System for and method of laser radiation source maintenance

Abstract

A system for and method of using performance data to generate natural language maintenance recommendations such as troubleshooting in which a machine learning model generates scores based on the performance data and a translator which may use fuzzy logic transposes the scores into natural language queries which a natural language processor uses to generate the natural language maintenance recommendations.

Description

SYSTEM FOR AND METHOD OF LASER RADIATION SOURCE MAINTENANCECROSS-REFERENCE TO RELATED APPLICATIONS

[0001] This application claims priority to US Application No. 63 / 778,006, filed March 26, 2025, titled SYSTEM FOR AND METHOD OF LASER RADIATION SOURCE MAINTENANCE, which is incorporated herein by reference in its entirety.FIELD

[0002] The present disclosure relates to a system and method for maintenance of a laser radiation source, for example, a deep ultraviolet radiation source.BACKGROUND

[0003] Photolithography is one of the processes by which semiconductor circuitry is patterned on a substrate such as a silicon wafer. An optical source generates deep ultraviolet (DUV) radiation used to expose a photoresist on the wafer. DUV radiation may include wavelengths from, for example, about 100 nanometers (nm) to about 400 nm. Often, the optical source is a laser radiation source (for example, an excimer laser) and the DUV radiation is a pulsed laser beam. The DUV radiation from the optical source interacts with a projection optical system, which projects the beam through a mask onto the photoresist on the silicon wafer. In this way, a layer of chip design is patterned onto the photoresist. The photoresist and wafer are subsequently etched and cleaned, and then the photolithography process repeats as necessary.

[0004] The laser radiation source is typically required to reliably produce an extremely large number of pulses over its operational lifetime. Satisfying this requirement in turn necessitates sophisticated maintenance of the laser radiation source. In particular, predictive maintenance methods are used to evaluate the health of the laser radiation source and anticipate maintenance actions which may need to be performed proactively to minimize downtime. These maintenance actions may include replacement of modules that have a significant likelihood of failure within a certain period of time based on performance parameters such as the age of the module, for example, in terms of the number of bursts of pulses the laser radiation source has produced (shot count). These performance parameters may be used to produce a numerical score that is indicative of the health of the laser radiation source. It may, however, be difficult for technicians performing maintenance to interpret the significance of these numerical scores in terms of understanding their significance and what maintenance actions should be undertaken as a result.

[0005] It is in this context in which the need for the presently disclosed subject matter arises.SUMMARY

[0006] Tire following presents a succinct summary of one or more embodiments in order to providea basic understanding of the disclosed subject matter. This summary is not an extensive overview of all contemplated embodiments. It is not intended to identify any elements of embodiments as being key or critical elements nor delineate the scope of any or ail embodiments. Its sole purpose is to present some concepts relating to one or more embodiments in a concise form as a prelude to the more detailed description that is presented later.

[0007] According to an aspect of an embodiment there is disclosed a maintenance system for a laser radiation source, the system comprising a machine learning model arranged to receive data from the laser radiation source relating to one or more parameters of the laser radiation source, the machine learning model being adapted to generate one or more numerical scores based at least in part on the data, a translator arranged to receive the numerical scores and adapted to generate a natural language expression indicative of values of tire numerical scores, and a natural language processor arranged to receive the natural language expression indicative of values of the numerical scores and adapted to generate a natural language expression of one or more maintenance actions based at least in part on the natural language expression indicative of values of the numerical scores.

[0008] The parameters may include at least one of diagnostics, configuration, and performance of a module of the laser radiation source. The maintenance action may include a troubleshooting plan for a module of the laser radiation source. The machine learning model may be an ensemble model. The ensemble model may be a random forest model.

[0009] The translator may be configured to employ fuzzy logic. The natural language processor may be configured as a large language model virtual assistant. The natural language expression may be a query for the virtual assistant. The numerical score may comprise a predictive score for a process module of the laser radiation source and the predictive score may indicate a likelihood of failure of the process module.

