Predictive device in a gas discharge light source

Abstract

An apparatus comprises a decision module configured to receive a performance metric relating to a performance condition of an optical system of a light beam, estimate effectiveness of a proposed change to the optical system based on the performance metric and a predetermined learning model, and output a change command to the optical system if the proposed change to the optical system is estimated to be effective.

Description

[0001] CROSS-REFERENCE TO RELATED APPLICATIONS

[0002] This application claims priority to U.S. Application No. 62 / 949,723, filed on December 18, 2019, entitled “PREDICTIVE APPARATUS IN A GASDISCHARGE LIGHT SOURCE,” which is incorporated herein by reference in its entirety. Technical Field

[0003] The present disclosure relates to an apparatus using a predictive model that predicts whether a change to an optical source of an optical lithography system will improve an operating condition of the optical source. Background Art

[0004] A gas discharge light source used in photolithography is called an excimer light source or laser. Typically, an excimer laser uses a combination of one or more rare gases and reactive gases, the rare gases may include argon, krypton or xenon, and the reactive gases may include fluorine or chlorine. An excimer laser can produce excimers (pseudo-molecules) under appropriate conditions of electrical simulation (providing energy) and high pressure (of the gas mixture), which exist only in an excited state. The excimers in the excited state produce amplified light in the ultraviolet range. An excimer light source can use a single gas discharge chamber or multiple gas discharge chambers. When the excimer light source is performed, the excimer light source produces a deep ultraviolet (DUV) beam. The DUV light can have a wavelength of, for example, about 100 nanometers (nm) to about 400 nm.

[0005] The DUV beam can be directed to a photolithography exposure apparatus or scanner, a machine that applies the desired pattern to a target portion of a substrate (e.g., a silicon wafer). The DUV beam interacts with a projection optical system that projects the DUV beam through a mask onto the photoresist of the wafer. In this way, one or more layers of the chip pattern are patterned onto the photoresist, and the wafer is subsequently etched and cleaned. Summary of the Invention

[0006] In some general aspects, an apparatus includes a decision module configured to: receive a performance metric related to a performance condition of an optical system that emits a light beam; estimate the effectiveness of a proposed change to the optical system based on the performance metric and a predetermined learning model; and output a change command to the optical system if the proposed change to the optical system is estimated to be effective.

[0007] Implementations may include one or more of the following features.For example, the decision module may be configured to output a maintain command to the optical system if the decision module estimates that a proposed change to the optical system is not valid.

[0008] The decision module can be configured to estimate the effectiveness of the proposed change by determining whether the performance condition of the beam is improved. The decision module can determine whether the performance condition of the beam is improved by determining whether the error rate of the beam quality is reduced.

[0009] The performance condition can include one or more of the following: a type of beam quality error of the beam, a number of discharge events occurring in the gas mixture over a period of time, one or more faults associated with the error of the beam quality of the beam, the beam quality of the beam, and the error of the beam quality of the beam. The performance condition can include a configuration of the optical system including a configuration related to one or more chambers containing the gas mixture, such as a pressure, a temperature, a setting, or an operational mode associated with one or more chambers containing the gas mixture. The performance condition can include error events associated with operational parameters or characteristics of the beam, an analysis of each error event with respect to a set of fault labels, and a change in the configuration of the optical system.

[0010] The decision module can be configured to estimate the effectiveness of the proposed change before implementing the change to the optical system.

[0011] The apparatus can include an interface module in communication with the decision module, the interface module providing performance metrics. The interface module includes a plurality of analysis sub-modules. The plurality of analysis sub-modules includes a beam quality detection sub-module including one or more spectral feature detection modules each configured to detect an error of a respective spectral feature of the beam and produce an error event signal indicative of the respective spectral feature error of the beam, and an energy detection module configured to detect an error of an energy of the beam and produce an error event signal indicative of the energy error of the beam. The spectral features can include a bandwidth or a wavelength of the beam, and the performance metrics can be produced based on the error events of the respective spectral features and the energy. The interface module can include a discharge count detection module configured to detect occurrences of discharge events in the gas mixture of the optical system and produce an error event signal indicative of a count of the discharge events over a period of time. The performance metrics can include data related to the signal produced from the discharge detection module. The period of time can be measured from a last refill of the gas mixture. The period of time can be measured from a replacement of one or more chambers containing the gas mixture with one or more new chambers in the optical system. The interface module can include a fault label module configured to analyze each beam quality error event with respect to a set of fault labels and produce a likelihood score classifying the beam quality error event to a known fault label. The performance metrics can include data related to the output from the fault label module.

[0012] The predetermined learning model may receive the performance metric as an input and may output an estimate. The predetermined learning model may include a support vector machine. The predetermined learning model may include a separating hyperplane that classifies the performance metric as yes or no, where a yes classification indicates that the proposed change is effective and a no classification indicates that the proposed change is not effective.

[0013] The predetermined learning model may be constructed based on the type, configuration, and / or age of the optical system.

[0014] The estimate of the effectiveness of the proposed change may indicate whether the performance condition of the optical system after the proposed change is improved relative to the condition of the optical system before the proposed change.

[0015] The decision module may be configured to evaluate the effectiveness of the proposed change by comparing the performance of the optical system after the proposed change with the performance of the optical system before the proposed change. The decision module may be configured to output a change command if the comparison indicates that the performance will improve by a predetermined amount.

[0016] The proposed change to the optical system may include one or more of: refilling the proposed gas mixture within the optical system and changing the configuration of the proposed optical system. The decision module may be configured to: further based on the detected change to the configuration of the optical system, estimate the effectiveness of the proposed refilling of the gas mixture within the optical system.

[0017] In another general aspect, a method includes receiving a training dataset based on a plurality of test optical systems, and generating a prediction model based on the training dataset, the prediction model estimating the effectiveness of a proposed change to the optical system based on a performance metric related to a performance condition of the optical system. The training dataset includes, for each of a plurality of changes to each test optical system, a plurality of performance condition values ​​related to the test optical system before the change, and a plurality of performance condition values ​​related to the test optical system after the change.

[0018] Implementations may include one or more of the following features. For example, a prediction model may be generated by comparing, for each change to a test optical system, a plurality of performance status values ​​associated with the test optical system after the change and a plurality of performance status values ​​associated with the test optical system before the change, the comparison result indicating the effectiveness of the change.

[0019] The prediction model can be a learning model. The learning model can include a support vector machine. The learning model can be any structure other than a support vector machine, as long as it can be used as a prediction model. Specifically, aspects of the learning model include: assembling a dataset, assembling input vectors with fault signature detection (FSD), training the dataset, testing the dataset, and then applying the learning model. For example, in other implementations, the learning model includes a neural network, a decision tree, or a K-nearest neighbor model.

[0020] The predictive model may be generated by mapping the performance metric to one of a maintain command or a change command, the maintain command and the change command being based on the estimated effectiveness of the proposed change to the optical system.

[0021] The training data set may include at least several thousand variations for a plurality of test optical systems.

[0022] The performance condition may include one or more of the following: a beam quality error rate of a light beam generated from the optical system, a type of beam quality error of the light beam generated from the optical system, a number of discharge events occurring in a gas mixture of the optical system within a certain period of time, one or more faults associated with errors in beam quality of the light beam generated from the optical system, an anomaly in operating efficiency of the optical system, and errors in one or more spectral characteristics of the light beam generated from the optical system. The performance condition may include one or more of error events associated with an operating parameter or characteristic of the light beam, an analysis of each error event with respect to a fault signature, and a change in configuration of the optical system.

[0023] The method may further include testing the prediction model before applying the performance metric related to the performance condition of the optical system to the prediction model. The prediction model may be tested by: using a test data set, for each of a plurality of changes to each of the plurality of test optical systems and for each of the plurality of test optical systems, the test data set including: a plurality of performance condition values ​​related to the test optical system before the change and a plurality of performance condition values ​​related to the test optical system after the change, the test data set being excluded from the training data set; applying the plurality of performance condition values ​​related to the test optical system before the change to the prediction model, and comparing each actual output of the prediction model with the associated performance condition value related to the test optical system after the change from the test data set.

[0024] The estimate of the effectiveness of the proposed change may indicate whether the performance condition of the optical system after the proposed change is improved relative to the condition of the optical system before the proposed change.

[0025] The method may also include adjusting the predictive model to reduce the likelihood that the predictive model estimates that the proposed change is ineffective.

[0026] The predictive model may be generated based on the type, configuration, and / or age of the optical system.

[0027] The predictive model may be configured to estimate the effectiveness of a proposed gas change to an optical system that is within or different from the test optical system.

[0028] The proposed changes to the optical system may include one or more of: refilling of the gas mixture within the proposed optical system and a change of the configuration of the proposed optical system.

[0029] In other general aspects, a method includes receiving a performance metric related to a performance condition of an optical system that transmits an optical beam; estimating the effectiveness of a proposed change to the optical system based on the performance metric and a predetermined learning model; and directing the change to the optical system if the proposed change is estimated to be effective.

[0030] Implementations may include one or more of the following features: For example, changes to the optical system may be directed by outputting change commands to the optical system.

[0031] The method may further comprise delaying the proposed change if it is estimated that the proposed change is invalid.The change may be delayed by outputting a maintain command to the optical system.

[0032] The method may further include, after delaying the change: receiving a performance metric related to a performance condition of the optical system; and estimating an effectiveness of the proposed change to the optical system based on the performance metric and a predetermined learning model.

[0033] The effectiveness of a proposed change may be assessed by determining whether the performance of the beam will improve by performing the change.Determining whether the performance of the beam improves may include determining whether an error rate in beam quality will decrease as a result of the change.

[0034] The performance condition may include one or more of the following: a type of beam quality error of the optical beam, a number of discharge events occurring in the gas mixture within a certain time period, one or more faults associated with an error in the beam quality of the optical beam, the beam quality of the optical beam, and an error in the beam quality of the optical beam. The performance condition may include one or more of error events associated with an operating parameter or characteristic of the optical beam, an analysis of each error event with respect to a fault signature, and a change in the configuration of the optical system.

[0035] The predetermined learning model may be configured to estimate the effectiveness of proposed changes to the optical system before the changes are implemented.

[0036] The performance condition may include a count of discharge events in the gas mixture over a period of time.

[0037] The predetermined learning model may receive a performance metric as input and output an estimate.

[0038] The predetermined learning model may include a support vector machine. The predetermined learning model may include a separating hyperplane that classifies the performance metric as yes or no, wherein a yes classification indicates that the proposed change will be effective and a no classification indicates that the proposed change is not effective. The estimate of the effectiveness of the proposed change may indicate whether the condition of the optical system after the proposed change is improved relative to the condition of the optical system before the proposed change. The gas mixture may include a gain medium, wherein a population inversion is configured to occur by stimulated emission when energy is supplied to the gas mixture.

[0039] The proposed changes to the optical system may include one or more of: refilling of the gas mixture within the proposed optical system and a change of the configuration of the proposed optical system.

