatus or module thereof.SUMMARY

[0005] The following presents a succinct summary of one or more aspects, in order to provide an introductory understanding of the presently disclosed subject matter. This summary is not an extensive overview of all contemplated implementations and is not intended to single out any elements as being key or critical. Nor is it intended to delineate the full scope of any or allimplementations. Its sole purpose is to present some concepts of one or more aspects in a streamlined form as a prelude to the more detailed description that is presented later.

[0006] According to one aspect, a process is provided of preparing a classifier for use in classifying in one or more categories of a status of a laser system or a module or a component thereof, the process including: applying a dimensionality reduction module to a high -dimensional data set including variables that include features relating to characteristics and / or state of the laser system or the module or the component thereof, the high-dimensional data set including a dimension for every measured property of the system or the module or the component thereof, direct or derived, and producing thereby a lower-dimensional data set from the high-dimensional data set; including label data of the one or more categories in the lower-dimensional data set, resulting in a labeled lower-dimensional data set; training a linear classifier using the labeled lower-dimensional data set to produce a trained lower-dimensional linear classifier.

[0007] Implementations can include one or more of the following.

[0008] Incrementally updating the training of the trained lower-dimensional classifier. Incrementally updating the higher-dimensional data set and the lower-dimensional data set, and incrementally updating the training of the trained lower-dimensional classifier.

[0009] The one or more categories can be whether a gas refill is needed within a prespecified interval. The one or more categories can include one or more of presence of an anomaly; fault prediction within a prespecified time; need for optimization; and success of optimization.

[0010] The dimensionality reduction module can use a singular value decomposition (SVD) of a column-centered, scaled matrix of the high-dimensional data set. The dimensionality reduction module can use an eigen value decomposition of a covariance matrix of the high-dimensional data set. The process can further include whitening the high-dimensional data set prior to applying the dimensionality reduction module.

[0011] The dimensionality reduction module can produce from the high-dimensional data set multiple principal components and the process can further include limiting the lower-dimensional data set to principal components belonging to the minimum set of principal components sufficient to account for a pre-selected proportion of the variability of the high-dimensional data set. The preselected proportion can be within the range of 95% to 99.9%.

[0012] The dimensionality reduction module can produce from the high-dimensional data set multiple principal components and the process can further include selecting a number of principal components for use in the lower-dimensional data set by an automatic optimization process. The automatic optimization process can include limiting the lower-dimensional data set to a number of principal components as determined by a computer-executable algorithmic implementation of finding an elbow point in a plot of standard deviations of the principal components arranged in order from the greatest to (or toward) the least.

[0013] The linear classifier can include logistic regression. The linear classifier can include a support vector machine (SVM) with a linear kernel.

[0014] In additional aspects, a process is provided of classifying the health of a laser system in one or more categories, the process including: receiving a high-dimensional variable from the laser system; transforming the high-dimensional variable into a lower-dimensional variable corresponding to a lower-dimensional linear classifier trained on a lower-dimensional data set produced from a highdimensional data set by a dimensionality reduction module without manual feature selection; and classifying the health of the laser system in the one or more categories by applying the trained lowerdimensional linear classifier to the lower-dimensional variable.

[0015] Implementations can include one or more of the following.

[0016] The process can further include incrementally updating the training of the linear classifier. The process can further include incrementally updating the dimensionality reduction module, the lower-dimensional data set, and the training of the linear classifier

[0017] The one or more categories can include (or can be) whether a gas refill is needed within a prespecified interval. The one or more categories can include whether one or more optimization, tuning, and / or mitigation processes are needed.

[0018] The dimensionality reduction module can be configured to produce principal components from the high-dimensional data set and the lower-dimensional data set can contain only such principal components as belong to the minimum set of principal components sufficient to account for a preselected proportion of the variability of the high-dimensional data set. The pre-selected proportion can be within the range of 95% to 99.9%. The dimensionality reduction module can be configured to produce principal components from a high-dimensional data set and select a number of principal components for use in the lower-dimensional data set by an automatic optimization process. The automatic optimization process can include limiting the lower-dimensional data set to a number of principal components as determined by a computer-executable algorithmic implementation of finding an elbow point in a plot of standard deviations of the principal components in order from the greatest to (or toward) the least.

[0019] The linear classifier can include logistic regression. The linear classifier can include a support vector machine (SVM) with a linear kernel.

[0020] In yet additional aspects, a process is provided of classifying a chamber of a laser system as requiring gas replacement within a prespecified interval, the process including: receiving a highdimensional variable from the laser system; transforming the high-dimensional variable into a lowerdimensional variable corresponding to a lower-dimensional linear classifier trained on a lowerdimensional data set produced from a high -dimensional data set by a dimensionality reduction module without manual feature selection; and classifying the chamber of the laser system as requiring gas replacement within a prespecified interval by applying the trained lower-dimensional linear classifier to the lower-dimensional variable.

[0021] Implementations can include one or more of the following.

[0022] The process can further include incrementally updating the training of the linear classifier. The process can further include incrementally updating the dimensionality reduction module, the lower-dimensional data set, and the training of the linear classifier.

[0023] In more additional aspects, a computer-implemented process includes: receiving by a computing device a high-dimensional variable from a laser system; transforming by the computing device the high-dimensional variable into a lower-dimensional variable corresponding to a lowerdimensional linear classifier trained on a lower-dimensional data set produced from a highdimensional data set by a dimensionality reduction module without manual feature selection; and classifying by the computing device the health of the laser system in one or more categories by applying the trained lower-dimensional linear classifier to the lower-dimensional variable.

[0024] Implementations can include one or more of the following.

[0025] The process can further include incrementally updating the training of the linear classifier by the computing device. The one or more categories can include (or can be) whether a gas refill is needed within a prespecified interval. The one or more categories can include whether one or more optimization, tuning, mitigation, and / or maintenance or replacement processes are needed.

