Process for maintaining modules of light sources in semiconductor manufacturing processes

Abstract

A computer-implemented process for maintaining a light source includes using a computer to: (1) form a plurality of pairs of performance parameters from a set of N performance parameters; (2) score each of the plurality of pairs to generate a pair score; and (3) sum the pair scores to generate a model score, for each of M machine learning models trained to each classify the status of a module of the light source as needing maintenance or not needing maintenance; implement the model having the highest score; and use the implemented model to iteratively classify the status of a particular module of a particular light source as needing maintenance or not needing maintenance over time; and perform maintenance on the particular module when indicated by the model, wherein M is an integer greater than or equal to 2 and N is an integer greater than or equal to 3.

Description

[Technical Field]

[0001] CROSS-REFERENCE TO RELATED APPLICATIONS

[0001] This application claims priority to U.S. Application No. 63 / 482,094, entitled "PROCESSES FOR MAINTENANCE OF MODULES OF LIGHT SOURCES IN SEMICONDUCTOR PHOTOLITHOGRAPHY," filed January 30, 2023, the entire contents of which are incorporated herein by reference. [Background technology]

[0002] The disclosed subject matter relates to the maintenance of light sources such as those used in integrated circuit photolithography manufacturing processes.

[0003]

[0003] Light, which may be laser radiation used in semiconductor photolithography, is typically supplied by a system called a light source. These light sources generate radiation as a train of pulses at a particular repetition rate, for example, in the range of about 500 Hz to about 6 kHz. Furthermore, the expected useful life of such a light source has traditionally been measured in terms of the number of pulses that can be predicted to occur before repair or replacement is required, typically expressed in billions of pulses.

[0004]

[0004] One system for generating light, such as laser radiation, at frequencies useful for semiconductor photolithography (e.g., deep ultraviolet (DUV) wavelengths) requires the use of a master oscillator power amplifier (MOPA) dual gas discharge chamber configuration. This configuration has two chambers: a master oscillator chamber (MO chamber) and a power amplifier chamber (PA chamber). These chambers and many other system components may be considered modules, and the entire light source may be considered a collection of modules. Each module generally has a lifetime that is shorter than the lifetime of the entire light source. Thus, throughout the lifetime of the light source, the health status of individual modules is evaluated to determine whether the module should be repaired or replaced, and the module is repaired or replaced according to such evaluation. Summary of the Invention

[0005]

[0005] In some general aspects, a computer-implemented process for maintaining a light source includes using a computer to, for each of M machine learning models, each trained to classify the state of a module of the light source as requiring maintenance or not requiring maintenance, and each having a set of N performance parameters representing the performance of the model, (1) form a plurality of pairs of performance parameters from the set of N performance parameters, (2) score each of the plurality of pairs to generate a pair score, and (3) sum the pair scores to generate a model score; implementing the model with the highest score; and using the implemented model to iteratively classify the state of a particular module of a particular light source as requiring maintenance or not requiring maintenance over time; and automatically or manually performing maintenance on the particular module in response to the implemented model classifying the particular module as requiring maintenance, wherein M is an integer greater than or equal to 2 and N is an integer greater than or equal to 3.

[0006]

[0006] Implementation of the computer-implemented process may include one or more of the following: The process may further include using a dataset to first generate M models and then training the M models; The process may further include repeating the process at some point after performing the process using a new dataset; The new dataset may include new data from a particular module; The new dataset may include new data from multiple modules and multiple light sources; The process may result in at least one of the N performance parameters being a performance parameter not used in generating and training the M models; Forming multiple pairs of performance parameters from the set of N performance parameters may include forming multiple pairs of performance parameters from performance parameters that are equal to or adjacent to each other in priority of the set of N performance parameters; Forming multiple pairs of performance parameters from the set of N performance parameters may include forming ordered pairs ordered according to the relative priority of each paired performance parameter; Forming multiple pairs of performance parameters from the set of N performance parameters may include forming N-1 pairs. Forming a plurality of pairs of performance parameters of the set of N performance parameters may include forming ordered pairs ordered according to the relative priority of each paired performance parameter.

