Apparatuses and processes for light source system status tracking and maintenance utilizing data to text machine translation

Abstract

A light source system includes two or more light source system modules and a controller in communication with the modules, the controller configured to (1) receive and store multivariate time series performance and / or diagnostic data from the light source system including data from the two or more modules and (2) periodically and / or in response to a triggering signal, produce a generated text sequence by transforming a transformer input data sequence into a text sequence using a sequence-to- sequence transformer, the transformer input data sequence generated at least in part from the stored multivariate time series performance and / or diagnostic data, the generated text sequence including an indication of one or more of (a) a status of the light source system, (b) a maintenance action for the light source system, and (c) a monitoring action for the light source system.

Description

APPARATUSES AND PROCESSES FOR LIGHT SOURCE SYSTEM STATUS TRACKING AND MAINTENANCE UTILIZING DATA TO TEXT MACHINE TRANSLATIONCROSS-REFERENCE TO RELATED APPLICATIONS

[0001] This application claims priority of US application 63 / 665,567 which was filed on June 28, 2025 and which is incorporated herein in its entirety by reference.TECHNICAL FIELD

[0002] The disclosed subject matter relates to status reporting and maintenance of light sources such as those used for integrated circuit photolithographic manufacturing processes, including laser-based deep ultraviolet (DUV) light sources, and particularly to status reporting and maintenance of such systems using data to text machine translation.BACKGROUND

[0003] Light, which can be laser radiation or generated therefrom, that is used for semiconductor photolithography can be supplied by a system that may be referred to as a light source. These light sources can produce light as a series of pulses at specified repetition rates, for example, in the range of about 500 Hz to about 8 kHz, or even higher. Additionally, such light sources can generally be expected to have useful lifetimes measured in terms of the number of pulses they are projected to be able to produce before requiring repair or replacement, typically expressed as billions of pulses.

[0004] One system for generating light at frequencies useful for semiconductor photolithography (such as at deep-ultraviolet (DUV) wavelengths) involves use of a master oscillator power amplifier (MOP A) dual-gas-discharge-chamber configuration. This configuration has two laser chambers, a master oscillator chamber (MO chamber) and a power ring amplifier chamber (PRA chamber). These chambers and many other system components can be structured as and / or regarded as modules, and the light source overall can be regarded as an ensemble of modules. Each module of a light source generally has a respective lifetime that is shorter than the desired lifetime of the overall light source. Thus, over the course of the lifetime of the light source, the health of the light source and / or the health of individual modules of the light source can be evaluated to determine whether specific modules of the light source should be repaired or replaced. Modules can be repaired or replaced in accordance with such evaluations.SUMMARY

[0005] In some general aspects, a light source system includes two or more light source system modules and a controller in communication with the two or more modules, the controller configured to (1) receive and store multivariate time series performance and / or diagnostic data from the light source system including data from the two or more modules and (2) periodically and / or in response toa triggering signal, produce a generated text sequence by transforming a transformer input data sequence into a text sequence using a sequence-to-sequence transformer, the transformer input data sequence having been generated at least in part from the stored multivariate time series performance and / or diagnostic data, the generated text sequence including an indication of one or more of (a) a status of the light source system, (b) a maintenance action for the light source system, and (c) a monitoring action for the light source system.

[0006] Implementations can include one or more of the following.

[0007] The multivariate time series performance and / or diagnostic data from the light source system can include numerical data. The controller can be further configured to produce one or more stored generated text sequences by storing the generated text sequence as (a) the generated text sequence in a non-user-approved form and / or (b) the generated text sequence in a user-approved form and / or (c) the generated text sequence in a user-edited form. The transformer input data sequence can be generated from (1) the multivariate time series performance and / or diagnostic data, and (2) at least one previously produced stored generated text sequence.

[0008] The sequence-to-sequence transformer can be a transformer trained using as training inputs sequences of data generated from multivariate time series data relating to performance and / or diagnostics of the light source system and / or of one or more light source systems of the same type, and using as training targets time-correlated human-generated and / or human-approved texts relating to the status, maintenance, and / or monitoring of the respective light source system or systems.

[0009] The sequence-to-sequence transformer can be a transformer trained using as training inputs (1) sequences of data generated from multi- variate time series data relating to performance and / or diagnostics of the light source system and / or of one or more light source systems of the same type, and (2) one or more time- and light source system-corelated human-generated and / or human approved texts relating to the status, maintenance, and / or monitoring of the respective light source system or systems.

[0010] The controller can be configured to generate the transformer input data sequence and include in the transformer input data sequence variable labels and associated variable values. The controller can be configured to generate the transformer input data and include in the transformer input data variable labels and associated variable values and associated time stamps. 21. The controller can be configured to receive and store multivariate asynchronous time series performance and / or diagnostic data from the light source system and the transformer input data sequence has been generated at least in part from the stored multivariate asynchronous time series performance and / or diagnostic data.

[0011] In additional general aspects, a process for automatically producing a text sequence relating to the status, maintenance, and / or monitoring of a light source system includes (1) forming a transformer input data sequence including data of a light source system by transforming multivariate time series data into string sequence data including variable labels and associated variable values and (2) producing a generated text sequence by machine-translating the transformer input data sequence into atext sequence relating to the status, maintenance and / or monitoring of the light source system using a sequence-to-sequence transformer.

[0012] Implementations can include one or more of the following.