[0026] In still other aspects, a non-transitory computer-readable storage medium includes executable instructions for causing a processor to perform operations, with the instructions including instructions for performing the following operations: receiving a high-dimensional variable from a laser system; transforming the high-dimensional variable into a lower-dimensional variable corresponding to a lower-dimensional linear classifier trained on a lower-dimensional data set produced from a highdimensional data set by a dimensionality reduction module without manual feature selection; and classifying by the health of the laser system in one or more categories by applying the trained lowerdimensional linear classifier to the lower-dimensional variable.

[0027] Implementations can include one or more of the following.

[0028] The one or more categories can include (or can be) whether a gas refill is needed within a prespecified interval.

[0029] In still more aspects, a laser maintenance system includes: a data collection system configured to receive data from a laser and store the data as high -dimensional variables; a transformer configured to receive from the data collection system the high-dimensional variables and transform the highdimensional variables into lower-dimensional variables; a trained lower-dimensional linear classifier configured to receive the lower-dimensional variables from the transformer, the trained lowerdimensional linear classifier being trained, on a lower-dimensional data set produced from a highdimensional data set by a dimensionality reduction module without manual feature selection, to classify a health status of the laser system in one or more categories.

[0030] Implementations can include one or more of the following.

[0031] The trained lower-dimensional linear classifier can be an incrementally updated linearclassifier. The one or more categories can include (or can be) whether a gas refill is needed within a prespecified interval. The one or more categories can include whether one or more optimization, tuning, and / or mitigation processes are needed.

[0032] Further aspects, implementations, features, and advantages of the subject matter of the present disclosure, as well as the structure and operation of the various implementations, are described in detail below with reference to the accompanying drawings.DESCRIPTION OF DRAWINGS

[0033] The accompanying drawings, which are incorporated herein and form part of the specification, illustrate the presently disclosed subject matter and, together with the description, further serve to explain the principles of the presently disclosed subject matter and to enable a person skilled in the relevant art to make and use the presently disclosed subject matter.

[0034] FIG. 1 is a diagram of an exposure system, in accordance with some aspects of the present disclosure.

[0035] FIG. 2 is a diagram of a light source , in accordance with certain aspects of the present disclosure.

[0036] FIG. 3 is a functional block diagram of a system for maintenance of a laser system in accordance with one or more aspects of the present disclosure.

[0037] FIG. 4 is a functional block diagram of a process for training a classifier to be used in a maintenance system in accordance with one or more aspects of the present disclosure.

[0038] FIG. 5 is a functional diagram of a computer system , in accordance with some aspects or implementations of the present disclosure or portions thereof.

[0039] Further features and advantages of the presently disclosed subject matter, as well as the structure and operation of various implementations of the presently disclosed subject matter, are described in detail below with reference to the accompanying drawings. It is noted that the scope of this disclosure is not limited to the specific implementations explicitly described herein. Such implementations are included herein for illustrative purposes only. Additional implementations will be apparent to persons skilled in the relevant art based on the teachings presented herein.DESCRIPTION

[0040] Various implementations are now described with reference to the drawings, wherein like reference numerals are used to refer to like elements throughout. In the following description, for purposes of explanation, numerous specific details are set forth in order to promote a thorough understanding of one or more implementations. It may be evident after reviewing this disclosure that in some or all instances any implementation described below can be practiced without adopting the specific design details described below. In some instances, well-known structures and devices are shown in block diagram form in order to facilitate description of one or more implementations.

[0041] Systems such as those described herein may render benefits in a wide range of applications and implementations. For the sake of having a specific nonlimiting example to facilitate description, one such application is in semiconductor photolithography. Also, the following example is in terms of a system for producing radiation in the deep ultraviolet (DUV) portion of the electromagnetic spectrum that is, light having a wavelength in a range of about 100 nanometers (nm) to about 300 nm. It will be apparent, however, that the principles elucidated herein may also be applied to systems that produce radiation in other portions of the spectrum, for example, the extreme ultraviolet (EUV) portion of the spectrum, that is, having a wavelength in a range of about 5 nm to 20 nm, as well as particle beams, such as ion beams or electron beams.

[0042] Referring to FIG. 1, an exposure system 100 includes a system controller 118, an output apparatus or exposure apparatus controller 116, and a light source system 102 that produces a pulsed light beam 104. The pulsed light beam 104 is directed to an exposure apparatus 106 such as stepper / scanner and / or a metrology tool that measures and analyzes critical dimensions on a mask or a substrate, such as a wafer 110 with high precision. The wafer 110 is positioned over a table 112 constructed to hold the wafer 110, and the table 112 connected to a positioner 114 configured to accurately position the wafer 110 in accordance with certain parameters. The exposure apparatus 106 operates to pattern features and / or precisely measure features on the wafer 110 using the pulsed light beam 104. In the exposure system 100, the components within the laser system or light source system 102 (as discussed and shown in FIG. 2 below), in combination with the components of an optical system 108 of the exposure apparatus 106, will determine the parameters of the light beam 104 at the wafer 110, and thereby the parameters of the features patterned, and / or the capabilities of the inspections performed, on the wafer 110 by the exposure apparatus 106.

[0043] FIG. 2 is a functional block diagram of an example configuration of a light source system 202 that can be used as the light source system 102 of FIG. 1.

[0044] FIG. 2 shows a DUV light source 202 in the form of a DUV gas-discharge pulsed laser system. In some embodiments, the DUV light source 202 produces a pulsed beam 204 that can serve as the light beam 104 of FIG. 1. FIG. 2 shows a two-chamber laser system including a seed laser stage 220, such as a solid state or gas discharge master oscillator, a power amplification stage 240 such as a single-pass power amplifier (“PA”), a power ring amplifier (“PRA”), or a power oscillator (“PO”). Single chamber systems and systems with three or more chambers can also be used. The DUV light source 202 further includes relay optics 230, and an output subsystem 250.