[0010] According to another aspect of an embodiment there is disclosed a maintenance method for a laser radiation source, the method comprising a step performed by the laser radiation source of generating performance data, supplying the performance data to a machine learning model, a step performed by the machine learning model of generating one or more numerical scores based at least in part on the performance data, translating the numerical scores into a natural language expression indicative of values of the numerical scores, and generating a natural language expression of one or more maintenance actions based at least in part on the natural language expression indicative of values of the numerical scores.

[0011] The machine learning model may be an ensemble model. The ensemble model may be a random forest model. The step of translating the numerical scores into a natural language expression may be performed using fuzzy logic. The step of generating a natural language expression of one or more maintenance actions may be performed by a natural language processor. Ihe natural language processor may be configured as a large language model virtual assistant. The natural language expression may be a query for the virtual assistant. The maintenance actions may be troubleshootingrecommendations. The numerical score may comprise a predictive score for a process module of the laser radiation source and wherein the predictive score indicates a likelihood of failure of the process module.

[0012] According to another aspect of an embodiment there is disclosed a method for decision - making in maintenance of a process module, the method comprising collecting operational data of the process module, generating a predictive score for the process module using a machine learning model, wherein tire predictive score indicates a likelihood of failure of the process module, applying fuzzy logic to the predictive score and the operational data to generate generalized descriptions of a status of the process module, and using a language model to process the generalized descriptions and provide action items for maintenance.

[0013] The predictive score may be grouped into predefined categories corresponding to the action items. Applying fuzzy logic may include assigning grades to individual features of the operating data, enabling interpretability of the predictive score. Hie method may further comprise uploading second operational data of the process module to the language model.

[0014] According to another aspect of an embodiment there is disclosed a system for maintenance of a laser module, the system comprising a data collection unit configured to gather operational data of the laser module, a machine learning module configured to generate a predictive score indicating a likelihood of failure of the laser module, a fuzzy logic engine configured to process the predictive score and the operational data to produce generalized text descriptions, and a language model interface configured to process the generalized text description and provide action items for maintenance. The system may further comprise a prompt generation module configured to create self-awareness prompts based on the generalized text description and predefined rules for the language model interface.

[0015] Further features and exemplary aspects of the embodiments, as well as the structure and operation of various embodiments, are described in detail below with reference to the accompanying drawings. It is noted that the scope of all possible embodiments is not limited to the specific embodiments described herein. Such specific embodiments are presented herein for illustrative purposes only. Additional embodiments will be apparent to persons skilled in the relevant art(s) based on the teachings contained herein.DRAWING DESCRIPTION[00161 The accompanying drawings, which are incorporated herein and form part of the specification, illustrate the embodiments and, together with the description, further serve to explain the principles of the embodiments and to enable a person skilled in the relevant art(s) to make and use the embodiments. The figures are not to scale unless otherwise indicated.

[0017] FIG. 1 is a functional block diagram of an overall broad conception of a photolithography system according to an aspect of the disclosed subject matter.

[0018] FIG. 2 is a schematic diagram of an overall broad conception of an illumination systemaccording to an aspect of the disclosed subject matter.

[0019] FIG. 3 is a graphical diagram illustrating certain principles of fuzzy logic according to an aspect of the disclosed subject matter.

[0020] FIG. 4 is a functional block diagram of a maintenance system according to an aspect of the disclosed subject matter.

[0021] FIG. 5 is a functional block diagram of a maintenance system according to an aspect of the disclosed subject matter.

[0022] FIG. 6 is a flow chart of a maintenance method according to an aspect, of the disclosed subject matter.DETAILED DESCRIPTION

[0023] Before describing specific embodiments in more detail, it is helpful to present an example environment in which embodiments of the present subject matter may be implemented. Referring to FIG. 1, a photolithography system 100 includes an illumination system 105. As described more fully below, the illumination system 105 includes a light source that produces a pulsed light beam 110 and directs it to an exposure apparatus, such as a stepper or a scanner 115 that patterns microelectronic features on a wafer 120. The wafer 120 is placed on a wafer table 125 constructed to hold wafer 120 and connected to a positioner configured to position the wafer 120 accurately in accordance with certain parameters.