[0007] Scoring each of the plurality of pairs may include, for each parameter, normalizing the parameter to the normal range and to the parameter target by adjusting the range of each parameter to match the normal range and / or shifting the range of each parameter as needed and by subtracting each parameter from the target for that parameter, and for each pair, taking the square root of the sum of the squares of the normalized parameters of the pair. Normalizing each parameter may include, for parameters in the form of a percentage, subtracting the parameter from the target percentage for that parameter, where the result of the subtraction is the normalized parameter, and, for parameters not in the form of a percentage, compressing, expanding, and / or shifting the range of the parameter to a value within the range of 0 to 1 and subtracting the parameter from the target value for that parameter. Scoring each of the plurality of pairs may further include, for each pair, multiplying the square root of each pair by Cl if both parameters are equal to or better than their respective targets, by negative C2 (-C2) if both parameters are worse than their respective targets, by negative C3 (-C3) if only the higher-priority parameter of the pair is worse than its target, and by negative C4 (-C4) if only the lower-priority parameter of the pair is worse than its target, where C2>C3>C4>0 and Cl>0. In the above process, the coefficients may have values ​​of Cl=C4=1, C2=0.75, and C3=0.5. Summing the pair scores to generate a score for each model may include a weighted sum.

[0008]

[0008] One or more implementations are described in detail in the accompanying drawings and the description below. Other features will be apparent from the following description and drawings, and from the claims. [Brief explanation of the drawings]

[0009] [Figure 1]

[0009] FIG. 1 is a schematic cross-sectional view showing an embodiment of a light source. [Figure 2]

[0010] 2 is a schematic diagram of a light source, such as the light source of FIG. 1, together with a lithographic exposure apparatus. [Figure 3]

[0011] 1 is a flowchart illustrating aspects of a model selection process. [Figure 4]

[0012] 1 is a flowchart illustrating aspects of an automatic model selection process. [Figure 5A]

[0013] FIG. 5 is a diagram of an implementation of a scoring process, such as the scoring process in the process of FIG. 4. [Figure 5B] FIG. 5 is a diagram of an implementation of a scoring process, such as the scoring process in the process of FIG. 4. [Figure 5C] FIG. 5 is a diagram of an implementation of a scoring process, such as the scoring process in the process of FIG. 4. [Figure 5D] FIG. 5 is a diagram of an implementation of a scoring process, such as the scoring process in the process of FIG. 4. [Figure 6]

[0014] FIG. 5 is a diagram of an implementation of a scoring process, such as the scoring process in the process of FIG. 4. [Figure 7]

[0015] 5 is a flowchart of an implementation of a scoring process, such as the scoring process in the process of FIG. 4. [Figure 8]

[0016] 8 is a flowchart of an implementation of a portion of a scoring process, such as the portion of the scoring process of FIG. 7. DETAILED DESCRIPTION OF THE INVENTION

[0010]

[0017] 1 , light source 100, which may be a deep ultraviolet (DUV) light source 100, may take the form of a dual-stage pulsed light source that generates pulse-amplified light beam 105 as light beam 105. 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.

[0011]

[0018] MO system 160 may include, for example, MO chamber module 161, in which an electrical discharge between electrodes (not shown) generates a laser gas discharge in the laser gas, creating an inverted population of energetic molecules, such as argon, krypton, or xenon, which is then narrowed to a relatively very narrow bandwidth in line narrowing module (“LNM”) 162 to generate relatively broadband radiation with a selected center wavelength. MO system 160 may also include MO output coupler (MO OC) 162, which may include a partially reflective mirror that, together with a reflective grating (not shown) in LNM 162, forms an oscillator cavity in which MO system 160 oscillates to form a seed output pulse, thereby forming a master oscillator. MO system 160 may also include line center analysis module (LAM) 163. As described above, LAM 180 includes, for example, an etalon spectrometer for fine wavelength measurement and a grating spectrometer for coarser resolution.

[0012]

[0019] The relay optics 170 may include an MO wavefront engineering box (WEB) 171 that functions to redirect the output of the MO system 160 towards the PA system 165, and may include, for example, beam expansion by a multi-prism beam expander (not shown) or coherence destruction, for example in the form of an optical delay path (not shown).

[0013]

[0020] The PA system 165 includes a PRA chamber module 166, which is also an oscillator formed, for example, by injection of the output light beam from the MO system 160 and output coupling optics (not shown) that may be incorporated into a PRAWEB 167 and redirected through a gain medium in the chamber 166 via a beam reverser 168. The PRAWEB 167 may incorporate a partially reflective input / output coupler (not shown), a maximum reflecting mirror for the nominal operating wavelength (which may be approximately 193 nm for an ArF system), and one or more prisms. The PA system 165 optically amplifies the output light beam from the MO system 160.

[0014]

[0021] The optical output subsystem 175 may include a bandwidth analysis module (BAM) 176 at the output of the PA system 165, which receives the output light beam pulses from the PA system 165 and extracts a portion of the light beam for metrology purposes, such as measuring the output bandwidth and pulse energy. The output light beam pulses then pass through an optical pulse stretcher module (OPuS) 177 and an output composite automatic shutter metering module (CASMM) 178. This module may also be the location of a pulse energy meter. One purpose of the OPuS 177 is to convert a single output pulse into a pulse train. Secondary pulses created from the original single output pulse may be delayed relative to each other. By distributing the original laser pulse energy among the train of secondary pulses, the effective pulse length of the light beam may be extended while simultaneously reducing the peak pulse intensity.