[0045] The seed laser stage 220 may include, e.g., a master oscillator (“MO”) chamber 224 which includes a pair of electrodes 223 and 225. The seed laser stage 220 may also include a master oscillator output coupler (“OC”) 228, which may comprise a partially reflective mirror (not shown), that forms, together with a reflective grating (not shown) in a line narrowing module (“LNM”) 222, an oscillator (optical cavity) in which a beam oscillates to form a seed laser output pulse. The seed laser stage 220 may also include a first spectrum analysis module 226. The relay optics 230 mayinclude an MO wavefront engineering box (“WEB”) 232 that may serve to redirect the output of the seed laser stage 220 toward the power amplification stage 240, and may include, a multi prism beam expander (not shown) and an optical delay path (not shown).

[0046] The power amplification stage 240 may include, for example, a power amplifier discharge chamber 244. The power amplifier discharge chamber 244 may include a pair of electrodes 243 and 245. The power amplifier discharge chamber 244 may be part of an oscillator. The oscillator may be formed or defined by (1) seed-beam injection and output coupling optics (not shown) that may be incorporated into a PRA wavefront engineering box (“PRA WEB”) 248 and (2) a beam reverser (“BR”) 242. The PRA WEB 248 may incorporate a partially reflective input / output coupler (not shown) and a maximally reflective mirror for the nominal operating wavelength (e.g., at around 193 nm for an ArF system) and one or more prisms. The PRA WEB 248 passes a portion of the power it receives to the output subsystem 250 as a beam 249.

[0047] In the output subsystem 250, a second spectrum analysis module 252 may receive the light beam and pick off a portion of the light beam for metrology purposes, e.g., to measure the output bandwidth and pulse energy. The laser light beam 249 of pulses then passes to an optical pulse stretcher (“OPuS”) 254, and then to an autoshutter, both within the output subsystem 250. In the implementation of the output subsystem 250 shown in FIG. 2, the autoshutter is in the form of, or included within, a combined autoshutter metrology module (“CASMM”) 256, which may also include a pulse energy meter. Alternatively, a separate pulse energy meter (not shown) may be provided separately from, or in the absence of, the CASMM 256.

[0048] The power amplifier discharge chamber 244 and the MO discharge chamber 224 are configured as chambers in which electrical discharges between the electrodes create an inverted population of high energy molecules, including, e.g., Ar, Kr, F2, and / or Xe to produce a relatively broad-band light amplification potential. The wavelength(s) that are permitted to oscillate, and accordingly receive significant amplification, can be line -narrowed to a relatively very narrow bandwidth around a center wavelength selected by adjustments made in the LNM 222.

[0049] As mentioned, laser systems such as the laser system or light source 202 just described sometimes require maintenance. For example, lasing gas or gas mixtures in the MO discharge chamber 224 and / or the PA discharge chamber 244 age over time and use, and periodically require replacement. The chambers 224, and 244 themselves age over time and use and require periodic replacement. Performing maintenance such as gas replacement and chamber replacement, among other maintenance tasks, can result in a significant downtime penalty. The negative impacts of downtime for gas replacement can be minimized by accurately predicting in advance when a gas replacement is likely to be required within an upcoming interval of time or light source usage . Correct prediction of the need for such maintenance processes allows operators of a facility employing one or more lithography apparatuses or metrology apparatuses to schedule needed downtime in advance so as to minimize disruption to production processes. Correct prediction of the need for maintenance alsoprevents out-of-spec operation of the light source and the associated lithography apparatus, reducing or preventing scrapped product. Correct prediction of the need for maintenance also prevents in-process or in-service failures which in some cases can damage and / or shorten the life of light source system equipment. While FIG.2 illustrates a two-chamber laser system, the present disclosure encompasses other configurations, such as single -chamber systems.

[0050] In one or more aspects of the present disclosure, a maintenance system 360, shown in diagrammatic form in FIG. 3, includes a machine learning model implemented in a processor or processors 370 (dashed outline), such as a general-purpose computer or a specific dedicated processor, with both processing and storage capabilities, or at least with access to storage. The processor or processors 370 may be local to the laser system 302 or remote, and may communicate with the laser system 302 directly over wire or optical fiber links, for example, or wirelessly. The processor(s) 370 can be distributed in various real or virtual locations, such as in a cloud service or services.

[0051] The maintenance system 360 of FIG. 3, which illustrates one or more aspects of the present disclosure, uses data 361 of the laser system 302 to proactively determine when the laser system 302 is in a state in which a maintenance action is needed. In some embodiments, the laser system 302 includes at least one of a light source and an exposure apparatus (e.g., a scanner or an inspection tool). The laser system 302 can be in the form of the laser system 202 of FIG. 2, for example. The data 361 from the laser system 302 is received and stored in a data collection system 362. The data 361 as received from the laser system 302 is in the form of high-dimensional data, that is, in the form of a collection of variables (“vectors” or “columns”) each having a large number of features, including derived data. Derived data in this usage means data produced with or by calculation from, or combination of, raw or primary data such as measurements and sensor outputs, and the like. Derived data of the laser system can include, for example, fault signatures, performance metrics, statistical measures such as means, medians, modes, maximum values, minimum values and standard deviations of raw data. In some embodiments, the derived data is referred to as a pulse characterization unit for monitoring and analyzing pulse characteristics.

[0052] Alternatively, in some implementations the data storage system 362 may receive highdimensional data 361 in a less organized form and then organize it into the high-dimensional data including many variables with many features each, including derived features, and store it in such form. The data storage system 362 may also derive all or some of the derived data or derived features, based on the data 361 from the laser system 302.