[0013] The data of the light source system can include data relating to performance and / or diagnostics of the light source system. Transforming the multivariate time series data into string sequence data can include ordering and tokenizing the multivariate time series data within a string sequence as a succession of values ordered by variable type. Transforming the multivariate time series data into string sequence data can include ordering and tokenizing the multi-variate time series data within a string sequence as a succession of values ordered by time. The process can further include producing one or more stored generated text sequences by storing the generated text sequence, and forming a transformer input data sequence can include transforming and combining both (1) the multivariate time series data and (2) one or more previously stored text sequences.

[0014] The sequence-to-sequence transformer can be trained using as inputs sequences of data of the light source system generated from multi-variate time series data relating to performance and / or diagnostics of the light source system, and / or of one or more light source systems of the same type as the light source system, and using as targets time-correlated human-generated and / or human-approved text relating to the status, maintenance, and / or monitoring of the respective light source system or systems.

[0015] The sequence-to-sequence transformer can be trained using as inputs sequences of data of the light source system generated from both (1) stored multi- variate time series data relating to performance and / or diagnostics of the light source system, and / or of one or more light source systems of the same type and (2) past-time-correlated stored human-generated and / or human-approved texts relating to the status, maintenance, and / or monitoring of the respective light source system or systems, and using as targets current- time-correlated stored human-generated texts relating to the status, maintenance, and / or monitoring of the respective light source system or systems.

[0016] Forming a sequence of data of the light source system can include transforming and tokenizing multi-variate time series data relating to performance and / or diagnostics of the light source system into string sequence data including variable labels with associated variable values. The process can further include, in response to the text sequence containing one or more previously specified words or phrases, triggering and / or sending an alert relating to and / or containing the content or a portion of the content of the text sequence. The process can further include, in response to the text sequence containing one or more previously specified words or phrases, automatically performing maintenance and / or an adjustment to the light source system. The process can include transforming multivariate asynchronous time series data into string sequence data including variable labels and associated variable values.

[0017] In still more general aspects, a process of training a sequence-to-sequence transformer for producing text sequences relating to the status, maintenance, and / or monitoring of a light sourcesystem includes: using as inputs at least sequences of data of the light source system generated from multi-variate time series data relating to performance and / or diagnostics of the light source system, and / or of one or more light source systems of the same type as the light source system, tokenized into string sequence data including variable labels with associated variable values and time stamps, and using as targets time-correlated human-generated and / or human-approved text relating to the status, maintenance, and / or monitoring of the respective light source system or systems.

[0018] Implementations can include one or more of the following.

[0019] The process can further use as inputs time-correlated human-generated texts relating to the status, maintenance, and / or monitoring of the respective light source system or systems. The process can use as inputs at least sequences of data of the light source system generated from multi-variate asynchronous time series data relating to performance and / or diagnostics of the light source system.

[0020] Implementations can include one or more of the following. . . .

[0021] The details of one or more implementations are set forth in the accompanying drawings and the description below. Other features will be apparent from the description and drawings, and from the claims.DRAWING DESCRIPTION

[0022] FIG. 1 is a schematic cross-sectional diagram of aspects of a light source.

[0023] FIG. 2 is a schematic diagram of a lithography exposure apparatus that can be used with a light source such as the light source of FIG. 1.

[0024] FIG. 3 is a process flow diagram showing elements of a process for automatically producing a text sequence relating to the status, maintenance, and / or monitoring of a light source system such as the light source of FIG. 1.

[0025] FIG. 4 is a diagram showing one or more aspects of a process for training a sequence-to- sequence transformer such as the sequence-to-sequence transformer of FIG. 3.

[0026] FIG. 5 is a diagram of data structures and / or sequences similar to those discussed with respect to FIGS. 3 and 4 above, useful in additional implementations of processes disclosed herein.

[0027] FIG. 6 is a diagram of additional data structures and / or sequences similar to those discussed with respect to FIGS. 3 and 4 above, useful in additional implementations of processes disclosed herein.

[0028] FIG. 7 is a diagram of further data structures and / or sequences similar to those discussed with respect to FIGS. 3 and 4 above, useful in additional implementations of processes disclosed herein.

[0029] FIG. 8 is a diagram of still further data structures and / or sequences similar to those discussed with respect to FIGS. 3 and 4 above, useful in additional implementations of processes disclosed herein.

[0030] FIG. 9 is a diagram of yet further data structures and / or sequences similar to those discussed with respect to FIGS. 3 and 4 above, useful in additional implementations of processes disclosed herein.

[0031] FIGS. 10 and 11 are diagrams showing implementations of tokenized input sequences such as those described in FIGS. 3-9.DETAILED DESCRIPTION

[0032] Referring to FIG. 1, a light source 100, which can be a deep UV (DUV) light source 100, can be in the form of a dual stage pulsed light source that produces a light beam 105 in the form of a pulsed amplified beam. The light source 100 includes a solid state or gas discharge master oscillator (MO) system 160, a power amplification (PA) system such as a power ring amplifier (PRA) system 165, relay optics 170, and an optical output subsystem 175.

[0033] The MO system 160 can include, for example, an MO chamber 161. In the MO chamber, electrical discharges between electrodes (not shown) can generate an inverted population of high energy molecules in a lasing gas, which molecules discharge and lase together to produce relatively broadband radiation. The relatively broadband radiation is line-narrowed to a relatively narrow bandwidth, and center-wavelength selected, in a line narrowing module (LNM) 162.