[0024] The photolithography system 100 uses a light beam 110 having a wavelength in the DUV range with a wavelength in a range of about 100 nm to about 400 nm. The minimum size of the microelectronic features that can be patterned on the wafer 120 depends on the wavelength of the light beam 110, with a lower wavelength resulting in a smaller minimum feature size. When the wavelength of the light beam 110 is about 248 nm or about 193 nm, the minimum size of the microelectronic features can be, for example, 50 nm or less. The bandwidth of the light beam 110 can be the actual, instantaneous bandwidth of its optical spectrum (or emission spectrum), which contains information on how the optical energy of the light beam 110 is distributed over different wavelengths.

[0025] The scanner 11 includes an optical arrangement having, for example, one or more condenser lenses, a mask, and an objective arrangement. The mask is movable along one or more directions, such as along an optical axis of the light beam 110 or in a plane that is perpendicular to the optical axis. The objective arrangement includes a projection lens and enables the image transfer to occur from the mask to the photoresist on the wafer 120. The illumination system 105 adjusts the range of angles for the light beam 110 impinging on the mask. The illumination system 105 also homogenizes (makes uniform) the intensity distribution of the light beam 110 across the mask.

[0026] Tire scanner 115 can include, among other features, a lithography controller 130, air conditioning devices, and power supplies for the various electrical components. The lithography controller 130 controls how layers are printed on the wafer 120. The lithography controller 130includes a memory' that stores information such as process recipes. A process program or recipe determines the length of the exposure on the wafer 120 based on, for example, the mask used, as well as other factors that affect the exposure. During lithography, one or more bursts of pulses of the light beam 110 illuminate the same area of the wafer 120 to constitute an illumination dose.

[0027] In some embodiments, the photolithography system 100 also includes a control system 135. In general, the control system 135 includes one or more of digital electronic circuitry, computer hardware, firmware, and software. The control system 135 also includes memory which can be read¬ only memory’ and / or random access memory. Storage devices suitable for tangibly embodying computer program instructions and data include all forms of non-volatile memory', including, by way of example, semiconductor memory devices, such as EPROM, EEPROM, and flash memory devices; magnetic disks such as internal hard disks and removable disks; magneto -optical disks; and CD-ROM disks,

[0028] The control system 135 can also include one or more input devices (such as a keyboard, touch screen, microphone, mouse, hand-held input device, etc.) and one or more output devices (such as a speaker or a monitor). The control system 135 also includes one or more programmable processors, and one or more computer program products tangibly embodied in a machine-readable storage device for execution by one or more programmable processors. The one or more programmable processors can each execute a program of instructions to perform desired functions by operating on input data and generating appropriate output. Generally, the processors receive instructions and data from the memory. Any of the foregoing may be supplemented by, or incorporated in, specially designed ASICs (application-specific integrated circuits). The control system 135 can be centralized or be partially or wholly distributed throughout the photolithography system 100.

[0029] In some embodiments, the present subj ect matter is implemented m an inspection or metrology system. During a semiconductor manufacturing process, specifically’ during various stages such as lithography, etching, deposition, and chemical mechanical polishing, the inspection or the metrology system can use DUV radiation to scan the surface of the wafer for pattern defects or any other anomalies that may impact yield. In addition, for reticle inspection, the inspection or metrology apparatus examines the reticles for defects such as missing features, extra features, or contamination, which could be transferred to the wafer during the lithography process. The inspection or metrology system may additionally perform overlay measurement, ensuring that different layers of the devices are aligned with each other during the manufacturing process. Moreover, the inspection or metrology apparatus may offer in-line process control by providing real-time feedback, allowing users to adjust process parameters on the fly. In at least one embodiment, such an inspection system and / or such a metrology system is integrated with the photolithography system 100.

[0030] FIG. 2 shows a gas discharge pulsed laser source as an example of an illumination system 105. The gas discharge pulsed laser source may include, e.g., a solid state or gas discharge seed laser system 140 including a master oscillator (‘‘MO”) chamber 165. The gas discharge pulsed laser sourcemay also include an energy amplification system 145 including an energy amplification lasing chamber 200 which may be, for example, a single-pass power amplifier (‘TA”), a power oscillator (“PO”), or a power ring amplifier (“PRA”). The gas discharge pulsed laser source may also include an energy amplification system 145, relay optics 150, and a laser system output subsystem 160.