[0053] As mentioned above, as the laser system 302 is continually sensed and monitored, data produced by the sensing and monitoring and the data derived therefrom can be organized into successive high-dimensional data vectors or columns (variables) each with many items or “features,” by the laser system 302 or the data collection system 362 or by both in combination. The features generally relate to the status or performance of the laser system as a whole and / or to the status orperformance of a particular lasing chamber or other apparatus or module within the laser system. The resulting high-dimensional variables 363 each having a large number of features are received by a transformer 364. The transformer 364 transforms the high-dimensional variables 363 into lowerdimensional variables (that is, projects them on to a reduced-dimension set of basis vectors) corresponding to a lower-dimensional trained linear classifier 366, and passes the lower-dimensional variables 365 to the linear classifier 366. In some embodiments, the linear classifier 366 includes at least one of logistic regression technique, linear SVM, a perceptron, naive Bayes classifier, and least squares classifier.

[0054] Based on one or more of the lower dimensional variable(s) 365, the lower-dimensioned trained linear classifier 366 classifies a health status of the laser system 302, such as whether a chamber or chambers of the laser system 302 require a gas replacement within a particular prespecified interval, for example. The interval may be a pre-specified number of laser pulses that a respective chamber participates in producing, a pre-specified time duration since a last service or refill of the respective chamber, or some combination of these, for example, or other interval measures.

[0055] One or more classifications 367 from the linear classifier 366 are received by a decision and communications system 368 that selects an appropriate response to the currently received classifications. In the case of a laser system health condition correctable or potentially correctable by an automatic process at the laser system 302, such as automatic processes that can take place during laser operation or during normal pauses in laser operation, a communication 369 can be sent to the laser system 302 to initiate a needed correction or optimization process. In the case of a more significant laser system health condition, such as a prediction for the requirement, within a predetermined interval, for a gas refill of a laser chamber or for a replacement of a laser chamber, a notification 371 can be sent to users such as service technicians or facility managers allowing for planning and scheduling of the needed maintenance. The decision and communications system 368 can be used to track performance of the laser system 302 over time, and / or to check for the success of optimization or correction processes, and / or in appropriate instances to follow a successive hierarchy of corrective actions to bring the laser system to within a desired performance range.

[0056] FIG. 4 is a diagram of an implementation of a process P400 for the preparation of a linear classifier 466 such as may be used as the linear classifier 366 of FIG. 3. A high-dimensional data set 472, having at least one of variables that include “raw” (as-measured or detected) and derived features relating to the characteristics and / or state of the laser system of interest and / or modules or apparatuses and components thereof, and / or of laser systems of the same general type and configuration, is processed by a dimensionality reduction module 474, such as a linear discriminant analysis (“LDA”) or a principal components analysis (“PCA”), without manual feature selection. In other words, the available data from the laser system of interest and / or of laser systems of the same general type and configuration is not pre-reduced or pre-selected before the dimensionality reduction module 474. That is, all available data is included, such that the high-dimensional data set includes a dimension formeasured property of the system (and / or similar systems) and / or modules or apparatuses and components thereof, direct or derived. Typical pre-processing such as interpolation or zeroing of missing values and centering and scaling of feature distributions is used as needed.

[0057] Projected features 475, such as discriminant components (“DCs”) or principal components (“PCs”) output from the dimensionality reduction module 474 are limited to a subset of the total or total possible projected features that is most representative of the variation of the high-dimensional data set 472 as a whole. For example, the lower-dimensional data set 476 produced by the dimensionality reduction module 474 can be limited to projected features belonging to a minimum set of the projected features sufficient to account for a preselected proportion of the variability of the high-dimensional data set 476. The preselected proportion can be 95%, 99%, or 99.9%, or any values or ranges encompassed by these, for example. Alternatively, the lower-dimensional data 476 set can be limited to a number of projected features or principal components determined by a computerexecutable algorithm for finding or approximating the finding of an elbow point in a plot of the standard deviations of projected features or principal components in order from the greatest to (or toward) the least. Such algorithms include but are not limited to those found in the FindPC package in R. Labels 473 for classification training can be added to (or remain in or be reinserted into) the lowerdimensional data 476 to form training data 477 supplied to train the linear classifier 466. The labels 473 may be automatically derived from aspects of the high-dimensional data set failures and / or out-of-range states and / or significant change states, as detected by existing or known failure signatures, employing a “look-back” from such states to include variables recorded within a preselected prediction interval (or respective intervals) preceding the respective state(s). Alternatively or in addition, expert classification may be employed to provide one or more desired labels if desired.

[0058] As may be appreciated from the foregoing and FIG. 4 together, an aspect of the current disclosure provides a process of preparing a classifier for use in classifying in one or more categories of the health of a laser system or a module or apparatus thereof, including: (1) performing a principal component analysis (PCA) of a high -dimensional data set comprising variables that include features relating to the performance and / or state of a laser system and / or modules or apparatuses and components thereof, and / or of laser systems of the same general type and configuration and / or modules and components thereof, with the high-dimensional data set including a dimension for every measured property of the system(s) and / or modules or apparatuses and components thereof, direct or derived, producing thereby a lower-dimensional data set from the high-dimensional data set; (2) including label data of the one or more categories in the lower-dimensional data set resulting in a labeled lower-dimensional data set; (3) training a linear classifier using the labeled lower-dimensional data set to produce a trained lower-dimensional linear classifier.

[0059] The linear classifier can be in the form of a logistic regression process or algorithm, or a support vector machine (SVM) with a linear kernel, or other suitable linear classifier, such as a perceptron, naive Bayes classifier, and least squares classifier. Use of a linear classifier allows forvery efficient training and classification processes. The overall health or performance of laser-based light source systems has generally not been believed to be a suitable application for linear classification, with existing laser health or performance classifiers generally employing nonlinear SVMs on the assumption of nonlinear behavior of the underlying system(s), but the methods described herein have shown increased classification accuracy (with respect to chamber or chamber gas failure prediction) and decreased variation relative to SVM with radial basis function (RBF) kernel. Use of linear classification also allows incrementally updating the training of the trained lower-dimensional classifier, such as incrementally updating the higher-dimensional data set and the resulting lower-dimensional data set and then incrementally updating the training of the trained lowerdimensional classifier. This can provide essentially continuous updating or optimization of the process and system for maintaining the laser based light source, eliminating delays produced during standard retraining processes and allowing the classifier(s) to reflect latest information.