[0034] The MO system 160 can also include an MO output coupler (MO OC) 162, which can include a partially reflective mirror, forming, with a reflective grating in the LNM 162 (not shown), an oscillator cavity in which light oscillates to generate a seed output pulse. The MO system 160 can also include a line-center analysis module (LAM) 163. The LAM 163 can include, for example, an etalon spectrometer for fine wavelength measurement and a coarser-resolution grating spectrometer.

[0035] The relay optics 170 can include an MO wavefront engineering box (WEB) 171 that serves to redirect the output of the MO system 160 toward the PRA system 165, and can include, for example, beam expansion, such as by a multi prism beam expander (not shown), and coherence busting, for such as by use of one or more optical delay paths (not shown).

[0036] The PA or PRA system 165 includes a power amplifier (PA) or power ring amplifier (PRA) chamber 166 (hereinafter the terms PRA or power ring amplifier will be used to represent both a power amplifier and a power ring amplifier interchangeably, and PRA system will be used to represent both power ring amplifier systems and power amplifier systems). Like the MO chamber 161, the PRA chamber 166 is an oscillator, oscillating in response to injection of the output light pulse or pulses from the MO system 160, and due to output coupling optics that can be incorporated into a PRA WEB 167 and can be redirect pulses back through a gain medium in the chamber 166, in cooperation with a beam reverser 168. The PRA WEB 167 can incorporate a partially reflective input / output coupler (not shown) and a maximally reflective mirror for the nominal operating wavelength, which can be at around 193 nm for an ArF system, and one or more prisms. The PRA system 165 optically amplifies the output light beam from the MO system 160.

[0037] The optical output subsystem 175 can include a bandwidth analysis module (BAM) 176 at the output of the PRA system 165. The BAM 176 can pick off for metrology purposes a portion of the light beam it receives. For example, the BAM can use the portion of the light beam to measure the output bandwidth and pulse energy of the light. The output light beam of pulses then passes through an optical pulse stretcher module (OPuS) 177 and an output combined autoshutter metrology module (CASMM) 178, which can include a pulse energy meter. One purpose of the OPuS 177 can be to convert a single output pulse into a pulse train. Secondary pulses created from the original single output pulse can 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 light beam can be expanded and at the same time the peak pulse intensity of the beam can be reduced.

[0038] The light source 100 is made up of modules. Each of the components (such as the MO chamber 161, the LNM 162, the MO WEB 171, the PRA chamber 166, the PRA WEB 167, the OPuS 177, the BAM 176) of the light source 100 are structured as and / or can be considered to be modules. The overall availability of the light source 100 is the direct result of the respective availabilities of these individual modules making up the light source 100. In other words, the light source 100 cannot operate properly and / or be “available” (such as for use in lithography processes) unless all of the modules making up the light source 100 are themselves operating properly or “available.” A controller 120 monitors these modules so that they can be adjusted, refreshed, or replaced, generally before they fail, in order to maintain the operation of the light source 100 and optimize and improve productivity of an output apparatus such as a lithography apparatus 210 of FIG. 2, described below. The controller 120 can collect and analyze data relating to performance and status or “health” and other properties of the modules and of the light source 100. The controller 120 can also provide maintenance alerts that can be used to perform automated and / or manual maintenance tasks for one or more specific modules including replacement tasks when needed to prevent failure of the light source 100.

[0039] Referring to FIG. 2, the amplified light beam 105 (FIG. 1) can be useful as a light beam 205 used by a photolithography exposure apparatus 210 to pattern features on a substrate or wafer 211. The wafer 211 is placed on a wafer table 212 constructed to hold the wafer 211 and connected to a positioner configured to position the wafer 211 accurately in accordance with certain parameters. The light beam 205 can have a wavelength in the deep ultraviolet (DUV) range, which can include wavelengths from, for example, about 100 nanometers (nm) to about 400 nm. For example, the light source 100 that produces such a light beam 105, 205 can be a gas discharge light source such as an excimer light source, or excimer laser that uses a combination of one or more noble gases, which can include argon, krypton, or xenon, and a reactive gas, which can include fluorine or chlorine as the gain medium. The light source 100 can be an excimer light source. Thus, for example, the gain medium can include argon fluoride (ArF), krypton fluoride (KrF), or xenon chloride (XeCl). If the gain medium includes argon fluoride, then the wavelength of the amplified light beam 205 is about193 nm and if the gain medium includes krypton fluoride, then the wavelength of the amplified light beam 205 is about 248 nm. The size of the microelectronic features patterned on the wafer 211 depends on the wavelength of the light beam 205, with a lower wavelength resulting in a smaller minimum feature size. When the wavelength of the light beam 205 is 248 nm or 193 nm, the minimum size of the microelectronic features can be, for example, 50 nm or less. The bandwidth of the light beam 205 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 205 is distributed over different wavelengths.

[0040] The photolithography exposure apparatus 210 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 205 or in a plane that is perpendicular to the optical axis. The objective arrangement includes a projection lens and enables an image transfer to occur from the mask to the photoresist on the wafer 211. The photolithography exposure apparatus 210 also includes an illumination system that adjusts the range of angles for the light beam 205 impinging on the mask. The illumination system also homogenizes (makes uniform) the intensity distribution of the light beam 205 across the mask.