[0031] In some embodiments, the seed laser system 140 may include a master oscillator output coupler ("‘MO OC”) 175, which may comprise a partially reflective mirror, forming with a reflective grating (not shown) in a line narrowing module (“LNM”) 170, an oscillator cavity in which the seed laser system 140 oscillates to form the seed laser output pulse, i.e., forming the master oscillator MO. The system may also include a line-center analysis module (“LAM”) 180. The LAM 180 includes an etalon spectrometer for fine wavelength measurement and a coarser resolution grating spectrometer. A first wavefront engineering box (“WEB”) 185 may serve to redirect tire output of the seed laser system 140 toward the energy amplification system 145, and may include, e.g., beam expansion with, e.g., a multi prism beam expander (not shown) and coherence busting, e.g., in the form of an optical delay path (not shown).

[0032] The energy amplification lasing chamber 200 forms a pulsed beam by seed beam injection and output coupling optics (not shown) that may be incorporated into a second WEB 210. A beam reverser 220 redirects the pulsed beam back through the gain medium in the energy amplification lasing chamber 200. The second WEB 210 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.

[0033] A bandwidth analysis module (“BAM”) 230 at the output of the energy amplification system 145 may receive the output laser light beam of pulses from the energy amplification system 145 and pick off a portion of the light beam for metrology purposes, e.g., to measure beam characteristics such as output bandwidth and pulse energy. The output pulsed beam then passes through an optical pulse stretcher (“OPuS”) 240 and an output combined autoshuter metrology module (“CASMM”) 250, which may also be the location of a pulse energy meter. One purpose of the OPuS 240 may be, e.g., to convert a single output laser pulse into a pulse train. Secondary pulses created from the original single output pulse may be delayed with respect to each other. By distributing the original laser pulse energy into a train of secondary pulses, the effective pulse length of the laser can be expanded and at tlie same time the peak pulse intensity reduced. The OPuS 240 can thus receive the laser beam from the second WEB 210 via the BAM 230 and direct the output of the OPuS 240 to the CASMM 250.

[0034] These various modules of the laser radiation source can be configured to provide data indicative of diagnostics, configuration, and performance of the module including its overall health. Data indicative of overall health may include an assessment of the likelihood of module failure within a future interval of defined duration. As an example, supervised machine learning models can be trained to predict module end-of-life (EOL). The supervised machine learning modeling may use data obtained from deinstalled modules. The model may retrospectively label each cycle (e.g., day) day ofthe deinstalled module’s service life as “No Technical Issue” or “Failure” depending on how the module was classified in service reports before it was deinstalled. Using daily laser performance data, the model can learn when an installed module is approaching a failure condition and return a numerical score indicating whether a failure is impending. The magnitude of the score may indicate a level of certainty of the impending failure. This scoring system allows users to proactively plan part replacements and effect daily evaluations of module performance.[00351 A disadvantage of a system in which the machine learning model returns numerical scores is that a complex machine learning model may generate scores that are difficult to interpret. For example, one learning model may be an ensemble model such as a random forest model. Ensemble machine learning models combine multiple individual models to improve overall performance. However, the complexity of ensemble models may make it difficult to interpret and understand the results they render. This may leave a user supplied only with those results unsure as to what maintenance actions to perform.

[0036] A natural language processing system such as one underlying a virtual assistant can provide a user with a natural language description of maintenance recommendations and other troubleshooting information. This is especially true when the natural language processing system has ingested relevant information such as field service documentation. A natural language processing system, however, is not configured to accept queries in the form of numerical scores.

[0037] Fuzzy logic is a form of logic that allows for reasoning about imprecise or vague information. Unlike traditional binary logic, which requires everything to be either true or false, fuzzy logic can operate based on the concept of partial truth, where truth values range between completely true and completely false. This makes fuzzy logic particularly useful for dealing with real -world scenarios where information is often not black and white.