[0040] In another general aspect, a non-transitory computer-readable medium stores instructions that, when executed by a computer, cause the computer to perform a method comprising: receiving a training dataset based on a plurality of test optical systems, the training dataset including, for each of a plurality of changes to each test optical system, a plurality of performance status values ​​associated with the test optical system before the change; and a plurality of performance status values ​​associated with the test optical system after the change; and generating a prediction model based on the training dataset, the prediction model estimating effectiveness of a proposed change to the optical system based on a performance metric associated with the performance status of the optical system. The optical system may be among or different from the test optical system.

[0041] In other general aspects, a non-transitory computer-readable medium stores instructions that, when executed by a computer, causes the computer to perform a method comprising: receiving a performance metric related to a performance condition of an optical system that emits a light beam; estimating the effectiveness of a proposed change to the optical system based on the performance metric and a predetermined learning model; and directing the change to the optical system if the proposed change is estimated to be effective.

[0042] In other general aspects, an apparatus includes a decision module configured to: receive a performance metric related to a performance condition of an optical system that emits a light beam; estimate the effectiveness of refilling a gas mixture of the optical system based on the performance metric and a predetermined learning model; and output a refill command to the optical system if the refilling of the gas mixture of the optical system is estimated to be effective.

[0043] In a further general aspect, a method includes receiving a training dataset based on a plurality of test optical systems, the training dataset including, for each gas refill of a plurality of gas refills for each test optical system, a plurality of performance condition values ​​associated with the test optical system before the gas refills; and a plurality of performance condition values ​​associated with the test optical system after the gas refills; and generating a prediction model based on the training dataset, the prediction model estimating effectiveness of gas refilling of a proposed gas mixture in the optical system based on performance metrics associated with the performance conditions of the optical system. The optical system can be among or different from the test optical system.

[0044] In other general aspects, a method includes: receiving a performance metric related to a performance condition of an optical system that transmits an optical beam; estimating effectiveness of a proposed gas refill of a gas mixture for the optical system based on the performance metric and a predetermined learned model; and directing gas refill of the optical system if the proposed gas refill is estimated to be effective.

[0045] In a further general aspect, a non-transitory computer-readable medium stores instructions that, when executed by a computer, cause the medium to perform a method comprising: receiving a training data set based on a plurality of test optical systems, the training data set comprising, for each of a plurality of gas refills for each test optical system: a plurality of performance condition values ​​associated with the test optical system before the gas refills; and a plurality of performance condition values ​​associated with the test optical system after the gas refills; and generating a prediction model based on the training data set, the prediction model estimating effectiveness of gas refilling in a proposed optical system based on performance metrics associated with the performance condition of the optical system, wherein the optical system is different from the test optical system.

[0046] In even other general aspects, a non-transitory computer-readable medium stores instructions that, when executed by a computer, causes the medium to perform a method comprising: receiving a performance metric related to a performance condition of an optical system that emits a light beam; estimating, based on the performance metric and a predetermined learning model, the effectiveness of a proposed gas refill in the optical system; and directing gas refilling of the optical system if the proposed gas refill is estimated to be effective.

[0047] Implementations of any of the techniques described above and herein may include processes, devices, control systems, instructions stored on non-transitory machine-readable computer media, and / or methods. Details of one or more implementations are set forth in the accompanying drawings and the following description. Other features will be apparent from the description and drawings, and from the claims. BRIEF DESCRIPTION OF THE DRAWINGS

[0048] Figure 1 is a block diagram of a prediction apparatus configured to predict whether a proposed change to an optical system improves operation of the optical system based at least in part on a performance condition of the optical system;

[0049] Figure 2A is a graph of a total number of error events versus time for a set of performance conditions of a first type, wherein a refill process has little effect on the error events for the set of performance conditions of the first type;

[0050] Figure 2B is a graph of a total number of error events versus time for a set of performance conditions of a second type, wherein a refill process has a greater impact on the error events for the set of performance conditions of the second type;

[0051] Figure 3 Is the realization of the optical system and Figure 1 a block diagram of an implementation of an interface module for communicating with a prediction device, the interface module being configured to acquire and analyze data from an optical system;

[0052] Figure 4 Is included with Figure 1 or Figure 3 A block diagram of an implementation of an optical system of a prediction device communicating with a gas supply system;

[0053] Figure 5 is a block diagram of an implementation of an optical system including a dual-stage optical source;

[0054] Figure 6 is a block diagram of a training apparatus configured to be built on Figure 1 or Figure 3 a learning model used in the decision module of the prediction device;

[0055] Figure 7 It can be done by Figure 6 A flowchart of an implementation of a process for building a learning model performed by a training device;

[0056] Figure 8 For testing use, for example Figure 7 A flowchart of the process of implementing the learning model constructed by the process;

[0057] Figure 9 is Figure 1 or Figure 3 a flowchart of an implementation of a process performed by a prediction device for predicting whether a proposed change to an optical system will improve the operation of the optical system;

[0058] Figure 10A is configured to receive Figure 1 and Figure 3 A block diagram of an implementation of a lithography exposure apparatus for generating a light beam using an optical system;

[0059] Figure 10B yes Figure 10A A block diagram of an implementation of a projection optical system in a lithography exposure apparatus;

[0060] Figure 11A is a table of examples of likelihood scores output from a set of fault marking algorithms applied to each error event, each error event corresponding to a performance condition of the optical system;

[0061] Figure 11B is shown how to use Figure 11A a table of output scores for each of the fault marking algorithms; and

[0062] Figure 12 is a table of examples of a set of changes to configuration parameters, each configuration parameter change corresponding to a performance condition of the optical system. DETAILED DESCRIPTION

[0063] refer to Figure 1Monitoring device 100 monitors the operation of optical system 105, which generates light beam 110 for use by output device 115. Monitoring device 100 predicts whether a proposed change to optical system 105 will improve the operation of optical system 105. For example, the proposed change may be a proposed refilling of gas mixture 120 within optical system 105. Monitoring device 100 makes this prediction based at least in part on performance conditions of optical system 105. In particular, machine learning is used to create a predetermined learning model 125, and monitoring device 100 uses the predetermined learning model 125 to determine or estimate which performance conditions are more relevant or most relevant to the operating condition or operation of optical system 105. For example, if the proposed change is a proposed refilling of gas mixture 120, predetermined learning model 125 determines or estimates which performance conditions are more relevant to the operating condition or operation of gas mixture 120. In this way, the monitoring device 100 is able to identify or predict the effectiveness 129 of the proposed gas refilling, i.e., whether the optical system 105 will be positively affected by the gas refilling of the proposed gas mixture 120. As another example, if the proposed change is a change in the configuration of the optical system 105, the predetermined learning model 125 determines whether the proposed change in the configuration of the optical system 105 will improve the operation of the optical system 105.

[0064] In the figures, a solid line connecting two elements represents a data path along which data (e.g., information and / or command signals) can flow. Data can flow over wireless or wired connections. Additionally, a dashed line represents an optical path along which light can propagate, while a double line (e.g., in FIG. 1 ) represents an optical path along which light can propagate. Figure 3 ) represents a fluid path along which a fluid can flow.

[0065] The output device 115 can be a lithography system (also referred to as a scanner), and the beam 110 can be a pulsed light beam used by the scanner to expose a wafer. The pulsed light beam 110 is a string of light pulses, where each pulse is formed by exciting the gas mixture 120. The pulsed light beam 110 has a repetition rate. The repetition rate is the number of light pulses that occur within a time measurement. For example, the repetition rate can be the number of light pulses that occur within a second. The repetition rate is determined by the number of times the gain medium within the gas mixture 120 is excited within the time measurement. Additionally, the pulsed light beam 110 is associated with one or more specifications regarding the quality of the pulsed light beam 110. These specifications can include, for example, acceptable values and / or ranges of values for various properties of the pulsed light beam 110. For example, the specifications can include values related to pulse energy, wavelength, bandwidth, repetition rate, and / or pulse duration, or can include values derived from measurements of any of these properties. When the pulsed light beam 110 is within the specification, all of the properties of interest to the current application are within the range or equal to the values of the properties as set forth in the specification. When the pulsed light beam 110 is not within the specification, one or more of the properties of interest to the current application are outside the range or not equal to the values of the properties. When the beam 110 is within the specification, the optical system 105 has optimal performance.

[0066] The gas mixture 120 has an actual gas lifetime and an assumed gas lifetime. The actual gas lifetime is the period of time that the gas mixture 120 is able to produce light pulses that are within the specification. The assumed gas lifetime and the actual gas lifetime can be measured as, for example, the total number of pulses produced by the gas mixture 120 or the total time that the optical system 105 is operated at a particular repetition rate. For example, when the gas mixture 120 reaches the end of its actual gas lifetime, the gas mixture 120 can no longer be able to produce light pulses that are within the specification.

[0067] Other events can cause the gas mixture 120 to be unable to produce light pulses that are within the specification. For example, a sealed chamber (a gas discharge chamber) that holds or contains the gas mixture 120 can reach the end of its lifetime. Or, one or more modules within the optical system 105 can reach the end of their lifetimes. Or, one or more modules within the optical system 105 can become misaligned.

[0068] The assumed gas lifetime is a conservative estimate of the actual gas lifetime based on knowledge of the performance of optical systems similar to optical system 105. The assumed gas lifetime can be a predetermined constant value, and the assumed gas lifetime makes it highly unlikely that the gas mixture in any optical system will have an actual lifetime less than the assumed lifetime. However, various optical systems and gas mixtures experience different conditions during operational use. Thus, the actual gas lifetime of a particular gas mixture (such as gas mixture 120) can differ from the actual gas lifetime of another gas mixture. Furthermore, because the assumed gas lifetime is a conservative estimate, the actual gas lifetime of a particular gas mixture used under normal operating conditions may be greater than the assumed gas lifetime.

[0069] In a typical optical system, the gas mixture 120 is replaced when its assumed lifespan is reached using a process known as a refilling process. The refilling process involves removing and replacing the gas mixture 120. Furthermore, the refilling process typically consumes (wastes) material from the gas mixture 120. Furthermore, the optical system 105 cannot be operated during the refilling process. For at least these reasons, it is desirable to reduce the number of refilling processes performed on the optical system 105. Furthermore, it is desirable to only perform a refilling process when such a refilling process will improve the operation of the optical system 105. During the refilling process, the contents of the gas mixture 120 are returned to the desired mixture, concentration, and pressure. For example, new gas is introduced in an amount sufficient to achieve a specific pressure and concentration of certain components of the gas mixture 120. After refilling, the operation of the optical system 105 should ideally be improved relative to its operation before refilling.

[0070] For example, reference Figure 2A , discusses the effect of the refill process 230A on a hypothetical gas mixture. In this example, a graph of the total number of error events for a set of first type of performance conditions [pci1] is shown versus time. The total number of error events for the set of first type of performance conditions is shown before the refill process 230A and after the refill process 230A (where a new gas mixture is being used in the optical system 105). This comparison shows that the refill process 230A does not reduce the number of error events for the set of first type of performance conditions. Therefore, Figure 2A Illustrated are situations where refilling was unsuccessful in preventing or reducing the observed error events.