[0015]

[0022] The light source 100 is comprised of multiple modules. Each component of the light source 100 (MO chamber 161, LNM 162, MOWEB 171, PRA chamber 166, PRAWEB 167, OPuS 177, BAM 176, etc.) is a module. The overall availability of the light source 100 is a direct result of the availability of each of the individual modules that make up the light source 100. That is, if all of the modules that make up the light source 100 are unavailable, the light source 100 is unavailable. The sensing and evaluation device 120 monitors these modules so that they may be refreshed or replaced before they fail, thereby maintaining operation of the light source 100 and optimizing and increasing productivity of the output device 210. The sensing and evaluation device 120 provides maintenance alerts that can be used to perform automated and / or manual maintenance tasks on one or more specific modules, including replacement tasks when necessary to prevent failure of the light source 100.

[0016]

[0023] Referring to FIG. 2, the amplified light beam 105 (FIG. 1) may 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 positioned on a wafer table 212 connected to a positioner constructed to hold the wafer 211 and configured to precisely position the wafer 211 according to specific parameters. The light beam 205 may have a wavelength in the deep ultraviolet (DUV) range, and may include wavelengths from about 100 nanometers (nm) to about 400 nm. For example, a light source 100 generating such a light beam 105, 205 may be a gas discharge light source, such as an excimer light source, or an excimer laser that uses a combination of one or more noble gases, which may include argon, krypton, or xenon, and a reactive gas, which may include fluorine or chlorine, as a gain medium. The light source 100 may be an excimer light source. Thus, for example, the gain medium may include argon fluoride (ArF), krypton fluoride (KrF), or xenon chloride (XeCl). When the gain medium includes argon fluoride, the wavelength of the amplified light beam 205 is approximately 193 nm, and when the gain medium includes krypton fluoride, the wavelength of the amplified light beam 205 is approximately 248 nm. The size of the microelectronic features patterned on the wafer 211 depends on the wavelength of the light beam 205, with lower wavelengths resulting in smaller minimum feature sizes. When the wavelength of the light beam 205 is 248 nm or 193 nm, the minimum size of the microelectronic features may be, for example, 50 nm or less. The bandwidth of the light beam 205 may be the actual instantaneous bandwidth of the optical spectrum (or emission spectrum), which contains information about how the optical energy of the light beam 205 is distributed across various wavelengths.

[0017]

[0024] The photolithography exposure apparatus 210 includes an optical arrangement having, for example, one or more condenser lenses, a mask, and an objective lens arrangement. The mask is movable along one or more directions, such as along the optical axis of the light beam 205 or in a plane perpendicular to the optical axis. The objective lens arrangement includes a projection lens and enables image transfer from the mask to the photoresist on the wafer 211. The photolithography exposure apparatus 210 also includes an illumination system that adjusts the angular range of the light beam 205 that strikes the mask. The illumination system also homogenizes (uniforms) the intensity distribution of the light beam 205 across the mask.

[0018]

[0025] The photolithography exposure apparatus 210 may also include a lithography controller 213 that controls, among other features, how layers are printed on the wafer 211. The lithography controller 213 includes a memory that stores information such as a process recipe. The process program or recipe determines the length of exposure on the wafer 211, the mask to be used, and other factors that affect the exposure. During lithography, multiple pulses of the light beam 205 illuminate the same area of ​​the wafer 211 and together make up an illumination dose.

[0019]

[0026] The quality of features produced on wafer 211 by photolithography exposure tool 210 depends directly on the quality and reliability of the light pulses from light source 100. Pulses with lower than desired power can lead to under-exposure of areas of wafer 211. Missing pulses can similarly lead to under-exposure. Shifts in the wavelength or bandwidth distribution can shift the position of the image and change the pattern produced on wafer 211.

[0020]

[0027] Information from various sources within light source 100 can be used to assess the need for maintenance or replacement of modules of light source 100, such as MO chamber module 161 and PRA chamber module 166. These may require gas replacement or even complete module replacement, for example, if the module's performance (and thus the light source's performance) begins to degrade.

[0021]

[0028] It is important to perform maintenance such as gas changes and module replacements in a timely manner. Maintenance performed too early reduces the percentage of time the light source is available for productive operation. Maintenance performed too late can lead to improperly processed products, which can have very high sunk production costs and may even have to be discarded.