[0060] The dimensionality reduction module process uses a matrix factorization framework. In some embodiments, the matrix factorization framework includes a direct extract decomposition technique, such as a singular value decomposition (SVD) of a column -centered, scaled matrix of the highdimensional data set to extract the top k singular values and their corresponding singular vectors for projecting the data onto a lower-dimensional space spanned by these singular vectors. In other embodiments, the direct extract decomposition technique includes an eigen value decomposition of a covariance matrix of the high-dimensional data set. In some embodiments, the matrix factorization framework includes an approximate decomposition technique, such as a power iteration by iteratively updating a vector to compute the principal eigenvector of a data matrix and projecting the data on to a lower-dimensional space spanned by such principal eigenvector data dimensionality reduction. For computational accuracy and efficiency, given that the high -dimensional data sets of laser systems or laser-based light sources tend toward sparseness, an SVD process optimized for efficient operation with sparse data can be used, for example, randomized SVD by generating a low -dimensional approximation using random projection. In some embodiments, whitening is used as part of and / or prior to the dimensionality reduction module. Use of efficient dimensionality reduction module processes together with linear classification results in a system capable of taking in and efficiently giving effect to all available data, possibly with the one exception of the highest frequency data sets such as data produced on a per-pulse basis. This data may be represented in the process by burst statistics data or samples thereof. Systems for obtaining burst statistics are disclosed, for example, in U.S. Patent Application Publication No. 2022 / 0365445 published November 17, 2022, and titled “Burst Statistics Data Aggregation Filter.” (Note that any patent applications, patents, and printed publications cited herein are incorporated herein by reference in their entireties, except for any definitions, subject matter disclaimers or disavowals, and except to the extent that the incorporated material is inconsistent with the express disclosure herein, in which case the language in this disclosure controls.) In addition (or in some implementations alternatively) to burst data or burststatistics, the available data includes properties measured or tracked for the light source or laser system, which can include one or more of (1) measurements of, for example, beam energy, beam wavelength, and beam bandwidth, (2) pulse repetition rate(s), (3) pulse counts, (4) chamber and / or gas age (by pulse and / or by time), (5) derived quantities such as rates of change in various measurements, failure or failure prediction metrics, optimization metrics, deviation from expected evolution of performance over time, (6) one or more statuses or states of the system, (7) metadata of the system such as software packages and versions, parts numbers and assemblies and ages, model numbers, plant ID, and (8) others.

[0061] The one or more categories on which the linear classifier is trained can include whether a gas refill is needed within a prespecified interval, and / or one or more of presence of an anomaly, fault prediction within a prespecified time; need for optimization; success of optimization; and others.

[0062] Classifier(s) described above with respect to FIGS. 3 (the linear classifier 366) and 4 (linear classifier 466) are used, as generally explained above with respect to FIG. 3, in a process of classifying the health of a laser system in one or more categories. The process includes (1) receiving a high-dimensional variable from the laser system, including at least one of a light source and an exposure apparatus; (2) transforming the high-dimensional variable into a lower-dimensional variable corresponding to a lower-dimensional linear classifier trained on a lower-dimensional data set produced from a high-dimensional data set by a dimensionality reduction module without manual feature selection; and (3) classifying the health of the laser system in the at least one aspect by applying the trained lower-dimensional linear classifier to the lower-dimensional variable. The process can further include incrementally updating the training of the linear classifier, such as by incrementally updating the high-dimensional data set, the dimensionality reduction module, the resulting lower-dimensional data set, and the training of the linear classifier. The one or more categories can be the category of whether a gas refill is needed within a prespecified interval, and or the additional category or categories of whether one or more optimization, tuning, and / or mitigation processes are needed.

[0063] In another aspect of the present disclosure, processes generally of the types described above can be implemented in a computer. For example, a computer-implemented process can include (1) receiving by a computing device a high-dimensional variable from a laser system; (2) transforming by the computing device the high-dimensional variable into a lower-dimensional variable corresponding to a lower-dimensional linear classifier trained on a lower-dimensional data set produced from a highdimensional data set by dimensionality reduction module without manual feature selection; and (3) classifying by the computing device the health of the laser system in one or more categories by applying the trained lower-dimensional linear classifier to the lower-dimensional variable. The process can further include incrementally updating the training of the linear classifier by the computing device. The one or more categories can include, or can be, whether a gas refill is needed within a prespecified interval. The one or more categories can additionally or alternatively includewhether one or more optimization, tuning, mitigation, and / or maintenance or replacement processes are needed.

[0064] In another aspect, the processes generally of the types described above can be implemented in a non-transitory computer-readable storage medium. For example, the non-transitory computer-readable storage medium can include executable instructions for causing a processor to perform operations, the instructions including instructions for performing the following operations: (1) receive a high-dimensional variable from a laser system; transform the high-dimensional variable into a lower-dimensional variable corresponding to a lower-dimensional linear classifier trained on a lowerdimensional data set produced from a high -dimensional data set by dimensionality reduction module without manual feature selection; and classify by the health of the laser system in one or more categories by applying the trained lower-dimensional linear classifier to the lower-dimensional variable. The executable instructions can further include instructions for incrementally updating the training of the linear classifier by the computing device. The one or more categories can include, or can be, whether a gas refill is needed within a prespecified interval. The one or more categories can additionally or alternatively include whether one or more optimization, tuning, mitigation, and / or maintenance or replacement processes are needed.

[0065] These and various other aspects of the processes and systems described herein, and / or features or parts thereof, can be implemented using one or more well-known computer systems, such as, for example, the computer system 1200 represented in the functional diagram of FIG. 5, which can be any well-known computer capable of performing the functions described herein.