[0071] In comparison, reference Figure 2B , the effect of the refill process 230B on the hypothetical gas mixture is discussed. Figure 2BA graph of the total number of error events versus time for a set of second type performance conditions [pci2] is shown. The total number of error events for the set of second type performance conditions is shown before and after the refill process 230B (where a new gas mixture is being used in the optical system 105). This comparison shows that the number of error events for the set of second type performance conditions is reduced by the refill process 230B. Thus, Figure 2B The diagram illustrates a situation where refilling successfully prevented or reduced the observed error events. It can be believed that the set of performance conditions of the second type is more related to the operating conditions or operation of the gas mixture 120 than the other set of performance conditions of the first type. In particular, and generally speaking, the error event rate of performance conditions that are more related to the operating conditions or operation of the gas mixture 120 is reduced by the operation of the refill process 230.

[0072] However, it is not well understood which performance conditions are affected by a particular refilling procedure, especially when various combinations of many performance conditions are considered in the determination. This analysis is complex, at least in part, because of the large amount of information required to analyze the performance conditions. For example, the monitoring device 100 may receive information related to many different types of performance conditions. In some implementations, the monitoring device 100 monitors and analyzes more than 30 types of performance conditions (or error rates for performance conditions). Therefore, the monitoring device 100 uses a data-driven machine learning approach (via a predetermined learning model 125) to avoid performing unnecessary refills in a futile attempt to eliminate or reduce error rates for performance conditions unrelated to the operating conditions or operation of the gas mixture 120. With the help of the predetermined learning model 125, the monitoring device 100 identifies which combinations of performance conditions and error rates will be remedied by the refilling procedure (i.e., can be expected to be remedied by the refilling procedure). With the help of the predetermined learning model 125, the monitoring device 100 can also identify which combinations and ratios of error events will be remedied by another change to an aspect of the optical system 105 that does not include the refilling procedure (i.e., can be expected to be remedied by another change).

[0073] The performance condition includes any information related to whether the properties of the pulsed light beam 110 meet the specifications (or can be used to determine whether the properties of the pulsed light beam 110 meet the specifications). Each performance condition can be binary numerical data with only two possible values, where one value indicates that the properties of the pulsed light beam 110 meet the specifications and the other value indicates that the properties of the pulsed light beam 110 do not meet the specifications. In some implementations, the performance condition is non-binary numerical data. For example, the performance condition can be a value generated by a measurement module (such as Figure 3The performance condition may be a measurement value detected or sensed by a device as part of the measurement module 360. In another example, each performance condition may be a numerical value representing a difference between a measurement value and an expected or ideal value.

[0074] The performance conditions may be related to operating parameters or characteristics of the optical beam 110. For example, the performance conditions may include the energy, wavelength, pulse duration, repetition rate, and bandwidth of the optical beam 110. The performance conditions may include error events associated with any operating parameter or characteristic of the optical beam 110 or the optical system 105. For example, an error event is an event in which an operating parameter or characteristic of the optical beam 110 or the optical system 105 exceeds a threshold value, and when this occurs, the error event (including data associated with a time window surrounding the error event) is recorded and / or stored in memory.

[0075] In some implementations, an error event can be analyzed against a set of fault-marking algorithms (multiple fault-marking algorithms). Each fault-marking algorithm is designed to represent a specific failure mode or failure signature. Each fault-marking algorithm outputs a score in an attempt to classify the error event as a known failure mode or failure signature. For example, an algorithm for a specific failure signature may output a likelihood score for the presence of that specific fault signature in the error event. Figure 11A shows an example of the output of a set of fault marking algorithms for each error event (BQi, where i ranges from 1 to N), and Figure 11B It is shown how the scoring output from each of the fault-marking algorithms may be used, as discussed in more detail below.

[0076] In some implementations, as discussed below, the performance conditions relate to changes in the configuration of the optical system 105. For example, the performance conditions may include one or more of a change in the concentration of a component within the gas mixture 120, a change in the temperature of the gas mixture 120, a change in the voltage applied to an energy source that provides energy to the gas mixture 120, and a change in conditions related to how new components are added to the gas mixture 120.

[0077] The performance conditions may relate to operating parameters or characteristics of components within the optical system 105. For example, the performance conditions may include the number of pulses output from the gas mixture 120, the number of pulses of the optical beam 110 output from the optical system 105, and the efficiency or efficiency anomalies associated with components within the optical system 105, such as an optical oscillator or an optical amplifier. The performance conditions may include a fault indicator or operating state of the optical system 105.

[0078] The performance condition includes one or more of the following items: a type of beam quality error of the light beam 110, a number of discharge events occurring in the gas mixture 120 within a certain period of time, one or more faults associated with the error in the beam quality of the light beam 110, the beam quality of the light beam 110, and the error in the beam quality of the light beam 110.

[0079] In one example, a performance condition may include "MO Dropout," which is a single pulse "dropout" of energy output in the gas discharge chamber (which holds / contains the gas mixture 120) of the master oscillator (MO) of the optical system 105. For example, if the MO is an emission pulse of 1 millijoule (mJ), and the energy of the beam emitted from the MO drops to 0.1 mJ for a single pulse and then returns to 1 mJ, this may indicate a MO dropout event. Typically, MO dropout is believed to be related to the age of the gas discharge chamber and not to the gas operating condition. In another example, a performance condition may include "MO Rollover," which refers to the MO moving into an operating region where an increase in voltage results in a loss of energy. In short, this can be considered a long-term loss of efficiency of the MO gas discharge chamber. Typically, MO rollover conditions are believed to improve with refilling.

[0080] Reference again Figure 1 Monitoring device 100 includes a decision module 127 configured to receive performance metrics 107. Performance metrics 107 relate to a performance condition of optical system 105. That is, performance metrics 107 include information about one or more performance conditions of optical system 105. Performance metrics 107 may be a linear array of values ​​representing the performance conditions of optical system 105. Alternatively, performance metrics 107 may be a number or value determined or calculated based on the values ​​of the performance conditions of optical system 105.

[0081] Each performance condition provides an indication related to the health or operation of optical system 105. As discussed above, not all performance conditions are related to the health or operation of optical system 105 or gas mixture 120. Furthermore, some combinations of performance conditions may be more indicative of the health or operation of gas mixture 120 than other combinations of performance conditions. Some combinations of performance conditions may also change their impact on the health or operation of optical system 105 over time. For example, some performance conditions may have a greater impact on the health or operation of optical system 105 after optical system 105 has been used for a certain period of time than they did earlier in the life of optical system 105. Information in performance metrics 107 related to performance conditions of optical system 105 may include information related to errors or error rates for performance conditions or faults associated with the performance conditions. Therefore, in some cases, if the error rate for a particular combination of performance conditions increases, monitoring device 100 may determine that refill process 230 is needed.

[0082] The decision module 127 is configured to estimate the effectiveness 129 of the proposed change to the optical system 105 based on the received performance metric 107 and the predetermined learning model 125. The proposed change can be a proposed refill of the gas mixture 120 of the optical system 105, or a proposed change of the configuration of the optical system 105. The decision module 127 is also configured to output a command 109 to the optical system 105 that depends on or is directly related to the estimate of effectiveness 129. For example, if it is estimated that the proposed refill of the gas mixture 120 of the optical system 105 will be effective 129, the command 109 can be a refill command instructing the optical system 105 to refill the gas mixture 120.

[0083] Specifically, the predetermined learning model 125 receives the performance metric 107 as its input and outputs an estimate of effectiveness 129 for use by the decision module 127. The performance metric 107 includes information about the performance conditions that existed before the proposed change (such as the proposed gas refill). The estimate of effectiveness 129 is a prediction of whether the proposed (and yet to occur) change will result in a significant reduction in the error rate of the performance conditions of the optical system 105.

[0084] In some implementations, decision module 127 outputs a maintenance command, such as command 109, to optical system 105, for example, to extend use of gas mixture 120 if decision module 127 estimates that refilling of gas mixture 120 of optical system 105 as proposed will not be effective 129. In other implementations, decision module 127 delays sending command 109 to optical system 105 if decision module 127 estimates that refilling of gas mixture 120 of optical system 105 as proposed will not be effective 129; by delaying command 109, optical system 105 maintains its current state.

[0085] In still other implementations, instead of outputting a command, decision module 127 outputs a proposal to another controller or even to a field service engineer, and the controller / field service engineer can make a decision as to whether the proposed refill should be performed.

[0086] Decision module 127 may include or have access to one or more programmable processors, and each may execute a program of instructions to perform a desired function by operating on input data and generating an appropriate output. Decision module 127 can be implemented in any of digital electronic circuitry, computer hardware, firmware, or software. In further implementations, decision module 127 has access to a memory configured to store information output from decision module 127, information from optical system 105, or even performance metrics 107. Such information may be used by decision module 127 in various ways during operation of device 100. The memory may be read-only memory and / or random access memory, and may provide a storage device suitable for tangibly embodying computer program instructions and data. Monitoring device 100 may also include one or more input devices (such as a keyboard, a touch-enabled device, an audio input device) and one or more output devices (such as an audio output or a video output).

[0087] When energy is provided to the gas mixture 120 (e.g., by an energy source 324, such as Figure 3When the light beam 110 is generated by stimulated emission from a population inversion occurring in the gain medium of the gas mixture 120, the gas mixture 120 generates the light beam 110. The concentration state of the gas mixture 120 provides an indication of the relative amounts of the chemical components within the gas mixture 120. For example, the gas mixture 120 may include a mixture of a gain medium and a buffer gas, where the gain medium is a laser-active entity within the gas mixture 120, which may be a single atom, molecule, or pseudomolecule. The population inversion occurs in the gain medium. The gain medium may include a noble gas and a halogen, while the buffer gas may include an inert gas. Useful inert gases include, for example, argon, krypton, or xenon. The halogen gas may be, for example, fluorine. The inert gas may include, for example, helium or neon. For example, the gas mixture 120 may include argon fluoride (ArF), which generates the light beam 110 at a wavelength of approximately 193 nanometers (nm). As another example, the gas mixture 120 may include krypton fluoride (KrF), which generates the light beam 110 at a wavelength of approximately 248 nm.

[0088] The predetermined learning model 125 can be any structure capable of assembling a data set, assembling an input vector with a fault signature detection (FSD), a training data set, a test data set, and then applying a learning model. For example, in some implementations, the learning model includes a neural network, a decision tree, or a K-nearest neighbor model.

[0089] In some implementations discussed in detail herein, the predetermined learning model 125 is a support vector machine. The support vector machine classifies each input (performance metric 107) into one of two classes related to effectiveness 129. In this example, the first class is positive, meaning that the proposed change (such as the proposed gas refill) will be effective and improve the performance of the optical system 105; and the second class is neutral or non-positive, meaning that the proposed change (such as the proposed gas refill) will be ineffective and therefore will not improve the performance of the optical system 105.