[0022]

[0029] Machine learning can be applied to determine the appropriate time to perform maintenance. Data from various parts of the light source, and possibly also data from the associated lithography apparatus, can be used by the machine learning model to determine the optimal time to maintain the light source module. Machine learning and the execution of the machine learning model can be performed by various computing hardware. In various implementations, a computing module configured to perform machine learning and / or execute the machine learning module can include an electronic processor, electronic storage, and an I / O interface. The electronic processor includes one or more processors suitable for executing computer programs, such as general-purpose or special-purpose microprocessors, and any one or more processors of any type of digital computer. Generally, the electronic processor receives instructions and data from read-only memory, random-access memory, or both. The electronic storage can be volatile memory, such as RAM, or non-volatile memory. The electronic storage can store data and information used to train or execute models, components of the control system, and / or other data or instructions. The information can be stored, for example, in a lookup table or database.

[0023]

[0030] A two-class machine learning classification model can be generated and trained on data generated from the operation of existing light sources used in manufacturing and / or from the operation of prototype light sources or new light source designs during program testing. Standard machine learning techniques can be used to train the classification model to distinguish between two classes or conditions: maintenance-required and non-maintenance-required. A maintenance-required condition can be defined, for example, as a condition in which a specific fault or a specific out-of-specification condition of the module occurs (in the training data set) within a specific number of light pulses of the light source or within a specific time period. Similarly, a non-maintenance-required condition can be defined, for example, as a condition in which a specific fault or a specific out-of-specification condition of the module does not occur (in the training data set) within a specific time period. Optionally, additional artificial fault or out-of-specification instances can be generated, such as by extrapolating from existing data, and added to the training data set. This can help improve the performance of the trained model, as two-class classification models generally perform best when trained with approximately equal numbers of instances of each class (maintenance-required and non-maintenance-required).

[0024]

[0031] Using machine learning tools for various types of models (including classification and regression trees (CART), random forests, feed-forward neural networks, etc., and combinations thereof), multiple models can be automatically generated and trained using a training dataset. Multiple models can be ranked using any of a variety of statistics, such as: (1) the true positive (TP) rate, which is the proportion of "true positives" among "positives," where a "positive" is a classification (decision) by the model that requires maintenance; true "Positive" is based on the model that maintenance is necessary. correct(1) classification (decision) (i.e., a classification [decision] that maintenance is needed when maintenance was or will actually be needed within the training dataset, within a specific number of light pulses, or within a specific time), and (2) true negative (TN) rate, which is the proportion of "negatives" (classifications / decisions that "no maintenance is needed") that are true (i.e., "correct") negatives. Next, we need a way to select which model to implement when performing maintenance on a light source module.

[0025]

[0032] Further complications may arise when business or engineering needs or other considerations may make it desirable to include factors ("data") in the model selection process that are not used in the model training process. Such data may take the form of data related to new or changed metrics used or to be used in managing light sources, metrics that cannot be easily represented in a classification model, or a combination of these. For example, the number of light pulses from a light source that are "missed" (i.e., not generated) between the time it is classified as "maintenance required" and the actual or predicted failure of the light source, minus some safety interval, may be referred to as "lost pulses." The lost pulse metric may not be easily represented within a classification model or may be a new metric or a metric that changes from time to time, and therefore may not be a metric relevant to model training.

[0026]

[0033] 3 and 4 show flowcharts illustrating aspects of a process that may be performed automatically to select a model from among multiple models based on both parameters such as TP% and TN% that are included in or specific to the model training process, and parameters such as missing pulses that may not be used in the model training process.

[0027]

[0034] FIG. 3 shows a preliminary process 320 that includes at least two steps. The first step is to collect requirements in the form of parameters to be used in evaluating the model (and its performance in use for maintenance purposes) (321). The parameters may be, for example, TP%, TN%, or lost pulses (LP). Collecting requirements includes collecting target values ​​for each parameter and collecting or assigning and applying a utility function or priority order for the parameters. The relative priority of the goals and parameters and / or associated targets may be determined by the associated production process and / or the business goals and requirements associated therewith. For example, LP may be more important than TP%, which may be more important than TN%. The goals may be specified in terms of number of pulses (e.g., 20,000) or time (e.g., 2 days or 7 days), with TP% and TN% specified as percentages (e.g., 60% and 95%, respectively). The second step is to select a pairing method and, optionally, a weighting scheme (322). Performance parameters may be paired by forming pairs of performance parameters that are adjacent to each other in priority order. For example, if the performance parameters are TP%, TN%, and LP, and the priorities from highest to lowest are LP, TP, and TN, pairs may be formed as (LP, TP) and (TP, TN), resulting in N-1 pairs for N parameters. Alternatively, one additional pair may be formed from the lowest and highest ranking performance parameters, such as (LP, TN), and included in the total pairs, resulting in N pairs for N performance parameters. Other suitable pairing methods, such as methods that generate other total pairs, may be used as needed. The pairs may take the form of ordered pairs, ordered according to the relative priority of each paired performance parameter. Once the performance parameters have been identified and ranked and a pairing method has been selected, an automated process, such as the automated process shown in the flowchart of FIG. 4, may be used.