[0041] The photolithography exposure apparatus 210 can also include, among other features, a lithography controller 213 that controls how layers are printed on the wafer 211. The lithography controller 213 includes a memory that stores information such as process recipes. A process program or recipe determines the length of the exposure on the wafer 211, the mask used, and other factors that affect the exposure. During lithography, a plurality of pulses of the light beam 205 illuminates the same area of the wafer 211 to together constitute an illumination dose.

[0042] The quality of the features produced on the wafer 211 by the photolithography exposure apparatus 210 depends directly upon the quality and reliability of the light pulses from the light source 100. Pulses having lower than desired power can result in underexposure of an area of the wafer 211. Missing pulses can similarly result in underexposure. Shifts in wavelength or bandwidth distribution can result in shifts in image position and alterations in patterns produced at the wafer 211.

[0043] Data relating to performance and / or status of the various modules of the light source 100, or relating to the performance and / or status of the light source 100 as a whole can be received by the controller 120 from the modules of the light source 100 and stored. Technicians can monitor the performance and / or status of the modules and / or light source based on current and stored data. Technicians can also make entries in an electronic logbook documenting the performance and / or status of the modules and / or light source. Entries can be made at fixed time intervals, such as weekly, and / or at irregular time intervals whenever the performance and / or status of the light source or one or more modules thereof requires action. One or more user interfaces 120uif, such as touch screens, workstations, wireless handheld devices, web-based or local network interface applications running on any of these, and the like can be in communication with the controller 120, whether remotely orlocally or both, to facilitate logbook entries. Entries can include recommendations regarding maintenance actions and / or monitoring of one or more modules and / or the light source. For instance, a recommendation for no action can be recorded as “no action” or “no action needed” or the like. A recommendation for monitoring can be recorded, for example, as “monitor ‘x’ variable for values above / below ‘y’” or the like. A need for a maintenance action can be recorded, for example, as “perform maintenance action ‘z’ on or before time ‘t’” or the like.

[0044] As explained in greater detail below, in the presently disclosed processes and apparatuses, entries for the technician’ s logbook or its equivalent can be generated by a trained machine learning process, such as a transformer, utilizing, at least in part, data-to-text machine translation. The previously trained machine learning process translates from the data relating to the performance and / or the status of the various modules of the light source 100, and / or of the light source 100 as a whole, to an appropriate English language (or other language) logbook entry. The entries can be generated in final form or can be generated subject to approval and / or editing by one or more technicians before becoming “final.” The controller 120 can store the entries in their final form.

[0045] FIG. 3 is a process flow diagram showing elements of a process 330 for producing a text according to the present disclosure. Performance and / or diagnostic data of a light source such as the light source 100 of FIG. 1 is available and / or stored in the form of multivariate time series data (MVTS) 332, represented in the figure as including various values 340 within various series variables V (A, B, C, D, E, and so forth, rightward in the grid 333), distributed over time T (downward in the grid 333). As represented in the figure, a variable may be produced and / or received and stored at fixed time intervals, such as variables A and B, or at varying intervals, such as variable C, such that the variables taken together can constitute a multivariate asynchronous time series, with asynchronous as used here meaning that the variables are not synchronized in time.

[0046] To generate a (proposed or final) logbook entry in the form of a generated text sequence (GTS), a transformer input data sequence TIDS is generated from the MVTS data by transforming the MVTS data into string sequence data 334a or 334b. The MVTS data can be transformed into string sequence data by ordering the MVTS data (within a recent time interval, such as beginning after the last logbook entry, for instance) as a succession of values 350 ordered by variable type, then by time (334a). Alternatively, the MVTS data can be transformed into string sequence data by ordering the MVTS data as a succession of values 350 over time (such as by individual times of the values, or as groups of values within preselected time intervals), then by variable type if any values have the same time or time interval (334b). Other defined, repeatable processes of transformation into string sequence data can also be used.

[0047] The input data sequence is then translated into a generated text sequence GTS 338, relating to one or more of (a) a status of the associated light source system, (b) a maintenance action for the light source system, and (c) a monitoring action for the light source system, by a previously trained sequence-to-sequence transformer SST 336. The process 330 and the SST 336 can be implemented inor in connection with a controller such as the controller 120, allowing a light source such as the light source 100 to provide or propose appropriate logbook entries periodically or in response to a signal or request.

[0048] The SST can be an attention-based process of a type or of types known to those of skill in the art of machine learning and / or machine translation. See, for example, Vaswani, Ashish, Shazeer, Noam, Parmar, Niki, Uszkoreit, Jakob, Jones, Llion, Gomez, Aidan N., Kaiser, Lukasz and Polosukhin, Illia; “Attention is All you Need,” presented at the meeting of the Advances in Neural Information Processing Systems 30: Annual Conference on Neural Information Processing Systems 2017, December 4-9, 2017, Long Beach, CA, USA, 2017, paper and abstract accessible at https: / / papers.nips.cc / paper_files / paper / 2017 / hash / 3f5ee243547dee91fbd053clc4a845aa- Abstract.html and related citing works. Additional attention techniques can be applied such as soft attention, hard attention, global attention, and local attention.