[0090] The support vector machine can utilize a separating hyperplane that classifies input data (i.e., performance metric 107) into a first class or a second class, and such a hyperplane is capable of analyzing multi-dimensional performance metrics 107. The support vector machine also defines boundaries and constraints associated with the analysis of the input data in order to separate the data in an efficient manner, thereby outputting a classification.

[0091] The predetermined learning model 125 may be constructed based on the type, configuration, changes in configuration, and / or age of the optical system 105 , as will be discussed below.

[0092] See also Figure 3In some implementations, the monitoring device 100 can be implemented as the monitoring device 300, and the optical system 105 can be designed as an optical system 305 that generates the light beam 310. The optical system 305 includes an optical source 340 that generates the light beam 310 from the gas mixture 320, and a gas supply system 350 in fluid communication with the gas mixture 320.

[0093] Although Figure 3 3. Although not shown, optical source 340 may include other gas mixtures and other optical components used in conjunction with gas mixture 320 to produce light beam 310. These other gas mixtures may be in fluid but separate communication with gas supply system 350.

[0094] Gas mixture 320 is part of a gas subsystem 322 within optical source 340. Gas subsystem 322 may include other components, such as a container 323 that forms a sealed chamber (gas discharge chamber) that holds gas mixture 320 and an energy source 324 for exciting the gain medium within gas mixture 320. Energy source 324 may include a cathode and an anode, and gas discharge chamber 323 may enclose cathode and anode 214b and gas mixture 320. The potential difference between the cathode and anode creates an electric field in gas mixture 320. The electric field provides energy to the gain medium within gas mixture 320 sufficient to cause a population inversion and generate light pulses via stimulated emission. The repeated creation of this potential difference creates a train of light pulses that ultimately constitutes light beam 310. A "discharge event" is the application of a voltage that creates a potential difference sufficient to cause a discharge in the gain medium of gas mixture 320 and the emission of light pulses.

[0095] Figure 5 An implementation of a gas subsystem 322 and an optical source 340 comprising two chambers, each holding or retaining its own gas mixture, is shown.

[0096] The gas supply system 350 includes one or more gas sources, fluid conduits configured to supply gas to the chambers of the gas subsystem 322 , and a valve system including one or more fluid control valves between the gas sources and the chambers. Figure 4 An implementation 450 of the gas supply system 350 is shown.

[0097] Monitoring system 300 includes a decision module 327 that, like decision module 127 , receives performance metrics 307 related to the performance conditions of optical system 305 . That is, performance metrics 307 include information about one or more performance conditions of optical system 305 .

[0098] Additionally, monitoring system 300 includes an interface module 360 ​​in communication with decision module 327. Interface module 360 ​​is configured to obtain and analyze data from optical system 305 and light beam 310, calculate and establish performance metric 307, and then provide performance metric 307.

[0099] Interface module 360 ​​includes multiple analysis submodules 360i, a memory 361 that receives and stores information from one or more of the analysis submodules 360i, and an output submodule 362 that accesses output from one or more of the analysis submodules 360i and the memory 361 and constructs performance metrics 307. In the illustrated example, analysis submodules 360i include four analysis submodules 360A, 360B, 360C, and 360D. Each analysis submodule 360i is configured to interact with a specific aspect of optical system 305, including the possibility of interacting with light beam 310. Each analysis submodule 360i includes hardware that senses, detects, or receives data or information related to its specific aspect of optical system 305. Interface module 360 ​​prepares a performance status of optical system 305 based on this data or information and, based on the performance status, constructs performance metrics 307 for use by decision module 327. The interface module 360 ​​may include or have access to one or more programmable processors and may each execute a program of instructions to perform a desired function by operating on input data and generating appropriate output. The interface module 360 ​​may be implemented in any of digital electronic circuitry, computer hardware, firmware, or software.

[0100] In this implementation, at least two of the analysis submodules 360i are beam quality detection submodules 360C and 360D. For example, the beam quality detection submodule 360C can be a spectral feature detection submodule configured to detect errors in the spectral features of the light beam 310 and generate a signal indicative of the corresponding spectral feature errors of the light beam 310. The spectral features of the light beam 310 can be any characteristics of the spectrum (or emission spectrum) of the light beam 310. The spectrum contains information about how the light energy or power is distributed as a function of the wavelength or light frequency of the light beam 310. Thus, for example, the spectral feature of the light beam 310 can be the wavelength or the width of the spectrum (referred to as the bandwidth) at a specific light energy or power. As another example, the beam quality detection submodule 360D can be an energy detection submodule configured to detect errors in the energy of the light beam 310 and generate a signal indicative of the energy error of the light beam 310.

[0101] The beam quality detection submodules 360C and 360D may include corresponding sensors, which may be positioned at any location where the beam 310 can be sensed. For example, the sensors may be in the optical system 305, between the optical system 305 and the output device 115, or in the output device 115.

[0102] The output of each of submodules 360C and 360D can be actual measurement data from a sensor, or it can be an average of data received from the sensor over a specific time period. For example, if beam quality detection submodule 360D is an energy detection submodule, it may include an energy sensor for estimating dose, and the output may have two values: one indicating that the dose is within specification and the other indicating that the dose is not within specification. Dose is the amount of light energy delivered to the wafer area. To determine dose, the energy sensor measures the amount of energy over a specific time period and also counts the number of pulses of beam 310 emitted during that time period. In these implementations, the energy sensor may include a detector that measures energy and a beam splitter in the path of beam 310. The beam splitter directs a portion of the light in each pulse to the detector. The detector measures the amount of energy over a specific time period. Furthermore, the number of pulses occurring within that time period can be derived from the energy measured by the detector. For example, if the detected energy is greater than a threshold, a pulse is considered present. If the detected energy is below the threshold, a pulse is not present. Thus, data from the energy sensor can be used to determine a beam quality metric based on dose.

[0103] The output from the analysis submodules 360C, 360D may include error events associated with any operating parameter or characteristic of the light beam 310 or the optical system 305. As discussed above, an error event is an event in which an operating parameter or characteristic of the light beam 310 or the optical system 305 exceeds a threshold value. When this occurs, the error event (including data associated with a time window surrounding the error event) is recorded and / or stored in the memory 361 for access by the output module 362.

[0104] At least one analysis submodule 360i is a discharge count detection submodule 360B configured to detect the occurrence of discharge events in the gas mixture 320 and generate a signal indicating a count of discharge events over a period of time. The period of time may be measured from the last time the gas mixture 320 was refilled or from the replacement of one or more chambers containing the gas mixture 320 within the optical system 305.

[0105] At least one of the analysis submodules 360i is a fault signature submodule 360A, configured to analyze each error event relative to a set of fault signatures and generate a likelihood score for classifying the error event as a known fault signature. The algorithm for a particular fault signature may output a likelihood score for the presence of that particular fault signature in the error event. Figure 11A shows an example of the output of a set of fault marking algorithms for each error event (BQi, where i ranges from 1 to N), and Figure 11B It shows how the scoring output from each fault-flagging algorithm is used, as discussed in more detail below.

[0106] refer to Figure 4 , shows an implementation of a gas supply system 450. The gas supply system 450 includes: one or more gas sources 451A, 451B, 451C; a conduit for supplying gas to the chamber 323 in the gas subsystem 322; and a valve system 452, including one or more fluid control valves between the gas sources 451A, 451B, 451C and the chamber in the gas subsystem 322. The gas sources 451A, 451B, 451C can supply gas to multiple chambers, for example, such as when the optical source 340 includes multiple stages (each stage including a gas discharge chamber having a gas mixture), as described with reference to Figure 5 As discussed. The gas sources 451A, 451B, 451C can be, for example, sealed gas bottles and / or canisters. As an example, the gas mixture 320 can contain a halogen (such as fluorine) and other gases (including argon, neon, and possibly other substances with different partial pressures (summed to a total pressure P)). In addition, one or more gas sources 451A, 451B, 451C are connected to the chamber 323 through a set of fluid control valves within the valve system 452. With this design, gases can be injected into the chamber 323 along with specific relative amounts of components of the gas mixture 320. For example, if the gain medium in the gas mixture 320 is argon fluoride (ArF), one of the gas sources 451A can contain a mixture of the following gases, which mixture includes the halogen fluorine, the rare gas argon, and one or more other rare gases (such as a buffer gas, including an inert gas such as neon). The mixture described can be referred to as a tri-mixture. In this example, gas source 451B may contain a mixture of gases including argon and one or more other gases in addition to any fluorine. The mixture described may be referred to as a dual mixture. Although only three gas sources 451A, 451B, and 451C are shown, gas supply system 450 may have fewer than three or more than three gas sources.

[0107] The decision module 327 can communicate with the valve system 452 using one or more signals to cause the valve system 452 to transfer gas from the particular gas source 451A, 451B, 451C into the chamber 323 during gas refill. Additionally or alternatively, the decision module 327 can communicate with the valve system 452 using one or more signals to cause the valve system 452 to initially bleed all gas from the chamber 323 prior to refilling, and such bleeded gas can be vented to the gas waste dump 490.

[0108] Although not shown, the fluid control valves of valve system 452 may include multiple valves assigned to each chamber of gas subsystem 322 or each chamber of optical source 340. For example, valve system 452 may include an injection valve that allows gas to enter and exit the chamber at a first rate, and a chamber fill valve that allows gas to enter and exit the chamber at a second rate different from the first rate.

[0109] refer to Figure 5 , shows an implementation 505 of optical system 305. Optical system 505 is a dual-chamber optical system 505 including an optical source 540 having a first gas subsystem 522A and a second gas subsystem 522B in optical communication with first gas subsystem 522A. First gas subsystem 522A is a master oscillator system, and second gas subsystem 522B is a power amplifier system. Master oscillator system 522A includes a master oscillator gas discharge chamber 523A, and power amplifier system 522B includes a power amplifier gas discharge chamber 523B. Master oscillator gas discharge chamber 523A includes two elongated electrodes as energy sources 524A, which provide a pulsed energy source to gas mixture 520A within chamber 523A. Power amplifier gas discharge chamber 523B includes two elongated electrodes as energy sources 524B, which provide a pulsed energy source to gas mixture 520B within chamber 523B.

[0110] Master oscillator system 522A provides a pulsed amplified optical beam (referred to as a seed beam) 508 to power amplifier system 522B. Master oscillator gas discharge chamber 523A contains gas mixture 520A, which includes a gain medium in which amplification occurs, and includes an optical feedback mechanism such as an optical resonator. The optical resonator is formed between spectroscopic optics 541 on one side of master oscillator gas discharge chamber 523A and output coupler 542 on a second side of master oscillator gas discharge chamber 523A. Power amplifier gas discharge chamber 523B contains gas mixture 520B, which includes a gain medium in which amplification occurs when developed using seed beam 508 from master oscillator system 522A. If power amplifier system 522B is designed as a regenerative ring resonator, it is described as a power ring amplifier, and in this case, sufficient optical feedback can be provided from the ring design. The power amplifier system 522B may also include a beam return (e.g., a reflector) 543 that returns the beam (e.g., via reflection) back into the power amplifier gas discharge chamber 523B to form a loop and a ring path (where the input entering the ring amplifier intersects the output exiting the ring amplifier). For example, the master oscillator system 522A may emit a pulsed seed beam 508 having a seed pulse energy of approximately 1 millijoule (mJ) per pulse, and these seed pulses may be amplified to approximately 10 to 15 mJ by the power amplifier system 522B.