[0028]

[0035] FIG. 4 is a flowchart of a computer-implemented automated process 430 for maintaining a light source. After multiple models are generated for evaluation (431, a total of M models), the performance parameters (of the set of N performance parameters) for each of the M models are normalized and paired, each pair is scored to generate a "paired score," and then the paired scores are summed to generate a model score (432). The model with the highest score is then selected (433) and implemented (i.e., its recipe is implemented) for use in maintaining at least a specific module of a specific light source (434), resulting in a "implemented module." The implemented model is then used to repeatedly classify specific modules over time. If the implemented module classifies a specific module as requiring maintenance (435, "Yes" branch), the necessary maintenance action is taken (436). The action can be performed automatically. The process can then return to repeatedly classifying specific modules (435). If the classification is "no maintenance required" (435, "no" branch), the process may return (435) to reclassify the particular module. If desired, a time delay (not shown) may be implemented between repeated classification steps. After at least some of the repeated classification steps (435) and / or performing required maintenance actions (436) (represented by dashed arrows), or after each instance of these steps (represented by solid arrows that do not follow the dashed arrows), an evaluation (437) may be performed to determine whether a model update is appropriate. For example, a model update may be appropriate after a certain number of classifications (435), or after a certain number of maintenance actions (436), or after a certain amount of time has passed, or after a certain amount of new data has been collected from the operation of the particular light source and module, and / or additional similar light sources and modules, making a new data set (including or consisting entirely of new data) available for model development and training, or based on a combination of these or other factors. If it is determined that updating the model is appropriate (437, "Yes" branch), the process begins again to generate multiple models for evaluation (431).

[0029]

[0036] For each performance parameter, a parameter may be better than the target or goal for that parameter if it is numerically greater than the target, such as a bowling score. Alternatively, a performance parameter may be better than the target or goal for that parameter if it is numerically less than the target, such as a golf score. TP% and TN% are, for example, better if they are greater than their respective target values, while lost pulses (LP) are better if they are less than the target because LP should be minimized.

[0030]

[0037] For a pair of performance parameters P1 and P2, where P1 is the higher-priority parameter of the two and is represented on the vertical axis of the grid, and P2 is represented on the horizontal axis of the grid, with the origin of the grid representing the targets for both parameters, the results shown in Figures 5A-5D are generated as follows: If both P1 and P2 are better when they are greater than their targets (which may be referred to as high P1 / high P2 or simply high / high), as in Figure 5A, the preferred quadrant for plotting the pair is the upper right, or quadrant I. If P1 is better when it is higher than its target and P2 is better when it is lower than its target (high / low), then we have the situation shown in Figure 5B, where the "good-good" quadrant is the upper left (quadrant II). If both P1 and P2 are better than their targets when they are lower than their targets (low / low), then we have the result shown in Figure 5C, where both P1 and P2 are "good" in the lower left (quadrant III) of the grid, and therefore this quadrant is the good / good quadrant, as shown. If P1 is better when it is below target and P2 is better when it is above target (low / high), then we have the situation shown in Figure 5D, where the "good-good" quadrant is in the lower right (quadrant IV). The classifications shown in Figures 5A-5D can be used in scoring pairs of performance parameters for a given model, as described below with reference to Figures 6 and 7.

[0031]

[0038] FIG. 6 shows a grid corresponding to that of FIG. 5A and illustrates part of the scoring process for high / high pairs (pairs that are better when both parameters are greater than their targets). The scoring method for pair P1, P2 may involve determining the radius R from the grid origin to point (P2, P1) (in other words, determining the square root of the sum of the squares of P1 and P2), and then multiplying by a factor depending on the quadrant in which point P is located. For example, for P in quadrant 1 (or on an axis), R may be multiplied by 1. For P in quadrant III, R may be multiplied by −1 (because both parameters are worse than the targets). For P in quadrant II, R may be multiplied by −0.5, and for P in quadrant IV, R may be multiplied by −0.75. This is because a high-priority parameter (P1) being worse than its target (as in quadrant III) is considered worse than a low-priority parameter (P2) being worse than its target (as in quadrant II). The flowchart in Figure 7 shows the process of Figure 6 extended to all combinations of performance parameters within pairs, as well as further details of the scoring process.