[0038] In particular, and according to an aspect of an embodiment, a fuzzy logic system can act as a bridge between a machine learning model and a natural language processing system by translating data, e.g., numerical scores generated by the machine learning model into natural language queries to which the natural language processing system can respond. The three technologies and methods (machine learning model, fuzzy logic, and natural language processing) may be integrated to create a self-aware module that can reference service documentation to provide users with a better understanding of the reasons the machine learning model is returning various scores (or combinations of scores) and what maintenance actions should be considered in view of those scores.

[0039] In other words, according to an aspect of an embodiment, the technology combines the results of the machine learning model, fits the model using fuzzy logic methods, and returns a self-assessment prompt that can be fed into a natural language processing system for recommendations and troubleshooting techniques.

[0040] The machine learning model may be an ensemble model. Ensemble models are machine learning models that combine multiple models to make more accurate predictions. Multiple modelsare trained on the same task and the predictions from each model are combined to create a single prediction. The three main classes of ensemble learning methods are bagging, stacking, and boosting.0041] lire ensemble model may be a random forest model. Random forest is an ensemble machine learning algorithm used for classification and regression tasks. It uses a collection of decision trees that work together to make predictions. Each tree is trained on a random subset of the data (with replacement) and a random subset of features. This process is a type of bagging or bootstrap aggregating.

[0042] Natural language processing is a subset of artificial intelligence that enables virtual assistants to understand (NLU) queries in natural language and to generate (NLG) responses in natural language. The virtual assistant can understand the intent of a query and respond in natural language.

[0043] In fuzzy logic, fuzzy levels, e.g., “high, “ ‘’medium,” and ’‘low” are created for a range of values for each feature. This procedure is referred to as “fuzzification.” One or more membership functions are used to determine a degree of membership in the fuzzy sets with values ranging between 0 and 1. Any value between 0 and 1 represents the degree of uncertainty that the value belongs in the set. FIG. 3 illustrates an example of assigning membership values using three triangular membership functions, A point X on the horizontal axis has three “truth values,” L, M, and H, one for each of the three membership functions low, medium, and high represented by the triangles going from left to right in the figure. The vertical line at X represents a particular value that the three arrows (truth values) gauge based on where the vertical line intercepts the boundary of each fuzzy set. Because the arrow labeled L points to zero, this value may be interpreted as “not low,”; i.e. X has zero membership in the fuzzy set “low'.” Tire arrow' M (pointing at. 0.16) may be described as “slightly medium” and the arrow' H (pointing at 0.63) may be described as “fairly high.” Therefore, value at X has 0.16 membership in the fuzzy set “medium” and 0.63 membership in the fuzzy set “high.”

[0044] As an example, an implementation may start with a database of laser data. This laser data may include historical daily performance data for deinstalled modules. This data may be used to create trained and validated machine learning models. Once a model has been, trained, validated, and deployed, future daily laser performance data for an installed and operational module may be used to generate numerical scores indicative of laser module diagnostics, configuration, or performance. For example, a numerical score may range from 1 to 5 grouped into levels of increasing criticality, e.g., extend module service (healthy module with no apparent issues), monitor module condition, investigate module condition, and replace module.

[0045] These scores may be paired with the features that were used to generate the scores. Fuzzy logic may then be applied to each score to convert these feature / score pairs to generalized text. This data transformation will allow' the data to be used by a large language model.

[0046] As a specific example, assume the machine learning model generates a numerical score in the range of I -5 for three parameters or features. It will be understood that this number of fixatures i s small to simplify the example and that in actual practice the number of features will typically be greater.Assume these features are shot count, chamber voltage, and chamber electrode position. Assume further that the machine learning model generates numerical scores as follows:SHOT COUNT: 4 (indicating a relatively high shot count)CHAMBER_VOLTAGE: 3 (indicating that the chamber is using a medium voltage to generate a pulse having the desired optical energy)ELECTRODE POSITION: 4 (indicating that the electrode is highly eroded).