[0111] The gas mixture used in the respective discharge chambers 523A, 523B (e.g., gas mixtures 520A, 520B) can be a combination of gases suitable for producing an amplified light beam around the desired wavelength, bandwidth, and energy. For example, as discussed above, the gas mixtures 520A, 520B can include argon fluoride (ArF) (emitting light at a wavelength of approximately 193 nm) or krypton fluoride (KrF) (emitting light at a wavelength of approximately 248 nm).

[0112] refer to Figure 6, the training device 650 establishes the learning model 125. The training device 650 includes a training module 652, which is configured to receive a training data set 654 and generate a learning model 125 based on the training data set 654. The training module 652 may include or have access to one or more programmable processors, and each may execute a program of instructions to perform a desired function by operating on input data and generating an appropriate output. The training module 652 can be implemented in any of digital electronic circuitry, computer hardware, firmware, or software. In another implementation, the training module 652 accesses a memory configured to store information output from the training module 652, information used to generate the training data set 654, or the training data set 654. The memory may be read-only memory and / or random access memory, and may provide a storage device suitable for tangibly embodying computer program instructions and data. The training module 652 may also include one or more input devices (such as a keyboard, a touch-enabled device, an audio input device) and one or more output devices (such as an audio output or a video output).

[0113] The training data set 654 is formed by a plurality of test optical systems 605t-i (where i=1, 2, 3, ... X) and is based on one or more system changes k performed on each of the test optical systems 605t-i. The test optical system 605t-i may or may not be the same as the optical system 105 to be analyzed by the decision module 127, 327. Each system change k may be a refill of the gas mixture 620-i within the test optical system 605t-i or a change in the configuration of the test optical system 605t-i.

[0114] The training data set 654 includes two multidimensional matrices, each referred to as a pc set (i, k). The first matrix [pc set prior (i, k)] includes a plurality of performance status values ​​associated with each test optical system 605t-i and each system change k achieved before the system change for that test optical system 605t-i. The second matrix [pc set after (i, k)] includes a plurality of the same performance status values ​​associated with each test optical system 605t-i and each system change k achieved after the system change for that test optical system 605t-i. The value i in the set corresponds to the test optical system 605t-i in which the system change was performed (i ranges from 1 to X), and the value k corresponds to the specific system change performed on the i-th test optical system 605t-i (k ranges from 1 to Y). The total number Y of system changes k for each test optical system 605t-i can vary. That is, more system changes can be performed on some test optical systems 605t-i than on other test optical systems 605t-i.

[0115] Each plurality of performance condition values ​​pc set includes a plurality of performance conditions, and the total number of different performance conditions in the plurality of performance conditions can be any number and depends only on how many performance conditions are monitored or tracked by interface module 360. For example, in some cases, there may be dozens of performance conditions in the plurality of values ​​pc set.

[0116] Thus, the training data set 654 includes several sets of performance condition values ​​for each test optical system 605t-i, as several gas refills are performed for each of the test optical systems 605t-i. For example, the training data set 654 may include multiple performance condition values ​​for hundreds or thousands of gas refills, which may be associated with dozens or hundreds of different test optical systems 605t-i. Additionally, the training data set 654 may include multiple performance condition values ​​for dozens, hundreds, or thousands of changes to the configuration of the test optical system 605t-i.

[0117] Because the learning model 125 is constructed from so many different system changes k (including gas refills and changes to the configuration) and so many different test optical systems 605t-i, the learning model 125 can be modular. This means that the learning model 125 can be used on any other optical system 105 having a similar design to the test optical system 605t-i used to generate the training data set 654 that is input to the training module 652.

[0118] The learning model 125 is a prediction model 125. In some implementations, the learning model 125 comprises a support vector machine, as discussed above, when used by the decision module 127 ( Figure 1 ), the support vector machine classifies each input (performance metric 107) into one of two classes related to effectiveness 129. The estimate of effectiveness 129 of a proposed system change (such as gas refill) indicates whether the performance condition of the optical system 105 will improve as a result of the proposed system change. The first class is positive, which means that the proposed system change will be effective and improve the performance condition of the optical system 105; and the second class is neutral or non-positive, which means that the proposed system change will not be effective and, therefore, will not improve the performance condition of the optical system 105.

[0119] The learning model 125 initially produces a simulated output that indicates which side of the hyperplane the observation (performance metric 107) lies on and how far away it is from the hyperplane. The values ​​of the simulated output range between +1 and 1, and the threshold can be 0 in simplified form.

[0120] The learning model 125 determines a binary output (positive or neutral class) based on the simulated data. Thus, the learning model 125 can assign any positive value to the first positive class (+1) and any negative value to the second neutral class (-1). The learning model 125 assigns any value close to 0 with slightly less confidence. Thus, for example, given an input vector κ (performance metric 107), a simulated output that is positive and has a small magnitude (e.g., +0.2) indicates that the input vector κ is on the "positive" side of the hyperplane but close to the hyperplane, and is therefore subject to some uncertainty when being classified as the positive class. On the other hand, a simulated output that is negative and has a larger magnitude (e.g., 0.9) indicates that the input vector κ is on the "negative" side of the hyperplane but relatively far from the hyperplane, and is therefore more likely to be correctly classified as the neutral class.

[0121] refer to Figure 7 , the learning model 125 is trained according to process 760. The training module 652 performs process 760. Initially, the training module 652 receives the training data set 654 (762). Next, the training module 652 generates the learning model 125 based on the training data set 654 (764). In some implementations, the generated learning model 125 is validated (766) using a test data set, as discussed below. The generated learning model 125 is output for use by the decision module 127.

[0122] The training module 652 can generate the learning model 125 (764) by comparing, for each system change k and each test optical system 605t-i, a plurality of performance status values ​​pcset(i, k) measured after the system change with a plurality of performance status values ​​pcset(i, k) measured before the system change. Ultimately, the generation (764) of the learning model 125 includes a mapping of the performance metric 107 (including the plurality of performance status values) to a specific command 109, which can be a refill command or a maintenance command or a command for changing the configuration of the optical system 105.

[0123] As mentioned, process 760 may also include an optional process 766 of testing the learning model 125. Before the decision module 127 uses the learning model 125, the learning model 125 may be tested 766 to ensure that the learning model 125 operates with an appropriate set of output constraints. Test 766 may be performed to determine the accuracy of the learning model 125. For example, test 766 may determine that the learning model 125 correctly predicted that a gas refill was unnecessary 58% of the time, thereby resulting in a 58% time saving. Test 766 may determine that the learning model 125 correctly predicted that a gas refill was necessary 28% of the time. Test 766 may determine that the learning model 125 uncertainly predicted that a gas refill was necessary 6% of the time and incorrectly predicted that a gas refill was unnecessary 8% of the time. In this example, the accuracy of the learning model 125 is 86%. The results of test 766 may be used to adjust the learning model 125 so that the percentage of time that the learning model 125 incorrectly predicts that a gas refill is unnecessary is less than a low value (such as 3%). By adjusting the learning model 125 in this manner, the risk of damaging the optical system 105 when the learning model 125 incorrectly predicts that gas refilling is unnecessary can be reduced. The learning model 125 can be adjusted using configurable parameters that adjust the separating hyperplane in the learning model 125, and the adjustment makes the decisions of the learning model 125 more conservative.

[0124] In some implementations, the adjustment is performed by changing the threshold to a value slightly away from 0. For example, if the purpose of the adjustment is to bias the learning model 125 to make more positive decisions (thus a first positive class of +1), the threshold can be changed to -0.2. In other implementations, the adjustment is performed by inserting a penalty function during process 760 that assigns a penalty score to each possible outcome, which is then proportionally adjusted by the penalty function to each observation during training, such that the resulting learning model 125 becomes inherently biased towards the outcome with the lowest penalty coefficient.

[0125] Additionally, in some implementations, the training module 652 generates the learning model 125 based on the type, configuration, and / or age of the optical system 105 (764). This information may be obtained from the interface module 360, which may monitor the optical system 105. For example, for a relatively new optical system 105, some performance conditions will improve with gas refilling, while for a relatively old optical system 105, other performance conditions will improve with gas refilling. As another example, MO loss is often associated with the end of life of the gas discharge chamber 523A of the master oscillator system 522A within the optical system 505. Thus, if the learning model 125 receives information indicating that MO loss is occurring at a high rate and that the chamber 523A is old, the learning model 125 is less likely to attribute problems with the optical system 505 to the operating conditions of the gas mixture 520A within the chamber 523A. On the other hand, if learning model 125 receives information indicating that MO dropouts of master oscillator system 522A are occurring at a high rate and chamber 523A is new, learning model 125 is more likely to infer that the problem with optical system 505 is related to the operating conditions of gas mixture 520A within chamber 523A.

[0126] refer to Figure 8 , performing a process 766 of testing the learning model 125. Process 766 can be performed by a dedicated testing module. Process 766 includes receiving a test data set 868 (870). Test data set 868 includes a set of test performance condition values ​​pc-tset(i, k) measured before system change k and a set of test performance condition values ​​pc-tset(i, k) measured after the set of system change k. Test data set 868 can be created in a manner similar to training data set 654, except that test data set 868 is not used during training process 760. In this way, test data set 868 does not bias the training process 760 with its tendency.

[0127] Next, the test data set 868 is applied to the learning model 125 resulting from the training process 760 (872). The test data set 868 can be applied to the learning model 125 by inputting the test performance condition values ​​pc-t set(i, k) measured before a set of system changes k into the learning model 125, and then estimating the effectiveness 129 of each system change (e.g., each gas refill or configuration change of the gas mixture 120) of the optical system 105 (872). The effectiveness 129 is a prediction of whether the system change (i.e., gas refill or configuration change) will result in an acceptable reduction in the error rate of the performance condition of the optical system 105.

[0128] The dedicated test module determines whether the learning model 125 is acceptable by comparing the effectiveness 129 output from the learning model 125 with the set of test performance condition values ​​pc-t set (i, k) actually measured after the system changes a set of k (874). At 874, it can be determined whether the effectiveness 129 has an accuracy greater than a certain percentage. For example, the dedicated test module determines how often the learning model 125 accurately predicts that gas refills are necessary and how often the learning model 125 accurately predicts that gas refills are unnecessary. As another example, the dedicated test module can compare the accuracy of the effectiveness 129 using the test performance condition pc-t set (i, k) with the accuracy of the effectiveness 129 using the training data set 654. If the two accuracies are within a few percentage points of each other, the learning model 125 can be considered to be well-fitting and generalizable. The effectiveness 129 output from the dedicated test module indicates whether the system change will significantly reduce the error of the test performance condition values ​​and is compared with the set of test performance condition values ​​pc-t set (i, k) actually measured after the system changes a set of k.