[0049] Before the SST is used to produce a generated text sequence (GTS) as shown in FIG. 3, the SST 336 is trained. FIG. 4 is a process diagram representing aspects of processes of training the SST 336. Training of the SST 336 is performed iteratively using as inputs training data sequences TDSs 337 such as TDSa, TDSb, TDSc, and so on, generated from previously generated and stored MVTS data relating to performance and / or diagnostics of the light source system in use, and / or of one or more light source systems of the same type as the light source system in use. The training data sequences TDSs 337 (TDSa, TDSb, TDSc, and so on), can be produced from respective sets of MVTS data (not shown) by transforming each respective set into string sequence data, as discussed above with respect to FIG. 3. Note that to save space in FIG. 4, only TDSs 337 corresponding to the alternative of FIG. 3 item 334a is shown, but either alternative (334a or 334b), or other transformations, can be used. While different transformations can be chosen, note also that the same transformation from MVTS data to sequence-form data is used for both training of an SST and for production of generated text sequences (GTSs) using that SST. Time-correlated (and system- correlated) text training sequences TTSs 335 (such as human-generated and / or human-approved texts HTa, HTb, HTc, and so on), relating to the status, maintenance, and / or monitoring of the respective light source system or systems are used for training targets. Time-correlated, in this context, means that each human-generated and / or human-approved text, such as text HTa, is paired with a corresponding training data sequence, TDSa in this case, formed from data that temporally preceded and led up to the text HTa. System-correlated, in this context, means that each human-generated and / or human-approved text, such as HTa, is paired with a corresponding training data sequence, TDSa in this case, from the same light source system. These correlations are represented graphically by the double-ended arrows in FIG. 4.

[0050] FIGS. 5-9 are diagrams of various alternative data structures and / or sequences that can be used for training of the SST and for production of generated text sequences (GTSs) by the trained SST as generally described above.

[0051] FIG. 5 is a diagram of data structures and / or sequences similar to those discussed with respect to FIGS. 3 and 4 above. With reference to FIG. 5, the row R represents a time sequence of sets of multivariate time series data from a most recent or currently referenced set MVTSD-0 to a least recent set MVTSD-n, together with a series of texts in the form of logbook entries from a most recent or currently reference text LE-0 to a least recent text LE-n, for a given light source system. Training data sequence TDSa can be produced by transforming into a data sequence the currently referenced set MVTSD-0 of multivariate time series data. A time- and system- correlated training text sequence in the form of human approved and / or human generated text HTa can be sourced from the most recent or currently referenced text LE-0 for use as a training target, paired (as in FIG. 4, represented by a double-ended arrow) with the training data sequence TDSa for training the SST. Similar pairs of training data and associated targets can be generated for the data and text pairs MVTSD-1 and LE-1, MVTSD-2 and LE-2, and so forth. Similar pairs of training data and associated targets can also be generated for as many other light source systems of the same or sufficiently similar type for which the necessary data is available to enable production of a significant amount of training data. Also represented in FIG. 5 is the previously noted correspondence (represented by the dotted outline arrow) between the format of the training data sequences (337 of FIG. 4) such as TDSa of FIG. 5 and the transformer input data sequences TIDS 334 (and such as 334a and 334b of FIG. 3) used for production of generated text sequences (GTSs).

[0052] FIG. 6 is a diagram of alternative data structures and / or sequences. The row R is the same as described in FIG. 5 above, but in the implementation shown in FIG. 6, the training data sequence TDSa can be produced by transforming into a data sequence the currently referenced set MVTSD-0 of multivariate time series data as well as one or more earlier sets, such as one or more of MVTSD-1, MVTSD-2, and so forth. A time- and system- correlated training text sequence in the form of human approved and / or human generated text HTa, as in FIG. 5, can be sourced from the most recent or currently referenced text LE-0, for use as a training target, paired with the training data sequence TDSa for training the SST. As above, similar pairs of training data and associated targets can be generated for LE-1, LE-2, and so forth. Similar pairs of training data and associated targets can also be generated for as many other light source systems of the same or sufficiently similar type for which the necessary data is available to enable production of a significant amount of training data. Also represented in FIG. 6 is the previously noted correspondence (represented by the dotted outline arrow) between the format of the training data sequences such as TDSa and the transformer input data sequences TIDS 334 (and such as 334a and 334b of FIG. 3) used after training during production of generated text sequences (GTSs). Relative to the implementation in FIG. 5, longer data sequences (for both training and production of generated text) are used in this implementation.

[0053] FIG. 7 is a diagram of further alternative data structures and / or sequences. In FIG. 7, the row R is the same as described in FIG. 5 above, but in the implementation shown in FIG. 7, the training data sequence TDSa can be produced by transforming into a data sequence the currently referencedset MVTSD-0 of multivariate time series data, and by adding to the data sequence (such as by appending on one of its ends, for instance) the (immediately) previous text sequence LE-1. This forms a training data sequence such as TDSa including both a text sequence TS and a number-containing sequence NS, as shown. This training data is paired, as in the previous examples, with a time- and system- correlated training text sequence in the form of human approved and / or human generated text HTa sourced from the most recent or currently referenced text LE-0, the text or logbook entry immediately following the currently referenced set of multivariate time series data MVTSD-0. When the SST is trained in this manner, the transformer input data sequences TIDSs 334 (or such as in items 334a and 334b of FIG. 3) used for production of generated text sequences GTSs again have a format corresponding to the format of the training data sequences such as TDSa. Including a text sequence of the immediately previous text in the training data sequences and, after training, in the input data sequences, can increase the context from which the SST can learn (during training) and generate (during production of generated text).