[0066] The computer system 1200 of FIG. 5 is a non-limiting example of a computer system that may be used as, or as part of, the processor(s) 370 of FIG. 3. The computer system 1200 of FIG. 5 includes one or more processors (also called central processing units, or CPUs), such as a processor 1210. Processor 1210 is connected to a communication infrastructure or bus 1220.

[0067] One or more processors 1210 may each be a graphics processing unit (GPU). In an implementation, a GPU is a processor that is a specialized electronic circuit designed to process mathematically intensive applications. The GPU may have a parallel structure that is efficient for parallel processing of large blocks of data, such as mathematically intensive data common to computer graphics applications, images, videos, etc.

[0068] Computer system 1200 also includes user input / output device(s) 1230, such as monitors, keyboards, pointing devices, etc., that communicate with communication infrastructure 1220 through user input / output interface (s) 1240.

[0069] Computer system 1200 also includes a main or primary memory 1250, such as random-access memory (RAM). Main memory 1250 may include one or more levels of cache. Main memory 1250 has stored therein control logic (i.e., computer software) and / or data.

[0070] Computer system 1200 may also include one or more secondary storage devices or memory 1260. Secondary memory 1260 may include, for example, a hard disk drive 1280 and / or a removablestorage device or drive 1290. Removable storage drive 1290 may be a floppy disk drive, a magnetic tape drive, a compact disk drive, an optical storage device, tape backup device, and / or any other storage device / drive.

[0071] Removable storage drive 1290 may interact with a removable storage unit 1300. Removable storage unit 1300 includes a computer usable or readable storage device having stored thereon computer software (control logic) and / or data. Removable storage unit 1300 may be a floppy disk, magnetic tape, compact disk, DVD, optical storage disk, and / any other computer data storage device. Removable storage drive 1290 reads from and / or writes to removable storage unit 1300 in a well-known manner.

[0072] According to an example implementation, secondary memory 1260 may include other means, instrumentalities, or other approaches for allowing computer programs and / or other instructions and / or data to be accessed by computer system 1200. Such means, instrumentalities or other approaches may include, for example, a removable storage unit 1310. Examples of the removable storage unit 1310 may include a program cartridge and cartridge interface (such as that found in video game devices), a removable memory chip (such as an EPROM or PROM) and associated socket, a memory stick and USB port, a memory card and associated memory card slot, and / or any other removable storage unit and associated interface.

[0073] Computer system 1200 may further include a communication or network interface 1320. Communication interface 1320 enables computer system 1200 to communicate and interact with any combination of remote devices, remote networks, remote entities, etc. (individually and collectively referenced by reference number 1330). For example, communication interface 1320 may allow computer system 1200 to communicate with remote devices 1330 over communications path 1340, which may be wired and / or wireless, and which may include any combination of LANs, WANs, the Internet, etc. Control logic and / or data may be transmitted to and from computer system 1200 via communications path 1340.

[0074] In an implementation, a non-transitory, tangible apparatus or article of manufacture comprising a non-transitory, tangible computer useable or readable medium having control logic (software) stored thereon is also referred to herein as a computer program product or program storage device. This includes, but is not limited to, computer system 1200, main memory 1250, secondary memory 1260, and removable storage units 1290 and 1300, as well as tangible articles of manufacture embodying any combination of the foregoing. Such control logic, when executed by one or more data processing devices (such as computer system 1200), causes such data processing devices to operate as described herein.

[0075] Based on the teachings contained in this disclosure, it will be apparent to persons skilled in the relevant art(s) how to make and use implementations of this disclosure using data processing devices, computer systems and / or computer architectures other than that shown in FIG. 5. Inparticular, implementations may operate with software, hardware, and / or operating system implementations other than those described herein.

[0076] Although specific reference may have been made above to the use of implementations in the context of optical lithography, it will be appreciated that implementations may be used in other applications, for example imprint lithography, and where the context allows, is not limited to optical lithography.

[0077] It is to be understood that the phraseology or terminology herein is for the purpose of description and not of limitation, such that the terminology or phraseology of the present specification is to be interpreted by those skilled in relevant art(s) in light of the teachings herein.

[0078] It is to be appreciated that the Detailed Description section is intended to be used to interpret the claims. The Summary and Abstract sections may set forth one or more but not all exemplary contemplated implementations, and thus, are not intended to limit the implementations and the appended claims in any way.

[0079] The implementations have been described above with the aid of functional building blocks illustrating the implementation of specified functions and relationships thereof. The boundaries of these functional building blocks have been arbitrarily defined herein for the convenience of the description. Alternate boundaries can be defined so long as the specified functions are appropriately performed.

[0080] The foregoing description of the specific implementations will so fully reveal the general nature of the implementations that others can, by applying knowledge within the skill of the art, readily modify and / or adapt for various applications such specific implementations, without undue experimentation, without departing from the general concept of the implementations. Therefore, such adaptations and modifications are intended to be within the meaning and range of equivalents of the disclosed implementations, based on the teaching and guidance presented herein.

[0081] The above description includes examples of multiple implementations. It is, of course, not possible to describe every conceivable combination of components or methodologies for purposes of describing the these implementations, but one of ordinary skill in the art may recognize that many further combinations and permutations of various implementations are possible. Accordingly, the described implementations are intended to embrace all such alterations, modifications, and variations that fall within the spirit and scope of the appended claims. Furthermore, although elements of the described aspects and / or implementations may be described or claimed in the singular, the plural is contemplated unless limitation to the singular is explicitly stated. Additionally, all or a portion of any aspect and / or implementation may be utilized with all or a portion of any other aspect and / or implementation, unless stated otherwise.