[0047] The fuzzy logic module may then assign a membership degree to each feature. With a degree and value degree for each row of the data set, a fuzzy logic model can be created from which rales may be derived. For example, a rule degree may be determined as the product of the membership degrees as follows:SHOT_COUNT(high).75 * CHAMBER_VOLTAGE (med).5 * ELECTRODE_POSITION (high).8 = rule degree .3:

[0048] The rule becomes:If SHOT COUNT is High and CHAMBER VOLTAGE is Medium and ELECTRODE POSITION is High then Response is High with degree.3

[0049] This rale may then be used as a basis for forming a query to the natural language processor which can then return a prompt such as “Follow these steps to troubleshoot issues with high shot count voltage, medium chamber voltage, and high electrode position in the laser chamber” along with a detailing of the recommended steps such asverify shot countreset pulse counterevaluate need for gas refillcheck chamber voltage set pointcheck electrode positioning.

[0050] The user thus receives the benefit of having a targeted investigation plan instead of having to rely only on an increasing score to evaluate laser performance.

[0051] Thus machine learning based scoring which has previously been used to make end-of-life assessments is leveraged to facilitate addressing issues that can occur throughout the life of a module that need attention but may not need module replacement to resolve. This can help users perform troubleshooting more efficiently. In other words, instead of waiting for scores to indicate that module replacement is necessary a user could use the scores to quickly troubleshoot why a score increased and address the cause with measures less burdensome than module replacement.

[0052] FIG. 4 is a functional black diagram of a system for automatic generation of recommended maintenance (including troubleshooting) actions according to an aspect of an embodiment. In the arrangement shown in FIG. 4, a laser 300 generates data indicative of parameters of modules of the laser 300 such as diagnostic information, configuration, and performance. These may include parameters such as a shot count, chamber voltages, and the like. A machine learning model 310receives the data and generates scores tor each of the parameters. A translator 320, which may be implemented using fuzzy logic, translates or transposes the scores into natural language queries. A natural language processor 330 receives the natural language queries and generates recommendations in natural language that can be readily comprehended by service personnel such as users as part of a maintenance plan. In some embodiments when a single membership degree does not apply, for example, due to a higher-order uncertainty caused by environment noise or individual system variation, a type-2 fuzzy logic using a range of possibility degrees applies.

[0053] FIG. 5 is a hybrid functional block diagram and flow chart of a system for automatic generation of recommended maintenance actions according to another aspect of an embodiment. In the arrangement shown in FIG. 5, a laser 400 generates data indicative of parameters of modules of the laser 400 such as diagnostic information, configuration, and performance. This data is loaded into a laser database 410. The laser database 410 may also include information on characteristics for uninstalled laser modules. The information in the laser database 410 is used in a step S420 to train a machine learning model. As mentioned, the machine learning model may be an ensemble machine learning model and, and, in particular, a random forest machine learning model.[0054} The machine learning model trained in machine learning model training step S420 is used by a machine learning prediction system 430 to generate scores relating to the laser module characteristics indicated by the laser data. These scores are used to create a fuzzy logic model in a step S440 which is then used in a fuzzy logic classifier 450. The fuzzy logic classifier 450 translates the scores generated by the machine learning prediction system 430 into natural language and supplies them as a query’ to a natural language processor 460. Tire natural language processor 460 may include information such as field service documentation 470. The natural language processor 460 then generates maintenance recommendations 480 including troubleshooting recommendations.

[0055] FIG. 6 is a flow chart outlining steps of a method of performing maintenance on a laser system in accordance with an aspect of an embodiment. In a step S500 parameters relating to a laser system such as diagnostics, configuration, and performance of modules in the laser system are obtained. In a step S510 numerical scores are generated for the laser system parameters. As noted above, this step may’ be performed by a machine learning model such as a random forest ensemble model. In a step S520 the numerical scores are translated into natural language. As noted above, this step may be performed by using fuzzy logic. In a step S530, the natural language generated in step S520 is used to query a virtual assistant. In a step S540 the virtual assistant generates a natural language maintenance recommendation.

[0056] Thus, in some embodiments, machine learning, fuzzy logic, and natural language processing are integrated to create an essentially self-aware system that can reference internal data and provide users with insights based on numerical scoring to improve the maintenance decision -making process. Laser data (features) is used to train a machine learning model. The trained model generates scores for the features. Fuzzy logic is then applied to convert these scores and features into generalized textdescriptions, making the data usable by a natural language processor. The natural language processor creates prompts which then provides maintenance guidance.