[0129] If the dedicated testing module determines that the learned model 125 is acceptable (874), the learned model 125 is output (876) for use by the decision module 127. On the other hand, if the dedicated testing module determines that the learned model 125 results in too many incorrect predictions, the dedicated testing module may adjust the learned model 125 to reduce the number of incorrect predictions (878).

[0130] refer to Figure 9 , process 980 is performed by decision module 127 for predicting whether a proposed system change to optical system 105 will improve the operation of optical system 105. For example, process 980 predicts whether one or more improvements within optical system 105 will improve the operation of optical system 105. The one or more changes include gas refilling of gas mixture 120 within optical system 105, a change in the configuration of optical system 105, and both gas refilling of gas mixture 120 and a change in the configuration of optical system 105.

[0131] The decision module 127 receives a query (981) regarding whether a change in the state of the optical system 105 can be made. The query (981) can occur regularly, for example, it can occur at a set frequency, and the frequency can depend on factors such as the operating condition or age of the optical system 105. For example, the query (981) can occur every few minutes, every few hundred minutes, or every few days. The query (981) can be generated external to the optical system 105, or in some cases can be generated by the output device 115, or the query (981) can be generated by a field engineer or operator. The query (981) can be generated when a monitored aspect of the output device 115 or the optical system 105 is above a threshold.

[0132] For example, when a performance condition error rate rises above a threshold, a query (981) can be generated. The performance condition error rate can be an error rate for a set of performance conditions of the optical system 105. The performance condition error rate can be determined based on the output of the measurement module 360. The query (981) can be generated by the interface module 360 ​​and provided to the decision module 127.

[0133] The query ( 981 ) may include a request for gas refill of the gas mixture 120 of the optical system 105 .

[0134] Upon receiving the query (981), the decision module 127 receives the performance metric 107 (982), for example, from the interface module 360. The decision module 127 estimates the effectiveness 129 of the proposed system change of the optical system 105 (e.g., the proposed gas refilling of the gas mixture 120) based on the performance metric 107 and the predetermined learning model 125 (983). The decision module 127 (via the learning model 125) is therefore configured to estimate the effectiveness 129 of the proposed gas refilling of the gas mixture 120 before performing the gas refilling, so that unnecessary gas refilling is avoided or necessary gas refilling is performed. The decision module 127 may also determine whether it makes more sense to change the configuration of the optical system 105 in order to improve the operation of the optical system 105 (instead of or in addition to performing the gas refilling).

[0135] If it is estimated (982) that the proposed gas refill is valid, the decision module 127 directs 109 the gas refill in the optical system 105 (984).

[0136] Decision module 127 may instruct the gas refill by, for example, outputting a gas refill command 109 to optical system 105 instructing optical system 105 to refill the gas mixture ( 984 ).

[0137] The process 980 may also include delaying the gas refill (indicated by returning to step 981) if it is estimated 983 that the proposed gas refill is ineffective 129. Furthermore, the decision module 127 may alternatively delay the gas refill by outputting a maintenance command 109 to the optical system 105 to extend the use of the gas mixture. If the decision module 127 determines that a particular change in configuration will result in an improvement in the operation of the optical system 105, the decision module 127 may alternatively direct or recommend a change in the configuration of the optical system 105.

[0138] Thus, once the decision module 127 has decided to delay gas refilling (and extend the life of the gas mixture 120), the process 980 returns to wait for further state change queries (981), at which point the decision module 127 receives the performance metrics 107 (982) and estimates the effectiveness 129 of the proposed system changes to the optical system 105 (983). In this way, the gas mixture 120 is used as much as possible, but not excessively. The monitoring device 100 allows the optical system 105 to optimize the use of the gas mixture 120. In addition, the monitoring device 100 promotes conservation of resources and provides the possibility of reducing the number of gas refills.

[0139] Decision module 127 can evaluate the effectiveness 129 of the proposed refill by determining or estimating whether the performance of beam 110 produced by optical system 105 will be improved by gas refilling. For example, decision module 127 uses learning model 125 to determine whether the error rate of beam quality will decrease as a result of performing gas refilling.

[0140] The performance conditions monitored and included in the performance metrics 107 include one or more of the following: the type of beam quality error of the light beam 110, the number of discharge events that occur in the gas mixture 120 over a period of time, one or more faults associated with errors in the beam quality of the light beam 110, the beam quality of the light beam 110, and the errors in the beam quality of the light beam 110. The performance conditions may include a count of discharge events in the gas mixture 120 over a period of time or use. The monitored performance conditions may include all error events (i.e., events where an operating parameter or characteristic of the light beam 110 or the optical system 105 exceeds a threshold) and metadata associated with each error event. The performance conditions may also include a score output from a fault marking algorithm for each of the error events to thereby classify each error event as a known failure mode or failure signature. Additionally, the performance conditions may include changes in the configuration of the optical system 105.

[0141] refer to Figure 10A and Figure 10B, shows an implementation 1015 of the output device 115. In this implementation, the output device 115 is a lithographic exposure apparatus 1015 that includes a projection optical system 1091 through which the light beam 110 passes before reaching a wafer 1092, and a sensor system or metrology system including a sensor 1060. The lithographic exposure apparatus 1015 can be a liquid immersion system or a dry system.

[0142] The sensor 1060 may be part of or in communication with the metrology module 360. The sensor 1060 may be, for example, a camera or other device capable of capturing an image of the light beam 110 at the wafer 1092, or an energy sensor capable of capturing data describing the amount of light energy at the wafer 1092 in the xy plane.

[0143] For example, microelectronic features are formed on the wafer 1092 by exposing a radiation-sensitive photoresist material layer on the wafer 1092 using an exposure beam 1093 output from the projection optical system 1091. Figure 10B , the projection optical system 1091 includes a slit 1094, a mask 1095, and a projection objective lens including a lens system 1096. The lens system 1096 includes one or more optical elements. The light beam 110 enters the photolithography exposure device 1015 and is incident on the slit 1094, and at least some of the light beam 110 passes through the slit 1094 to form an exposure beam 1093. Figure 10A and Figure 10B In the example shown, slit 1094 is rectangular and shapes light beam 110 into an elongated rectangular shaped light beam (exposure beam 1093). A pattern is formed on mask 1095, and this pattern determines which portions of the shaped light beam are transmitted by mask 1095 and which are blocked by mask 1095. The design of the pattern is determined by the specific microelectronic circuit design to be formed on wafer 1092.

[0144] The decision module 127 does not only consider the error rate of the performance condition when determining whether to trigger gas refilling. Instead, the decision module 127 also uses information related to the type or kind of error within the performance metric 107 to estimate whether gas refilling is effective.

[0145] Next, an example of performance metric 107 is discussed. Performance metric 107 may include fewer or more elements than those discussed next, and this example is not meant to be limiting.

[0146] In this example, the performance metric 107 is defined using one or more of the following data: a first metadata set κ meta , including data related to error events (BQ) and / or states related to the optical system; a second data set κ FSD, comprising a set of scores for each error event BQ, the scores classifying the error event BQ as a known fault marker; and a third data set κ config , comprising a set of possible / detected changes to the configuration of optical system 105. Performance metric 107 is constructed from one or more of these data sets, as discussed next.

[0147] In some implementations, the performance metric 107 is composed of two datasets (specifically, the first metadata dataset κ meta and the second dataset κ FSD ) build. This is discussed below.

[0148] The first meta-data set κ meta The following information may be included: shotsGas, which is the total number of pulses of energy supplied to the gas mixture 320 since the last gas refill; shotsChamber, which is the total number of pulses of energy supplied to any gas mixture 320 within the gas discharge chamber 323 since the gas discharge chamber 323 was initially installed in the optical source 340; and typeBQ, which is the type of performance condition associated with the error event. meta Less or more information than listed may be included.

[0149] The second dataset κ FSD A score FSD may be included that is a set of scores determined for each error event BQ output from the fault flag submodule 360A within the interface module 360 ​​.

[0150] Each performance status error event BQ generates a first metadata set κ meta and the second dataset κ FSD For example, if there are 20 performance condition error events BQ before gas refill, there will be 20 rows, each containing one or more columns corresponding to metadata and a score set output from the fault marking submodule 360A to form a matrix. This matrix information can be further transformed into a linear array κ to be used as a performance metric 307 for the learning model 125, as shown below.

[0151] It can be assumed that in such a small time window before the proposed gas refill, the data for shotsGas and shotsChamber should not change significantly between performance status error events BQ. Therefore, the values ​​of shotsGas and shotsChamber at the time of gas refill can be used in the first metadata set κ for input into the linear array κ meta Used in.

[0152] For data type BQ, the corresponding input is defined as a fraction of performance condition error events BQ that are assigned to each type of performance condition error event. An explanation is provided next. In this particular example, there are five performance condition error events, 3 error events for the energy E of the light beam 110, 1 error event for the wavelength W of the light beam 110, and 1 error event for the bandwidth B of the light beam 110. This data is transformed into a 1 x 3 array [0.6, 0.2, 0.2] that can be used for the first metadata set κ meta As follows:

[0153]

[0154] Further, the input corresponding to the scores FSD is transformed in the same manner, resulting in a 1 x β array, where each element in the array represents a fraction of BQ identified in each of the β fault markers, and β corresponds to the number of fault markers in the set of fault markers analyzed by the fault marker submodule 360A. For example, β can be a value greater than 1, can be greater than 10, or can be greater than 20, and in one specific example discussed next, β is 28.

[0155] The linear array κ that can then be used as the performance metric 307 for the learning model 125 is given by the concatenation of each of the above arrays, as follows:

[0156] κ = [κ meta κ FSD ], or

[0157] κ = [shotsGas (1 x 1) shotsChamber (1 x 1) typeBQ (1 x 3) scoresFSD (1 x 28)], where κ meta = [shotsGas (1 x 1) shotsChamber (1 x 1) typeBQ (1 x 3)] and κ FSD = [scoresFSD (1 x 20)]; shotsGas (1 x 1) is the total number of pulses of energy supplied to the gas mixture 320 since the last gas refill; shotsChamber (1 x 1) is the total number of pulses of energy supplied to any gas mixture 320 within the gas discharge chamber 323 since the gas discharge chamber 323 was initially installed in the optical source 340; typeBQ (1 x 3) is a linear array of the three types of performance condition error events BQ; and scoresFSD (1 x 28) is a linear array of 28 scores, each determined by the fault marker submodule 360A within the interface module 360.

[0158] At step 982 of process 980, the cascaded linear array 307 is input into the learning model 125. The learning model 125, which may include a support vector machine, classifies the linear array κ 307 into one of two classes related to effectiveness 129. The first class is positive, meaning that the proposed gas refill will be effective and improve the performance of the optical system 105; and the second class is neutral or non-positive, meaning that the proposed gas refill will be ineffective and therefore will not improve the performance of the optical system 105. The decision module 127 determines effectiveness 129 based on the determined class and outputs a command 109 to the optical system 305.