[0032]

[0039] 7 is a flowchart of a process or sub-process 732 for scoring a model having a set of N performance parameters grouped into performance parameter pairs. Process 732 may be used, for example, as a step or process 432 within process 430 of FIG. 4. First, for a given model, for a first pair of performance parameters P1, P2, the parameters are normalized and R is set to the square root of the sum of the squares of P1 and P2 (741). Normalizing each parameter may involve, for example, for parameters in the form of a percentage, subtracting the parameter from a target percentage for the parameter, with the result of this subtraction being the normalized parameter, and for parameters not in the form of a percentage, compressing, expanding, and / or shifting the range of the parameter to a value within the range of 0 to 1 and subtracting the parameter from its target value. Next, depending on the classification of P1, P2 as high / high (742 "Yes" branch), high / low (743 "Yes" branch), low / low (744 "Yes" branch), or low / high (744 "No" branch), and depending on the quadrant of the points (P2, P1) listed in factor selection tables 745a, 745b, 745c, and 745d (see Figures 5A-5D), the previously obtained value of R is multiplied in each "Quadrant" column by a coefficient shown in each "Score S" column of factor selection tables 745a-745d. ("S" represents the score of the ith pair of scored pairs.) The individual scores S generated in this first part 748 of process 740 may be summed directly (solid arrow to 752), or, optionally, individual coefficients ωi-ω for each pair score S (shown in steps 751-1 to 751-Ptot) are calculated. Ptot4. The model scores may be scaled by (dashed arrow to 750). This process is repeated (753 "yes" branch) until there are no more pairs (out of the total number of pairs "Ptot") to score (753 "no" branch). The sum of the pair scores is then stored as the score for each model (754). If there are more models to score (755 "yes" branch), the above is repeated until each of the M models has been scored. Process 732, which can be used as step or process 432 in FIG. 4, may then proceed to the next step, such as step 433 in FIG. 4.

[0033]

[0040] A somewhat generalized and simplified version of the first portion 748 of process or sub-process 740 is shown in Figure 8 in the form of process or sub-process 848g. Process or sub-process 848g sorts each pair of performance parameters (at 856a, 856b, and 856c) using constants C1, C2, C3, and C4 that satisfy the conditions shown in (855) (i.e., C2 > C3 > C4 > 0 and C1 > 0) depending on which member of the pair is better or worse than its respective target. If both parameters are better than their respective targets (856a "Yes" branch), R is multiplied by C1. If both parameters are worse than their respective targets (856b "Yes" branch), R is multiplied by negative C2 (-C2). If only the higher priority parameter of the pair is worse than its target (856c "Yes" branch), R is multiplied by negative C3 (-C3); otherwise (if only the lower priority parameter of the pair is worse than its target), R is multiplied by negative C4 (-C4). The resulting pair scores may be summed directly, as in step 752 of process 740 of Figure 7, or may be scaled and then summed, as in process or subprocess 750 of Figure 7. In an implementation consistent with the coefficients of Figure 7, C1 = C4 = 1, C2 = 0.75, and C3 = 0.5.

[0034]