[0054] FIG. 8 is a diagram of more alternative data structures and / or sequences. In FIG. 8, the row R is the same as described in FIG. 5 above, but in the implementation shown in FIG. 8, the training data sequence TDSa can be produced by transforming into a data sequence the currently referenced set MVTSD-0 of multivariate time series data and one or more previous sets (MVTSD-1, MVTSD-2, and so forth) of multivariate time series data and by adding to each such data sequence corresponding text sequences (a text sequence from EE-1 to the sequential transformation of MVTSD-0, and one or more of: a text sequence from LE-2 to the sequential transformation of MVTSD-1, a text sequence from LE-3 to the sequential transformation of MVTSD-2, and so forth). This forms a training data sequence such as TDSa having alternating text sequences TS and a number-containing sequences NS as shown. This training data sequence is paired, as in the previous examples, with a time- and system- correlated training text sequence in the form of human approved and / or human generated text HTa sourced from the currently referenced text LE-0. When the SST is trained in this manner, the transformer input data sequences such as TIDS 334 (or 334a and 334b of FIG. 3) used for production of generated text sequences (GTSs) again has a format corresponding to the format of the training data sequences such as TDSa, with alternating text sequences TS and a number-containing sequences NS. Including multiple text and number sequences in the training data sequences and, after training, in the input data sequences can further increase the context from which the SST can learn (during training) and generate (during production of generated text).

[0055] FIG. 9 is a diagram of still more alternative data structures and / or sequences. In FIG. 9, the row R is again the same as described in FIG. 5 above, but in the implementation shown in FIG. 9, the training data sequence TDSa can be produced by transforming into a data sequence the currently referenced set MVTSD-0 of multivariate time series data and by adding two or more previous corresponding text sequences, such as by appending them to the front or back of the transformed data sequence. A text sequence generated from LE-1 and text sequences generated from one or more ofLE-2, LE-3, and so forth can be appended to the sequential transformation of MVTSD-0, as shown. This forms a training data sequence such as TDSa having multiple text sequences TS and a numbercontaining sequence NS. This training data sequence is paired, as in the previous examples, with a time- and system- correlated training text sequence in the form of human approved and / or human generated text HTa sourced from the currently referenced text LE-0. When the SST is trained in this manner, the transformer input data sequences such as TIDS 334 (and / or items 334a and 334b of FIG. 3) used for production of generated text sequences (GTSs) again have a format corresponding to the format of the training data sequences such as TDSa, with multiple text sequences TS and a numbercontaining sequences NS, as shown. Multiple text sequences are used with only one numbercontaining sequence (with “one” defined here as a sequence transformation of the multivariate time series data temporally between successive text or logbook entries). Using this format in the training data sequences and, after training, in the input data sequences, can further decrease the total data and calculation load for the SST relative to the implementation of FIG. 8, while still allowing a significant increase the context from which the SST can learn (during training) and generate (during production of generated text). Depending on the capacity of the SST and associated infrastructure, this implementation can provide training data and corresponding input data for production of generated text that reaches back in time to the placing in-service of a major module and / or of the light source system itself. This can allow the SST to learn aspects of module and / or light source system lifecycle characteristics.

[0056] As mentioned above in reference to FIG. 3, the multivariate time series data (MVTS) can be ordered within the transformer input data sequence (TIDS) as a succession of values 350 ordered by variable type, or by time, or by other repeatable ordering. The transformer input data sequences (TIDSs) can also be tokenized, meaning structured or arranged in ways that make machine learning easier and / or more productive. For example, as shown in FIG. 10, each value 350 can include, grouped together as an ordered group within the string, a variable label VE and a variable value V V. Alternatively, for example, as shown in FIG. 11 , each value 350 can include, grouped together as an ordered group within the string, a variable label VL and a variable value V V and a variable time stamp VT. The text training sequences (TTSs, FIG. 4) can also be tokenized as known in the art for text, such as English language text. Alternatively, if desired, the text training sequences such TTSs 335 of FIG. 4 can be tokenized in ways specifically adapted to the technical language, vocabulary, and style of light source system logbook entries or similar texts.

[0057] The processes described herein can be implemented, controlled, and / or triggered by a processor such as the processer 120 of FIG. 1 or some other processor. An implementing, controlling, and / or triggering processor, as well as the processor 120, can take the form of hardware, software, firmware, or other forms of processing capability, and need not be implemented in a single apparatus or location, but can be distributed physically and functionally, including across many processors and devices. Memory storage can be included in or merely used by the processor, or any combinationthereof. In response to the production of a generated text sequence containing one or more previously specified words or phrases, an alert can be triggered and / or sent relating to and / or containing the content or a portion of the content of the generated text sequence. Alternatively or in addition, in response a generated text sequence containing one or more previously specified words or phrases maintenance of and / or an adjustment to the light source system can be performed automatically.

[0058] The processes and apparatuses disclosed herein can allow a light source system or a machine learning or machine translation system associated therewith to produce generated text that functions as a prediction of, or a draft of, a logbook entry for a maintenance or status log of the light source system, saving time and resources and allowing machine-learned expertise to contribute to the maintenance and status monitoring of the light source system.