[0159] Figure 11A and Figure 11B shows how to form the subarray κ FSD In this example, there are N BQs (performance error events) since the last gas refill, and each performance error event BQ generates 32 scores through the fault marking submodule 360A, resulting in an [N×32] matrix, as shown in FIG. Figure 11A In other words, the fault marking submodule 360A performs 32 analyses, one for each error event file associated with the error event BQ. Figure 11A Each column in represents the likelihood score associated with each of the 32 analyses, and these likelihood scores have been adjusted so that a score greater than 1 means that the associated fault marker is believed to be present. In this example, BQ1 generates a score of 1.54 for the F101 fault marker, a score of 0.29 for the F102 fault marker, a score of 0.03 for the F103 fault marker, and so on. The goal is to transform this [N × 32] matrix into a [1 × 32] subarray κ FSD , [1×32] subarray κ FSD It can be cascaded with other sub-arrays to form an array κ which is used as the performance metric 307 .

[0160] Figure 11B FIGURE 3 illustrates how the interface module 360 ​​uses these likelihood scores. Specifically, the interface module 360 ​​(via the fault marking submodule 360A or the output submodule 362) thresholds each score so that Figure 11B Each element in the matrix represents a binary decision; that is, whether a fault marker is present or not. Next, the columns are summed so that each element represents the total number of times each fault marker appears in all error event files within a particular window. Each element is then divided by the total number of error event files so that each element represents a score indicating how prevalent each error marker is in the sample of error events BQ, and the final form is the subarray κ FSD .like Figure 11BAs shown, if the score is greater than or equal to 1.00, the score is assigned a value of 1, and if the score is less than 1.00, the score is assigned a value of 0.

[0161] In some implementations, the linear array κ that can be used as the performance metric 307 for the learning model 125 is composed not only of each subarray κ meta and κ FSD , and further includes a third subarray κ associated with a set of configuration changes of the optical system 105 config : For example, the linear array κ may correspond to:

[0162] κ=[κ meta κ FSD κ config ]

[0163] If configuration changes are used during training to form the learning model 125, then the third subarray κ config can be included in the performance metrics 107 because it means that the learning model 125 is equipped to analyze not only whether a proposed gas refill will improve the operation of the optical system 105, but also whether certain changes to the configuration of the optical system 105 will improve the operation of the optical system 105. As an example, the gas refill appears to reduce error events associated with one or more performance conditions, but the reduction in error events is actually due to a change in the configuration of the optical system 105 (occurring in time with or overlapping with the gas refill during training). The learning model 125 can distinguish between which change (gas refill or configuration change) or both changes (gas refill and configuration change) will lead to an improvement in the operation of the optical system 105.

[0164] In this example, the subarray κ config The array [κ c1 , κ c2 ,…κ c11 ] indicates that each κ ci This corresponds to a change in the configuration parameter (Δcpi) of the optical system 105. Figure 12 The table shows the κ for the optical system 105. ci For example, the optical system 105 may be Figure 3 The optical system 305 shown, or as Figure 5 The optical system 505 is shown. For example, Figure 12 In the table, κ c1 corresponds to a change in the concentration of one of the components within the chamber 323 / 523A / 523B of the corresponding gas subsystem 322 / 522A / 522B, respectively; c2corresponds to a change in the concentration of another of the components within the chamber 323 / 523A / 523B of the corresponding gas subsystem 322 / 522A / 522B; c3 corresponds to a change in properties related to how the components are injected into the chambers 323 / 523A / 523B of the respective subsystems 322 / 522A / 522B; and c8 Corresponding to a change in target temperature associated with a component of optical system 305 or 505 .

[0165] Thus, in this example, when determining the effectiveness 129 of gas refilling or configuration changes to the optical system 105, the decision module 127 also considers these configuration changes. For example, as discussed above, gas refilling can reduce the error rate of one of the error events BQ that is most relevant to gas operating conditions (such as dropouts on the master oscillator system 522A). However, the reduction in the error rate of error events BQ (in this case, dropouts on the master oscillator system 522A) can also be driven by configuration changes other than gas refilling, or regardless of gas refilling. For example, a change in the temperature of the chamber 523A of the master oscillator system 522A can result in a reduction in the error rate of MO dropouts on the master oscillator system 522A.

[0166] Other implementations are within the scope of the following claims.

[0167] For example, in other implementations, the learning model 125 includes a neural network, a decision tree, a K-nearest neighbor model (instead of a support vector machine), or any other machine learning model with a similar input / output structure. Aspects of the learning model include: assembling a dataset, assembling an input vector with fault signature detection (FSD), training the dataset, testing the dataset, and then applying the learning model.

[0168] Further aspects of the invention are set out in the following numbered clauses.

[0169] 1. A device comprising:

[0170] The decision module is configured as follows:

[0171] receiving a performance metric related to a performance condition of an optical system that transmits the light beam;

[0172] estimating the effectiveness of proposed changes to the optical system based on a performance metric and a predetermined learning model; and

[0173] If the proposed change to the optical system is estimated to be effective, a change command is output to the optical system.

[0174] 2. The apparatus according to clause 1, wherein the decision module is configured to output a maintain command to the optical system if the decision module estimates that the proposed change to the optical system is ineffective.

[0175] 3. The apparatus according to clause 1, wherein the decision module is configured to estimate the effectiveness of the proposed change by determining whether the performance condition of the beam is improved.

[0176] 4. The apparatus according to clause 3, wherein determining whether the performance condition of the beam is improved comprises determining whether an error rate of the beam quality is reduced.

[0177] 5. The apparatus according to clause 1, wherein the performance condition comprises one or more of: a type of error of the beam quality of the beam, a number of discharge events occurring in the gas mixture over a period of time, one or more faults associated with the error of the beam quality of the beam, the beam quality of the beam, the error of the beam quality of the beam, an error event associated with an operating parameter or characteristic of the beam, an analysis of each error event relative to a fault flag, and a change in configuration of the optical system.

[0178] 6. The apparatus according to clause 1, wherein the decision module is configured to estimate the effectiveness of the proposed change before implementing the change to the optical system.

[0179] 7. The apparatus according to clause 1, further comprising an interface module in communication with the decision module, the interface module providing the performance metric, wherein the interface module comprises a plurality of beam quality detection modules, the plurality of beam quality detection modules comprising:

[0180] one or more spectral feature detection modules, each spectral feature detection module configured to detect an error of a respective spectral feature of the beam and to generate an error event signal indicative of the error of the respective spectral feature of the beam; and

[0181] an energy detection module configured to detect an error of an energy of the beam and to generate an error event signal indicative of the error of the energy of the beam;

[0182] wherein the spectral feature comprises a bandwidth or a wavelength of the beam, and wherein the performance metric is generated based on the error events of the respective spectral feature and the energy.

[0183] 8. The apparatus according to clause 7, wherein the interface module comprises:

[0184] a discharge count detection module configured to detect occurrences of discharge events in the gas mixture of the optical system and to generate a signal indicative of a count of the discharge events over a period of time, wherein the performance metric comprises data related to the signal generated from the discharge count detection module.

[0185] 9. The apparatus of clause 8, wherein the time period is measured from when the gas mixture was last refilled or from when one or more chambers containing the gas mixture were added to the optical system.

[0186] 10. The apparatus of clause 7, wherein the interface module comprises:

[0187] a fault signature module configured to analyze each beam quality error event with respect to a set of fault signatures and to produce a likelihood score that classifies the beam quality error event as a known fault signature;

[0188] wherein the performance metric comprises data related to the output from the fault signature module.

[0189] 11. The apparatus of clause 1, wherein the predetermined learning model receives the performance metric as input and outputs an estimate.

[0190] 12. The apparatus of clause 1, wherein the predetermined learning model is a support vector machine.

[0191] 13. The apparatus of clause 12, wherein the predetermined learning model comprises a separating hyperplane that classifies the performance metric as yes or no, wherein a yes classification indicates that the proposed change is effective and a no classification indicates that the proposed change is not effective.

[0192] 14. The apparatus of clause 1, wherein the predetermined learning model is constructed based on the type, configuration, and / or age of the optical system.

[0193] 15. The apparatus of clause 1, wherein the estimate of the effectiveness of the proposed change indicates whether the performance condition of the optical system after the proposed change is improved relative to the condition of the optical system before the proposed change.

[0194] 16. The apparatus of clause 1, wherein the decision module is configured to estimate the effectiveness of the proposed change by comparing the performance condition of the optical system after the proposed change to the performance condition of the optical system before the proposed change, wherein the decision module is configured to output the change command if the comparison indicates that the performance condition is improved by a predetermined amount.

[0195] 17. The apparatus of clause 1, wherein the proposed change to the optical system comprises one or more of a proposed refilling of a gas mixture within the optical system and a proposed change to the configuration of the optical system.

[0196] 18. The apparatus of clause 17, wherein the decision module is configured to estimate the effectiveness of the proposed refilling of the gas mixture within the optical system based also on the detected change to the configuration of the optical system.

[0197] 19. A method comprising:

[0198] receiving a training data set based on a plurality of test optical systems, the training data set including, for each change of a plurality of changes to each test optical system:

[0199] a plurality of performance condition values related to the test optical system prior to the change; and

[0200] a plurality of performance condition values related to the test optical system after the change; and

[0201] generating a prediction model based on the training data set, the prediction model estimating effectiveness of a proposed change to an optical system based on a performance metric related to a performance condition of the optical system.

[0202] 20. The method of clause 19, wherein generating the prediction model comprises, for each change, comparing the plurality of performance condition values related to the test optical system after the change and the plurality of performance condition values related to the test optical system prior to the change, a result of the comparison indicating effectiveness of the change.

[0203] 21. The method of clause 19, wherein the prediction model is a learning model.

[0204] 22. The method of clause 21, wherein the learning model comprises a support vector machine.

[0205] 23. The method of clause 19, wherein generating the prediction model comprises mapping the performance metric to one of a maintain command or a change command, the maintain command and the change command based on an estimated effectiveness of the proposed change to the optical system.

[0206] 24. The method of clause 19, wherein the training data set comprises at least several thousand changes from the plurality of test optical systems.

[0207] 25. The method of clause 19, wherein the performance condition comprises one or more of: a beam quality error rate of a beam produced from the optical system, a type of beam quality error of a beam produced from the optical system, a number of electrical discharge events occurring in a gas mixture of the optical system over a period of time, one or more faults associated with an error in beam quality of a beam produced from the optical system, an anomaly in operating efficiency of the optical system, an error in one or more spectral characteristics of a beam produced from the optical system, an error event associated with an operating parameter or characteristic of the beam, an analysis of each error event relative to a fault signature, and a change in configuration of the optical system.