[0057] Although specific reference may have been made above to the use of embodiments in the context of optical lithography, it will be appreciated that embodiments may be used in other applications, for example imprint lithography, and where the context allows, is not limited to optical lithography. It is to be understood that the phraseology or terminology herein is for the purpose of description and not of limitation.

[0058] It is to be appreciated that the Detailed Description section, and not the Summary and Abstract sections, is intended to be used to interpret the claims. The Summary and Abstract sections may set forth one or more but not all exemplary embodiments and thus are not intended to limit the embodiments and the appended claims in any way.

[0059] The embodiments 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 and relationships thereof are appropriately performed.

[0060] The foregoing description of the specific embodiments will so fully reveal the general nature of the embodiments that others can, by applying knowledge within the skill of the art, readily modify and / or adapt for various applications such specific embodiments, without undue experimentation, without departing from the general concept of the embodiments. Therefore, such adaptations and modifications are intended to be within the meaning and range of equivalents of the disclosed embodiments, based on the teaching and guidance presented herein.

[0061] The breadth and scope of the embodiments should not be limited by any of the above¬ described exemplary’ embodiments but should be defined only’ in accordance with the following claims and their equivalents.

[0062] The implementations can be further described using the following clauses.1. A maintenance system for a laser radiation source, the system comprising:a machine learning model arranged to receive data from the laser radiation source relating to one or more parameters of the laser radiation source, the machine learning model being adapted to generate one or more numerical scores based at least in part on the data;a translator arranged to receive the numerical scores and adapted to generate a natural language expression indicative of values of the numerical scores; anda natural language processor arranged to receive the natural language expression indicative of values of the numerical scores and adapted to generate a natural language expression of one or more maintenance actions based at least m part on the natural language expression indicative of values of the numerical scores.2. The system of clause 1 wherein the parameters include at least one of diagnostics, configuration,and performance of a module of the laser radiation source.3. The system of clause 1 wherein the maintenance action includes a troubleshooting plan for a module of the laser radiation source.4. The system of clause 1 wherein the machine learning model is an ensemble model.5. The system of clause 4 wherein the ensemble model is a random forest model.6. The system of clause 1 wherein the translator is configured to employ fuzzy logic.7. The system of clause 1 wherein the natural language processor is configured as a large language model virtual assistant.8. The system of clause 7 wherein the natural language expression is a query for the virtual assistant.9. The system of clause 1 wherein the numerical score comprises a predictive score for a process module of the laser radiation source and wherein the predictive score indicates a likelihood of failure of the process module.10. A maintenance method for a laser radiation source, the method comprising:a step performed by the laser radiation source of generating performance data;supplying the performance data to a machine learning model;a step performed by the machine learning model of generating one or more numerical scores based at least in part on the performance data;translating the numerical scores into a natural language expression indicative of values of the numerical scores; andgenerating a natural language expression of one or more maintenance actions based at least in part on the natural language expression indicative of values of the numerical scores.11. The method of clause 10 wherein the machine learning model is an ensemble model.12. The method of clause 11 wherein the ensemble model is a random forest model.13. The method of clause 10 wherein the step of translating the numerical scores into a natural language expression is performed using fuzzy logic.14. The method of clause 10 wherein the step of generating a natural language expression of one or more maintenance actions is performed by a natural language processor.15. The method of clause 14 wherein the natural language processor is configured as a large language model virtual assistant.16. The method of clause 15 wherein the natural language expression is a query for the virtual assistant.17. The method of clause 10 wherein the maintenance actions are troubleshooting recommendations.18. The method of clause 10 wherein the numerical score comprises a predictive score for a process module of the laser radiation source and wherein the predictive score indicates a likelihood of failure of the process module.19. A method for decision-making in maintenance of a process module, the method comprising: collecting operational data of the process module;generating a predictive score for the process module using a machine learning model, wherein thepredictive score indicates a likelihood of failure of the process module;applying fuzzy logic to the predictive score and the operational data to generate generalized descriptions of a status of the process module; andusing a language model to process the generalized descriptions and provide action items for maintenance.20. The method of clause 19, wherein the predictive score is grouped into predefined categories corresponding to the action items.21. The method of clause 19, wherein applying fuzzy logic includes assigning grades to individual features of the operating data, enabling interpretability of the predictive score.22. The method of clause 19, further comprising uploading second operational data of the process module to the language model.23. A system for maintenance of a laser module, the system comprising:a data collection unit configured to gather operational data of the laser module;a machine learning module configured to generate a predictive score indicating a likelihood of failure of the laser module;a fuzzy logic engine configured to process the predictive score and the operational data to produce generalized text descriptions; anda language model interface configured to process the generalized text description and provide action items for maintenance.24. The system of clause 23 further comprising a prompt generation module configured to create self-awareness prompts based on the generalized text description and predefined rules for the language model interface.