[0059] Aspects and implementations of the present disclosure can be further described using the following clauses:1. A light source system including: two or more light source system modules; a controller in communication with the two or more modules, the controller configured to (1) receive and store multivariate time series performance and / or diagnostic data from the light source system including data from the two or more modules and (2) periodically and / or in response to a triggering signal, produce a generated text sequence by transforming a transformer input data sequence into a text sequence using a sequence-to-sequence transformer, the transformer input data sequence having been generated at least in part from the stored multivariate time series performance and / or diagnostic data, the generated text sequence including an indication of one or more of (a) a status of the light source system, (b) a maintenance action for the light source system, and (c) a monitoring action for the light source system.2. The light source system of clause 1, wherein the multivariate time series performance and / or diagnostic data from the light source system includes numerical data.3. The light source system of clause 1, wherein the controller is further configured to produce one or more stored generated text sequences by storing the generated text sequence as (a) the generated text sequence in a non-user-approved form and / or (b) the generated text sequence in a user-approved form and / or (c) the generated text sequence in a user-edited form.4. The light source system of clause 3, wherein the transformer input data sequence is generated from (1) the multivariate time series performance and / or diagnostic data, and (2) at least one previously produced stored generated text sequence.5. The light source system of clause 1, wherein the sequence-to-sequence transformer is a transformer trained, using as training inputs, sequences of data generated from multivariate time series data relating to performance and / or diagnostics of the light source system, and / or of one or more light source systems of the same type, and, using as training targets, time-correlated human-generated and / or human-approved texts relating to the status, maintenance, and / or monitoring of the respective light source system or systems.6. The light source system of clause 5, wherein the sequence-to-sequence transformer is a transformer trained using as training inputs (1) sequences of data generated from multi- variate time series data relating to performance and / or diagnostics of the light source system and / or of one or more light source systems of the same type, and (2) one or more time- and light source system- corelated humangenerated and / or human approved texts relating to the status, maintenance, and / or monitoring of the respective light source system or systems.7. The light source system of clause 1, wherein the controller is configured to generate the transformer input data sequence and include in the transformer input data sequence variable labels and associated variable values.8. The light source system of clause 1, wherein the controller is configured to generate the transformer input data and include in the transformer input data variable labels and associated variable values and associated time stamps.9. A process for automatically producing a text sequence relating to the status, maintenance, and / or monitoring of a light source system, the process including (1) forming a transformer input data sequence including data of a light source system by transforming multivariate time series data into string sequence data including variable labels and associated variable values and (2) producing a generated text sequence by machine-translating the transformer input data sequence into a text sequence relating to the status, maintenance and / or monitoring of the light source system using a sequence-to-sequence transformer.10. The process of clause 9, wherein the data of the light source system includes data relating to performance and / or diagnostics of the light source system.11. The process of clause 9, wherein transforming the multivariate time series data into string sequence data includes ordering and tokenizing the multi-variate time series data within a string sequence as a succession of values ordered by variable type.12. The process of clause 9, wherein transforming the multivariate time series data into string sequence data includes ordering and tokenizing the multivariate time series data within a string sequence as a succession of values ordered by time.13. The process of clause 9, further including producing one or more stored generated text sequences by storing the generated text sequence, and wherein forming a transformer input data sequence includes transforming and combining both (1) the multivariate time series data and (2) one or more previously stored text sequences.14. The process of clause 9, wherein the sequence-to-sequence transformer is trained using as inputs sequences of data of the light source system generated from multi-variate time series data relating to performance and / or diagnostics of the light source system, and / or of one or more light source systems of the same type as the light source system, and using as targets time-correlated human-generated and / or human-approved text relating to the status, maintenance, and / or monitoring of the respective light source system or systems.15. The process of clause 9, wherein the sequence-to-sequence transformer is trained using as inputs sequences of data of the light source system generated from both (1) stored multi- variate time series data relating to performance and / or diagnostics of the light source system, and / or of one or more light source systems of the same type and (2) past-time-correlated stored human-generated and / or human- approved texts relating to the status, maintenance, and / or monitoring of the respective light source system or systems, and using as targets current-time-correlated stored human-generated texts relating to the status, maintenance, and / or monitoring of the respective light source system or systems.16. The process of clause 9, wherein forming a sequence of data of the light source system includes transforming and tokenizing multi-variate time series data relating to performance and / or diagnostics of the light source system into string sequence data including variable labels with associated variable values.17. The process of clause 9, further including, in response to the text sequence containing one or more previously specified words or phrases, triggering and / or sending an alert relating to and / or containing the content or a portion of the content of the text sequence.18. The process of clause 9, further including, in response to the text sequence containing one or more previously specified words or phrases, automatically performing maintenance and / or an adjustment to the light source system.19. A process of training a sequence-to-sequence transformer for producing text sequences relating to the status, maintenance, and / or monitoring of a light source system, the process including: using as inputs at least sequences of data of the light source system generated from multi-variate time series data relating to performance and / or diagnostics of the light source system, and / or of one or more light source systems of the same type as the light source system, tokenized into string sequence data including variable labels with associated variable values and time stamps; and using as targets time- correlated human-generated and / or human-approved text relating to the status, maintenance, and / or monitoring of the respective light source system or systems.20. The process of clause 19, further using as inputs time-correlated human-generated texts relating to the status, maintenance, and / or monitoring of the respective light source system or systems.21. The light source system of clause 1, wherein the controller is configured to receive and store multivariate asynchronous time series performance and / or diagnostic data from the light source system and the transformer input data sequence has been generated at least in part from the stored multivariate asynchronous time series performance and / or diagnostic data.22. The process of clause 9, including transforming multivariate asynchronous time series data into string sequence data including variable labels and associated variable values.23. The process of claim 19, including using as inputs at least sequences of data of the light source system generated from multi-variate asynchronous time series data relating to performance and / or diagnostics of the light source system.

[0060] The above-described aspects and implementations and other implementations are within the scope of the following claims.