[0082] The implementations can be further described using the following clauses.1. A process of preparing a classifier for use in classifying in one or more categories of a status of a laser system or a module or a component thereof, the process comprising: applying a dimensionalityreduction module to a high-dimensional data set including variables that include features relating to characteristics and / or state of the laser system or the module or the component thereof, the highdimensional data set including a dimension for every measured property of the system or the module or the component thereof, direct or derived, and producing thereby a lower-dimensional data set from the high-dimensional data set; including label data of the one or more categories in the lowerdimensional data set, resulting in a labeled lower-dimensional data set; training a linear classifier using the labeled lower-dimensional data set to produce a trained lower-dimensional linear classifier.2. The process of clause 1, further including incrementally updating the training of the trained lowerdimensional classifier.3. The process of clause 1, further including incrementally updating the higher-dimensional data set and the lower-dimensional data set, and incrementally updating the training of the trained lowerdimensional classifier.4. The process of clause 1, wherein the one or more categories is whether a gas refill is needed within a prespecified interval.5. The process of clause 1, wherein the one or more categories includes one or more of presence of an anomaly; fault prediction within a prespecified time; need for optimization; and success of optimization.6. The process of clause 1, wherein the dimensionality reduction module uses a singular value decomposition (SVD) of a column -centered, scaled matrix of the high-dimensional data set.7. The process of clause 1, wherein the dimensionality reduction module uses an eigen value decomposition of a covariance matrix of the high -dimensional data set.8. The process of clause 1, further including whitening the high -dimensional data set prior to applying the dimensionality reduction module.9. The process of clause 1, wherein the dimensionality reduction module produces from the highdimensional data set multiple principal components and further including limiting the lowerdimensional data set to principal components belonging to the minimum set of principal components sufficient to account for a pre-selected proportion of the variability of the high-dimensional data set.10. The process of clause 9, wherein the preselected proportion is within the range of 95% to 99.9%.11. The process of clause 1, wherein the dimensionality reduction module produces from the highdimensional data set multiple principal components and further including selecting a number of principal components for use in the lower-dimensional data set by an automatic optimization process.12. The process of clause 11, wherein the automatic optimization process includes limiting the lowerdimensional data set to a number of principal components as determined by a computer-executable algorithmic implementation of finding an elbow point in a plot of standard deviations of the principal components arranged in order from the greatest to (or toward) the least.13. The process of clause 1, wherein the dimensionality reduction module produces from the highdimensional data set multiple principal components and further including limiting the lower-dimensional data set to a number of principal components as determined by a computer-executable algorithmic implementation of finding an elbow point in a plot of standard deviations of the principal components arranged in order from the greatest to (or toward) the least.14. The process of clause 1, wherein the linear classifier includes logistic regression.15. The process of clause 1, wherein the linear classifier includes a support vector machine (SVM) with a linear kernel.16. A process of classifying the health of a laser system in one or more categories, the process comprising: receiving a high-dimensional variable from the laser system; transforming the highdimensional variable into a lower-dimensional variable corresponding to a lower-dimensional linear classifier trained on a lower-dimensional data set produced from a high-dimensional data set by a dimensionality reduction module without manual feature selection; and classifying the health of the laser system in the one or more categories by applying the trained lower-dimensional linear classifier to the lower-dimensional variable.17. The process of clause 16, further including incrementally updating the training of the linear classifier.18. The process of clause 16, further including incrementally updating the dimensionality reduction module, the lower-dimensional data set, and the training of the linear classifier.19. The process of clause 16, wherein the one or more categories includes whether a gas refill is needed within a prespecified interval.20. The process of clause 16, wherein the one or more categories includes whether one or more optimization, tuning, and / or mitigation processes are needed.21. The process of clause 16, wherein the dimensionality reduction module produces principal components from the high data set and the lower-dimensional data set contains only principal components belonging to the minimum set of principal components sufficient to account for a preselected proportion of the variability of the high-dimensional data set.22. The process of clause 21, wherein the preselected proportion is within the range of 95% to 99.9%.23. The process of clause 16, wherein the dimensionality reduction module produces principal components from the high data set and further including selecting a number of principal components for use in the lower-dimensional data set by an automatic optimization process.24. The process of clause 23, wherein the automatic optimization process includes limiting the lowerdimensional data set to a number of principal components as determined by a computer-executable algorithmic implementation of finding an elbow point in a plot of standard deviations of the principal components, in order from the greatest to (or toward) the least.25. The process of clause 16, wherein the linear classifier includes logistic regression.26. The process of clause 16, wherein the linear classifier includes a support vector machine (SVM) with a linear kernel.27. A process of classifying a chamber of a laser system as requiring gas replacement within a prespecified interval, the process comprising: receiving a high-dimensional variable from the laser system; transforming the high-dimensional variable into a lower-dimensional variable corresponding to a lower-dimensional linear classifier trained on a lower-dimensional data set produced from a highdimensional data set by a dimensionality reduction module without manual feature selection; and classifying the chamber of the laser system as requiring gas replacement within a prespecified interval by applying the trained lower-dimensional linear classifier to the lower-dimensional variable.28. The process of clause 27, further including incrementally updating the training of the linear classifier.29. The process of clause 27, further including incrementally updating the dimensionality reduction module, the lower-dimensional data set, and the training of the linear classifier.30. A computer-implemented process including: receiving by a computing device a high-dimensional variable from a laser system; transforming by the computing device the high-dimensional variable into a lower-dimensional variable corresponding to a lower-dimensional linear classifier trained on a lower-dimensional data set produced from a high-dimensional data set by a dimensionality reduction module without manual feature selection; and classifying by the computing device the health of the laser system in one or more categories by applying the trained lower-dimensional linear classifier to the lower-dimensional variable.31. The computer-implemented process of clause 30, further including incrementally updating the training of the linear classifier by the computing device.32. The computer-implemented process of clause 30, wherein the one or more categories includes whether a gas refill is needed within a prespecified interval.33. The computer-implemented process of clause 30, wherein the one or more categories includes whether one or more optimization, tuning, mitigation, and / or maintenance or replacement processes are needed.34. Anon-transitory computer-readable storage medium including executable instructions for causing a processor to perform operations, the instructions including instructions for performing the following operations: receiving a high-dimensional variable from a laser system; transforming the highdimensional variable into a lower-dimensional variable corresponding to a lower-dimensional linear classifier trained on a lower-dimensional data set produced from a high-dimensional data set by a dimensionality reduction module without manual feature selection; and classifying by the health of the laser system in one or more categories by applying the trained lower-dimensional linear classifier to the lower-dimensional variable.35. The non-transitory computer-readable storage medium of clause 34, wherein the one or more categories includes whether a gas refill is needed within a prespecified interval.36. A laser maintenance system comprising: a data collection system configured to receive data from a laser and store the data as high-dimensional variables; a transformer configured to receive from thedata collection system the high-dimensional variables and transform the high-dimensional variables into lower-dimensional variables; a trained lower-dimensional linear classifier configured to receive the lower-dimensional variables from the transformer, the trained lower-dimensional linear classifier being trained, on a lower-dimensional data set produced from a high-dimensional data set by a dimensionality reduction module without manual feature selection, to classify a health status of the laser system in one or more categories.37. The system of clause 36, wherein the trained lower-dimensional linear classifier is an incrementally updated linear classifier.38. The process of clause 36, wherein the one or more categories includes whether a gas refill is needed within a prespecified interval.39. The process of clause 36, wherein the one or more categories includes whether one or more optimization, tuning, and / or mitigation processes are needed.