[0063] The above -described implementations and other implementations are within the scope of the following claims.

Claims

CLAIMS1. A maintenance system for a laser radiation source, the system comprising:a machine learning model arranged to receive data from the laser radiation source relating to one or more parameters of the laser radiation source, the machine learning model being adapted to generate one or more numerical scores based at least in part on the data:a translator arranged to receive the numerical scores and adapted to generate a natural language expression indicative of values of the numerical scores; anda natural language processor arranged to receive the natural language expression indicative of values of the numerical scores and adapted to generate a natural language expression of one or more maintenance actions based at least in part on the natural language expression indicative of values of the numerical scores.

2. The system of claim 1 wherein the parameters include at least one of diagnostics, configuration, and performance of a module of the laser radiation source.

3. Tire system of claim 1 wherein the maintenance action includes a troubleshooting plan for a module of the laser radiation source.

4. The system of claim 1 wherein the machine learning model is an ensemble model.

5. The system of claim 4 wherein the ensemble model is a random forest model.

6. The system of claim 1 wherein the translator is configured to employ fuzzy logic.

7. The system of claim 1 wherein the natural language processor is configured as a large language model virtual assistant.

8. The system of claim 7 wherein the natural language expression is a query for the virtual assistant.

9. The system of claim 1 wherein the numerical score comprises a predictive score for a process module of the laser radiation source and wherein the predictive score indicates a likelihood of failure of the process module.

10. A maintenance method for a laser radiation source, the method comprising:a step performed by the laser radiation source of generating performance data;supplying the performance data to a machine learning model;a step performed by the machine learning model of generating one or more numerical scores based at least in part on the performance data;translating the numerical scores into a natural language expression indicative of values of the numerical scores; andgenerating a natural language expression of one or more maintenance actions based at least in part on the natural language expression indicative of values of the numerical scores.

11. The method of claim 10 wherein the machine learning model is an ensemble model.

12. The method of claim 10 wherein the step of translating the numerical scores into a natural language expression is performed using fuzzy logic.

13. The method of claim 10 wherein the step of generating a natural language expression of one or more maintenance actions is performed by a natural language processor.

14. The method of claim 10 wherein the maintenance actions are troubleshooting recommendations.

15. The method of claim 10 wherein the numerical score comprises a predictive score for a process module of the laser radiation source and wherein the predictive score indicates a likelihood of failure of the process module.

16. A method for decision-making in maintenance of a process module, the method comprising:collecting operational data of the process module;generating a predictive score for the process module using a machine learning model, wherein the predictive score indicates a likelihood of failure of the process module;applying fuzzy logic to the predictive score and the operational data to generate generalized descriptions of a status of the process module; andusing a language model to process the generalized descriptions and provide action items for maintenance.

17. The method of claim 16, wherein the predictive score is grouped into predefined categories corresponding to the action items.

18. Tire method of claim 16, wherein applying fuzzy logic includes assigning grades to individual features of the operating data, enabling interpretability of the predictive score.

19. The method of claim 16, further comprising uploading second operational data of the process module to the language model.

20. A system for maintenance of a laser module, the system comprising:a data collection unit configured to gather operational data of the laser module;a machine learning module configured to generate a predictive score indicating a likelihood of failure of the laser module;a fuzzy logic engine configured to process the predictive score and the operational data to produce generalized text descriptions;a language model interface configured to process the generalized text description and provide action items for maintenance; anda prompt generation module configured to create self-awareness prompts based on the generalized text description and predefined rules for the language model interface,

Images