[0041] Aspects and implementations of the present disclosure can be further described using the following clauses. 1. A computer-implemented process for maintaining a light source, comprising: Using a computer, For each of M machine learning models, each trained to classify the status of a module of a light source as requiring maintenance or not requiring maintenance, and each having a set of N performance parameters that describe the performance of the model: forming a plurality of pairs of performance parameters from the set of N performance parameters; Scoring each of the plurality of pairs to generate a pair score; summing the pair scores to generate a model score; and Implementing the model with the highest score and using the implemented model to iteratively classify the status of a particular module of a particular light source as needing maintenance or not needing maintenance over time; automatically or manually performing maintenance on a particular module in response to the implemented model classifying the particular module as requiring maintenance; Including, The method wherein M is an integer greater than or equal to 2 and N is an integer greater than or equal to 3. 2. The computer-implemented process of clause 1, further comprising: first generating M models and then training the M models using the dataset. 3. The computer-implemented process of clause 2, further comprising, at some point after performing the process, repeating the process using a new data set. 4. The computer-implemented process of clause 3, wherein the new dataset includes new data from a particular module. 5. The computer-implemented process of clause 4, wherein the new data set includes new data from multiple modules and multiple light sources. 6. The computer-implemented process of clause 2, wherein at least one of the N performance parameters is a performance parameter that is not used in generating and training the M models. 7. The computer-implemented process of clause 1, wherein forming multiple pairs of performance parameters from the set of N performance parameters includes forming multiple pairs of performance parameters from performance parameters that are equal to or adjacent to each other in priority in the set of N performance parameters. 8. The computer-implemented process of clause 7, wherein forming a plurality of pairs of performance parameters from the set of N performance parameters includes forming ordered pairs ordered according to the relative priority of each performance parameter being paired. 9. The computer-implemented process of clause 7, wherein forming a plurality of pairs of performance parameters from the set of N performance parameters includes forming N-1 pairs. 10. The computer-implemented process of clause 1, wherein forming multiple pairs of performance parameters from the set of N performance parameters includes forming ordered pairs ordered according to the relative priority of each performance parameter being paired. 11. Scoring each of the plurality of pairs comprises: for each parameter, normalizing the parameter to the standard range and to the parameter's target by adjusting the range of each parameter to match the standard range and / or by shifting the range of each parameter as needed and subtracting each parameter from the target for that parameter; for each pair, taking the square root of the sum of the squares of the normalized parameters of the pair; 2. The computer-implemented process of clause 1, comprising: 12. Normalizing each parameter is For a parameter in the form of a percentage, subtracting the parameter from a target percentage for that parameter, the result of the subtraction being a normalized parameter; 12. The computer-implemented process of claim 11, comprising: for parameters not in the form of a percentage, compressing, expanding, and / or shifting the range of the parameter to a value within the range of 0 to 1 and subtracting the parameter from the target value for that parameter. 13. Scoring each of the plurality of pairs comprises: For each pair, the square root of each pair, Cl if both parameters are equal to or better than their respective targets, If both parameters are worse than their respective targets, then negative C2 (-C2) Negative C3 (-C3) if only the higher priority parameter of the pair is worse than its target, further including multiplying by negative C4 (-C4) if only the lower priority parameter of the pair is worse than its target; 12. The computer-implemented process of clause 11, wherein C2>C3>C4>0 and Cl>0. 14. The computer-implemented process of clause 13, wherein C1=C4=1, C2=0.75, and C3=0.5. 15. The computer-implemented process of clause 1, wherein summing the paired scores to generate a score for each model comprises a weighted sum. 16. A DUV light source configured to perform the computer-implemented process of any one of clauses 1-15. 17. A module configured to be installed within a DUV light source and configured to perform the computer-implemented process of any one of clauses 1 to 15. 18. A computing system configured to communicate with a DUV light source and configured to perform the computer-implemented process of any one of clauses 1-15.

[0035]

[0042] The above aspects and implementations, as well as other implementations, are within the scope of the following claims.

Claims

1. 1. A computer-implemented process for maintaining a light source, comprising: Using a computer For each of M machine learning models, each trained to classify the status of a module of a light source as requiring maintenance or not requiring maintenance, and each having a set of N performance parameters that describe the performance of the model: forming a plurality of pairs of performance parameters from the set of N performance parameters; scoring each of the plurality of pairs to generate a pair score; and summing the paired scores to generate a model score; implementing the model with the highest score and using the implemented model to iteratively classify the status of a particular module of a particular light source as needing maintenance or not needing maintenance over time; performing automatic or manual maintenance on the particular module in response to the implemented model classifying the particular module as requiring maintenance; Including, The method, wherein M is an integer greater than or equal to 2 and N is an integer greater than or equal to 3.

2. The computer-implemented process of claim 1 , further comprising first generating the M models and then training the M models using a dataset.

3. The computer-implemented process of claim 2 , further comprising, at some point after performing said process, repeating said process using a new data set.

4. The computer-implemented process of claim 3 , wherein the new data set includes new data from the particular module.

5. The computer-implemented process of claim 4 , wherein the new data set includes new data from multiple modules and multiple light sources.

6. The computer-implemented process of claim 2 , wherein at least one of the N performance parameters is a performance parameter that is not used in generating and training the M models.

7. 2. The computer-implemented process of claim 1, wherein forming a plurality of pairs of performance parameters from the set of N performance parameters comprises forming a plurality of pairs of performance parameters from performance parameters that are equal to or adjacent to each other in priority from the set of N performance parameters.

8. 8. The computer-implemented process of claim 7, wherein forming a plurality of pairs of performance parameters from the set of N performance parameters comprises forming ordered pairs ordered according to a relative priority of each paired performance parameter.

9. 8. The computer-implemented process of claim 7, wherein forming a plurality of pairs of performance parameters in the set of N performance parameters comprises forming N-1 pairs.

10. 2. The computer-implemented process of claim 1, wherein forming a plurality of pairs of performance parameters from the set of N performance parameters comprises forming ordered pairs ordered according to a relative priority of each paired performance parameter.

11. Scoring each of the plurality of pairs comprises: for each parameter, normalizing the parameter to the standard range and to the target for the parameter by adjusting the range of each parameter to match the standard range and / or by shifting the range of each parameter as needed and subtracting each parameter from the target for that parameter; for each pair, taking the square root of the sum of the squares of the normalized parameters of the pair; The computer-implemented process of claim 1 , comprising:

12. Normalizing each parameter is For a parameter in the form of a percentage, subtracting the parameter from a target percentage for that parameter, the result of the subtraction being the normalized parameter; and 12. The computer-implemented process of claim 11, comprising: for parameters not in the form of a percentage, compressing, expanding, and / or shifting the range of the parameter to a value within the range of 0 to 1 and subtracting the parameter from a target value for that parameter.