Claims

CLAIMS1. A light source system comprising: two or more light source system modules; a controller in communication with the two or more modules, the controller configured to (1) receive and store multivariate time series performance and / or diagnostic data from the light source system including data from the two or more modules and (2) periodically and / or in response to a triggering signal, produce a generated text sequence by transforming a transformer input data sequence into a text sequence using a sequence-to-sequence transformer, the transformer input data sequence having been generated at least in part from the stored multivariate time series performance and / or diagnostic data, the generated text sequence including an indication of one or more of (a) a status of the light source system, (b) a maintenance action for the light source system, and (c) a monitoring action for the light source system.

2. The light source system of claim 1, wherein the multivariate time series performance and / or diagnostic data from the light source system comprises numerical data.

3. The light source system of claim 1, wherein the controller is further configured to produce one or more stored generated text sequences by storing the generated text sequence as (a) the generated text sequence in a non-user-approved form and / or (b) the generated text sequence in a user-approved form and / or (c) the generated text sequence in a user-edited form.

4. The light source system of claim 3, wherein the transformer input data sequence is generated from (1) the multivariate time series performance and / or diagnostic data, and (2) at least one previously produced stored generated text sequence.

5. The light source system of claim 1, wherein the sequence-to-sequence transformer is a transformer trained, using as training inputs, sequences of data generated from multivariate time series data relating to performance and / or diagnostics of the light source system, and / or of one or more light source systems of the same type, and, using as training targets, time-correlated human-generated and / or human-approved texts relating to the status, maintenance, and / or monitoring of the respective light source system or systems.

6. The light source system of claim 5, wherein the sequence-to-sequence transformer is a transformer trained using as training inputs (1) sequences of data generated from multi- variate time series data relating to performance and / or diagnostics of the light source system and / or of one or more light source systems of the same type, and (2) one or more time- and light source system- corelated human-generated and / or human approved texts relating to the status, maintenance, and / or monitoring of the respective light source system or systems.

7. The light source system of claim 1, wherein the controller is configured to generate the transformer input data sequence and include in the transformer input data sequence variable labels and associated variable values.

8. The light source system of claim 1, wherein the controller is configured to generate the transformer input data and include in the transformer input data variable labels and associated variable values and associated time stamps.

9. A process for automatically producing a text sequence relating to the status, maintenance, and / or monitoring of a light source system, the process comprising (1) forming a transformer input data sequence including data of a light source system by transforming multivariate time series data into string sequence data including variable labels and associated variable values and (2) producing a generated text sequence by machine-translating the transformer input data sequence into a text sequence relating to the status, maintenance and / or monitoring of the light source system using a sequence-to-sequence transformer.

10. The process of claim 9, wherein the data of the light source system includes data relating to performance and / or diagnostics of the light source system.

11. The process of claim 9, wherein transforming the multivariate time series data into string sequence data comprises ordering and tokenizing the multivariate time series data within a string sequence as a succession of values ordered by variable type.

12. The process of claim 9, wherein transforming the multivariate time series data into string sequence data comprises ordering and tokenizing the multi-variate time series data within a string sequence as a succession of values ordered by time.

13. The process of claim 9, further comprising producing one or more stored generated text sequences by storing the generated text sequence, and wherein forming a transformer input data sequence includes transforming and combining both (1) the multivariate time series data and (2) one or more previously stored text sequences.

14. The process of claim 9, wherein the sequence-to-sequence transformer is trained using as inputs sequences of data of the light source system generated from multi-variate time series data relating toperformance and / or diagnostics of the light source system, and / or of one or more light source systems of the same type as the light source system, and using as targets time-correlated human-generated and / or human-approved text relating to the status, maintenance, and / or monitoring of the respective light source system or systems.

15. The process of claim 9, wherein the sequence-to-sequence transformer is trained using as inputs sequences of data of the light source system generated from both (1) stored multi- variate time series data relating to performance and / or diagnostics of the light source system, and / or of one or more light source systems of the same type and (2) past-time-correlated stored human-generated and / or human- approved texts relating to the status, maintenance, and / or monitoring of the respective light source system or systems, and using as targets current-time-correlated stored human-generated texts relating to the status, maintenance, and / or monitoring of the respective light source system or systems.

16. The process of claim 9, wherein forming a sequence of data of the light source system includes transforming and tokenizing multi-variate time series data relating to performance and / or diagnostics of the light source system into string sequence data including variable labels with associated variable values.

17. The process of claim 9, further comprising, in response to the text sequence containing one or more previously specified words or phrases, triggering and / or sending an alert relating to and / or containing the content or a portion of the content of the text sequence.

18. The process of claim 9, further comprising, in response to the text sequence containing one or more previously specified words or phrases, automatically performing maintenance and / or an adjustment to the light source system.

19. A process of training a sequence-to-sequence transformer for producing text sequences relating to the status, maintenance, and / or monitoring of a light source system, the process comprising: using as inputs at least sequences of data of the light source system generated from multivariate time series data relating to performance and / or diagnostics of the light source system, and / or of one or more light source systems of the same type as the light source system, tokenized into string sequence data including variable labels with associated variable values and time stamps; and using as targets time-correlated human-generated and / or human-approved text relating to the status, maintenance, and / or monitoring of the respective light source system or systems.

20. The process of claim 19, further using as inputs time-correlated human-generated texts relating to the status, maintenance, and / or monitoring of the respective light source system or systems.

Images