[0083] The above-described implementations and other implementations are within the scope of the following claims.

Claims

CLAIMS1. A process of preparing a classifier for use in classifying in one or more categories of a status of a laser system or a module or a component thereof, the process comprising:applying a dimensionality reduction module to a high-dimensional data set comprising variables that include features relating to characteristics and / or state of the laser system or the module or the component thereof, the high-dimensional data set including a dimension for measured property of the system or the module or the component thereof, direct or derived, producing thereby a lowerdimensional data set from the high-dimensional data set;including label data of the one or more categories in the lower-dimensional data set resulting in a labeled lower-dimensional data set;training a linear classifier using the labeled lower-dimensional data set to produce a trained lower-dimensional linear classifier.

2. The process of claim 1, further comprising incrementally updating the training of the trained lower-dimensional classifier.

3. The process of claim 1, further comprising incrementally updating the higher-dimensional data set and the lower-dimensional data set, and incrementally updating the training of the trained lowerdimensional classifier.

4. The process of claim 1, wherein the one or more categories is whether a gas refill is needed within a prespecified interval.

5. The process of claim 1, wherein the one or more categories includes one or more of presence of an anomaly, fault prediction within a prespecified time; need for optimization; and success of optimization.

6. The process of claim 1, wherein the dimensionality reduction module uses a singular value decomposition (SVD) of a column -centered, scaled matrix of the high -dimensional data set.

7. The process of claim 1, wherein the dimensionality reduction module uses an eigen value decomposition of a covariance matrix of the high -dimensional data set.

8. The process of claim 1, further comprising whitening the high -dimensional data set prior to applying the dimensionality reduction module.

9. The process of claim 1, wherein the dimensionality reduction module produces from the highdimensional data set multiple principal components and further comprising limiting the lowerdimensional data set to principal components belonging to the minimum set of principal components sufficient to account for a preselected proportion of the variability of the high -dimensional data set.

10. The process of claim 1, wherein the dimensionality reduction module produces from the highdimensional data set multiple principal components and further comprising selecting a number of principal components for use in the lower-dimensional data set by an automatic optimization process.

11. The process of claim 10, wherein the automatic optimization process comprises limiting the lower-dimensional data set to a number of principal components as determined by a computerexecutable algorithmic implementation of finding an elbow point in a plot of standard deviations of the principal components arranged in order from the greatest to (or toward) the least.

12. The process of claim 1, wherein the dimensionality reduction module produces from the highdimensional data set multiple principal components and further comprising limiting the lowerdimensional data set to a number of principal components as determined by a computer-executable algorithmic implementation of finding an elbow point in a plot of standard deviations of the principal components arranged in order from the greatest to (or toward) the least.

13. The process of claim 1, wherein the linear classifier comprises logistic regression or a support vector machine (SVM) with a linear kernel.

14. A process of classifying a chamber of a laser system as requiring gas replacement within a prespecified interval, the process comprising:receiving a high-dimensional variable from the laser system;transforming the high-dimensional variable into a lower-dimensional variable corresponding to a lower-dimensional linear classifier trained on a lower-dimensional data set produced from a high-dimensional data set by a dimensionality reduction module without manual feature selection; and classifying the chamber of the laser system as requiring gas replacement within a prespecified interval by applying the trained lower-dimensional linear classifier to the lower-dimensional variable.

15. The process of claim 14, further comprising incrementally updating the training of the linear classifier.

16. The process of claim 14, further comprising incrementally updating the dimensionality reduction module, the lower-dimensional data set, and the training of the linear classifier.

17. A laser maintenance system comprising:a data collection system configured to receive data from a laser and store the data as highdimensional variables;a transformer configured to receive from the data collection system the high-dimensional variables and transform the high-dimensional variables into lower-dimensional variables;a trained lower-dimensional linear classifier configured to receive the lower-dimensional variables from the transformer, the trained lower-dimensional linear classifier being trained, on a lower-dimensional data set produced from a high-dimensional data set by a dimensionality reduction module without manual feature selection, to classify a health status of the laser system in one or more categories.

18. The system of claim 17, wherein the trained lower-dimensional linear classifier is an incrementally updated linear classifier.

19. The process of claim 17, wherein the one or more categories comprises whether a gas refill is needed within a prespecified interval.

20. The process of claim 17, wherein the one or more categories comprises whether one or more optimization, tuning, and / or mitigation processes are needed.

Images