13. Scoring each of the plurality of pairs comprises: For each pair, the square root of each pair, Cl if both parameters are equal to or better than their respective targets, negative C2 (-C2) if both parameters are worse than their respective targets, negative C3 (-C3) if only the higher priority parameter of the pair is worse than its target; further comprising multiplying by negative C4 (-C4) if only the lower priority parameter of the pair is worse than its target; 12. The computer-implemented process of claim 11, wherein C2>C3>C4>0 and Cl>0.

14. 14. The computer-implemented process of claim 13, wherein C1 = C4 = 1, C2 = 0.75, and C3 = 0.

5.

15. The computer-implemented process of claim 1 , wherein summing the paired scores to generate a score for each model comprises a weighted sum.

16. 1. A deep ultraviolet (DUV) light source, comprising: a processor; a memory that, when executed by the processor, For each of M machine learning models, each trained to classify the status of a module of a light source as requiring maintenance or not requiring maintenance, and each having a set of N performance parameters that describe the performance of the model: forming a plurality of pairs of performance parameters from the set of N performance parameters; scoring each of the plurality of pairs to generate a pair score; and summing the paired scores to generate a model score; implementing the model with the highest score and using the implemented model to iteratively classify the status of a particular module of a particular light source as needing maintenance or not needing maintenance over time; performing automatic or manual maintenance on the particular module in response to the implemented model classifying the particular module as requiring maintenance; a memory including machine-executable instructions configured to perform actions including: Including, A DUV light source, wherein M is an integer greater than or equal to 2 and N is an integer greater than or equal to 3.

17. 17. The DUV light source of claim 16, wherein the acts further include using a data set to first generate the M models and then train the M models.

18. 20. The DUV light source of claim 17, wherein the actions further include repeating the process at some point after performing the process using a new data set.

19. 20. The DUV light source of claim 18, wherein the new data set includes new data from the particular module.

20. 20. The DUV light source of claim 19, wherein the new data set includes new data from multiple modules and multiple light sources.

21. 18. The DUV light source of claim 17, wherein at least one of the N performance parameters is a performance parameter that is not used in generating and training the M models.

22. 17. The DUV light source of claim 16, wherein forming a plurality of pairs of performance parameters from the set of N performance parameters comprises forming a plurality of pairs of performance parameters from performance parameters that are equal to or adjacent to each other in priority in the set of N performance parameters.

23. 23. The DUV light source of claim 22, wherein forming a plurality of pairs of performance parameters of the set of N performance parameters comprises forming ordered pairs ordered according to a relative priority of each paired performance parameter.

24. 23. The DUV light source of claim 22, wherein forming a plurality of pairs of performance parameters of the set of N performance parameters comprises forming N-1 pairs.

25. 17. The DUV light source of claim 16, wherein forming a plurality of pairs of performance parameters of the set of N performance parameters comprises forming ordered pairs ordered according to a relative priority of each paired performance parameter.

26. Scoring each of the plurality of pairs comprises: for each parameter, normalizing the parameter to the standard range and to the target for the parameter by adjusting the range of each parameter to match the standard range and / or by shifting the range of each parameter as needed and subtracting each parameter from the target for that parameter; for each pair, taking the square root of the sum of the squares of the normalized parameters of the pair; 17. The DUV light source of claim 16, comprising:

27. Normalizing each parameter is For a parameter in the form of a percentage, subtracting the parameter from a target percentage for that parameter, the result of the subtraction being the normalized parameter; and 27. The DUV light source of claim 26, further comprising: for parameters not in the form of a percentage, compressing, expanding, and / or shifting the range of the parameter to a value within the range of 0 to 1 and subtracting the parameter from its target value.

28. Scoring each of the plurality of pairs comprises: For each pair, the square root of each pair, Cl if both parameters are equal to or better than their respective targets, negative C2 (-C2) if both parameters are worse than their respective targets, negative C3 (-C3) if only the higher priority parameter of the pair is worse than its target; further comprising multiplying by negative C4 (-C4) if only the lower priority parameter of the pair is worse than its target; 27. The DUV light source of claim 26, wherein C2>C3>C4>0 and Cl>0.

29. 29. The DUV light source of claim 28, wherein C1 = C4 = 1, C2 = 0.75, and C3 = 0.

5.

30. The DUV light source of claim 16 , wherein summing the pair scores to generate a score for each model comprises a weighted sum.

Images