[0208] 26. The method of clause 19, further comprising: testing the predictive model before applying the performance metric related to the performance condition of the optical system to the predictive model, wherein testing the predictive model comprises:

[0209] using a test data set, for each change in the plurality of changes and each test optical system in the plurality of test optical systems, the test data set including: a plurality of performance status values ​​associated with the test optical system before the change and a plurality of performance status values ​​associated with the test optical system after the change, the test data set being excluded from the training data set; and

[0210] A plurality of performance condition values ​​associated with the test optical system before the change are applied to the prediction model, and each actual output of the prediction model is compared with an associated performance condition value from the test data set associated with the test optical system after the change.

[0211] 27. The method of clause 19, wherein the estimate of the effectiveness of the proposed change indicates whether the performance condition of the optical system after the proposed change is improved relative to the condition of the optical system before the proposed change.

[0212] 28. The method of clause 19, further comprising: adjusting the prediction model to reduce the likelihood that the prediction model estimates that the proposed change is ineffective.

[0213] 29. The method of clause 19, wherein generating the predictive model is further based on the type, configuration and / or age of the optical system.

[0214] 30. The method of clause 19, wherein the predictive model is configured to estimate the effectiveness of proposed changes to the optical system.

[0215] 31. The method according to clause 19, wherein the proposed change to the optical system comprises one or more of: refilling of a gas mixture within the proposed optical system and a change of a configuration of the proposed optical system.

[0216] 32. The method of clause 19, wherein the optical system is different from the test optical system.

[0217] 33. A method comprising:

[0218] receiving a performance metric related to a performance condition of an optical system that transmits the light beam;

[0219] estimating the effectiveness of proposed changes to the optical system based on a performance metric and a predetermined learning model; and

[0220] If the proposed changes are estimated to be effective, changes to the optical system are directed.

[0221] 34. The method of clause 33, wherein instructing the change to the optical system comprises outputting a change command to the optical system.

[0222] 35. The method of clause 33, further comprising delaying the change if the proposed change is estimated to be ineffective.

[0223] 36. The method of clause 35, wherein delaying the change comprises outputting a maintain command to the optical system.

[0224] 37. The method of clause 35, further comprising, after delaying the change:

[0225] receiving a performance metric related to a performance condition of the optical system; and

[0226] estimating, based on the performance metric and a predetermined learning model, an effectiveness of the proposed change to the optical system.

[0227] 38. The method of clause 33, wherein estimating the effectiveness of the proposed change comprises determining whether the performance condition of the optical beam is improved.

[0228] 39. The method of clause 38, wherein determining whether the performance condition of the optical beam is improved comprises determining whether an error rate of the beam quality is reduced.

[0229] 40. The method of clause 33, wherein the performance condition comprises one or more of: a type of beam quality error of the optical beam, a number of discharge events in the gas mixture within a certain time period, one or more faults associated with an error of the beam quality of the optical beam, a beam quality of the optical beam, an error of the beam quality of the optical beam, an error event associated with an operating parameter or characteristic of the optical beam, an analysis of each error event relative to a fault flag, and a change in configuration of the optical system.

[0230] 41. The method of clause 33, wherein the predetermined learning model is configured to estimate the effectiveness of the proposed change prior to implementing the change to the optical system.

[0231] 42. The method of clause 33, wherein the performance condition comprises a count of discharge events in the gas mixture within a certain time period.

[0232] 43. The method of clause 33, wherein the predetermined learning model receives the performance metric as an input and outputs an estimate.

[0233] 44. The method of clause 33, wherein the predetermined learning model is a support vector machine.

[0234] 45. The method of clause 44, wherein the predetermined learning model comprises a separating hyperplane that classifies the performance metric as yes or no, wherein a yes classification indicates that the proposed change is effective and a no classification indicates that the proposed change is not effective.

[0235] 46. ​​The method of clause 44, wherein the estimate of the effectiveness of the proposed change indicates whether the condition of the optical system after the proposed change is improved relative to the condition of the optical system before the proposed change.

[0236] 47. The method according to clause 33, wherein the proposed change to the optical system comprises one or more of: refilling of a gas mixture within the proposed optical system and a change of a configuration of the proposed optical system.

[0237] 48. Method according to clause 47, wherein the estimation of the effectiveness of the refilling of the gas mixture within the proposed optical system is also based on possible changes to the configuration of the optical system.

[0238] 49. A non-transitory computer-readable medium storing instructions that, when executed by a computer, cause the computer to perform the following method, the method comprising:

[0239] A training data set based on a plurality of test optical systems is received, wherein for each of a plurality of variations of each of the test optical systems, the training data set comprises:

[0240] a plurality of performance condition values ​​associated with the test optical system prior to the alteration; and

[0241] a plurality of performance condition values ​​associated with the test optical system after the alteration; and

[0242] A predictive model is generated based on the training data set, the predictive model estimating the effectiveness of proposed changes to the optical system based on performance metrics related to the performance condition of the optical system.

[0243] 50. A non-transitory computer-readable medium storing instructions that, when executed by a computer, perform the following method, the method comprising:

[0244] receiving a performance metric related to a performance condition of an optical system that transmits the light beam;

[0245] estimating the effectiveness of proposed changes to the optical system based on a performance metric and a predetermined learning model; and

[0246] If the proposed changes are estimated to be effective, changes to the optical system are directed.

[0247] 51. An apparatus comprising:

[0248] The decision module is configured as follows:

[0249] receiving a performance metric related to a performance condition of an optical system that transmits the light beam;

[0250] estimating the effectiveness of refilling of the gas mixture of the proposed optical system based on the performance metric and the predetermined learning model; and

[0251] If it is estimated that refilling of the proposed optical system with the gas mixture is effective, a refilling command is output to the optical system.

[0252] 52. A method comprising:

[0253] A training data set based on a plurality of test optical systems is received, wherein for each gas refill of a plurality of gas refills for each test optical system, the training data set comprises:

[0254] a plurality of performance condition values ​​associated with the tested optical system prior to gas refill; and

[0255] a plurality of performance condition values ​​associated with the tested optical system after gas refill; and

[0256] A prediction model is generated based on the training data set, the prediction model estimating the effectiveness of gas refilling of the gas mixture in the proposed optical system based on a performance metric related to the performance condition of the optical system.

[0257] 53. A method comprising:

[0258] receiving a performance metric related to a performance condition of an optical system that transmits the light beam;

[0259] estimating the effectiveness of gas refilling of the gas mixture of the proposed optical system based on the performance metric and the predetermined learning model; and

[0260] If the proposed gas refill is estimated to be effective, gas refill of the optical system is directed.

[0261] 54. A non-transitory computer-readable medium storing instructions that, when executed by a computer, cause the computer to perform the following method, the method comprising:

[0262] A training data set based on a plurality of test optical systems is received, wherein for each gas refill of a plurality of gas refills for each test optical system, the training data set comprises:

[0263] a plurality of performance condition values ​​associated with the tested optical system prior to gas refill; and

[0264] a plurality of performance condition values related to the test optical system after the gas refill; and

[0265] generating a prediction model based on the training data set, the prediction model estimating the effectiveness of the proposed gas refill in the optical system based on the performance metric related to the performance condition of the optical system.

[0266] 55. A non-transitory computer-readable medium storing instructions that, when executed by a computer, cause the computer to perform a method comprising:

[0267] receiving a performance metric related to a performance condition of an optical system emitting a light beam;

[0268] estimating, based on the performance metric and a predetermined learning model, the effectiveness of a proposed gas refill in the optical system; and

[0269] if the proposed gas refill is estimated to be effective, directing the gas refill of the optical system.

[0270] The breadth and scope of the present disclosure should not be limited by any of the above-described exemplary embodiments, but should be defined only in accordance with the following claims and their equivalents.

Claims

1. A device for controlling an optical system, comprising: a decision module in communication with the optical system and the interface module, the decision module including a predetermined learning model and configured to: receiving, from the interface module, a performance metric related to a performance condition of the optical system that transmits the light beam; estimating the effectiveness of a proposed change to the optical system based on the performance metric and the predetermined learning model; as well as outputting a change command to the optical system if the proposed change to the optical system is estimated to be effective; Wherein the predetermined learning model receives the performance metric as input and outputs the estimate. 2 . The apparatus of claim 1 , wherein the decision module is configured to output a maintain command to the optical system if the decision module estimates that the proposed change to the optical system is invalid. 3 . The apparatus of claim 1 , wherein the decision module is configured to estimate the effectiveness of the proposed change by determining whether the performance condition of the light beam is improved.

4. The apparatus of claim 3 , wherein determining whether the performance condition of the light beam is improved comprises: It is determined whether the error rate of the beam quality is reduced.

5. The apparatus of claim 1 , wherein the performance condition comprises one or more of: a type of beam quality error of the optical beam, a number of discharge events occurring in the gas mixture within a certain period of time, one or more faults associated with an error in the beam quality of the optical beam, the beam quality of the optical beam, an error in the beam quality of the optical beam, error events associated with an operating parameter or characteristic of the optical beam, an analysis of each error event with respect to a fault signature, and a change in the configuration of the optical system. 6 . The apparatus of claim 1 , wherein the decision module is configured to: evaluate the effectiveness of a proposed change to the optical system before implementing the change.

7. The apparatus of claim 1 , further comprising the interface module in communication with the decision module and the optical system, the interface module providing the performance metric, wherein the interface module comprises a plurality of beam quality detection modules, the plurality of beam quality detection modules comprising: one or more spectral feature detection modules, each spectral feature detection module configured to: detect an error in a corresponding spectral feature of the light beam and generate an error event signal indicative of the error in the corresponding spectral feature of the light beam; as well as an energy detection module configured to: detect an error in the energy of the light beam and generate an error event signal indicating the energy error of the light beam; Wherein the spectral characteristic comprises a bandwidth or wavelength of the light beam, and wherein the performance metric is generated based on error events of the corresponding spectral characteristic and energy.

8. The apparatus according to claim 7, wherein the interface module comprises: A discharge count detection module is configured to detect occurrences of discharge events in a gas mixture of the optical system and generate a signal indicative of a count of discharge events over a period of time, wherein the performance metric includes data related to the signal generated from the discharge count detection module.

9. The apparatus of claim 8, wherein the period is measured from the last time the gas mixture was refilled, or from the time one or more chambers containing the gas mixture were added to the optical system.

10. The apparatus according to claim 7, wherein the interface module comprises: a fault signature module, interacting with aspects of the optical system, configured to: analyze each beam quality error event relative to a set of fault signatures and generate a likelihood score classifying the beam quality error event as a known fault signature; Wherein the performance metric includes data related to output from the fault-flagging module. The apparatus according to claim 1 , wherein the predetermined learning model is a support vector machine.

12. The apparatus of claim 11, wherein the predetermined learning model comprises a separating hyperplane that classifies the performance metric as yes or no, wherein a yes classification indicates that the proposed change is effective and a no classification indicates that the proposed change is not effective.

13. The apparatus according to claim 1, wherein the predetermined learning model is constructed based on a type, configuration and / or age of the optical system.

14. An apparatus according to claim 1, wherein the decision module is configured to estimate the effectiveness of the proposed change by comparing the performance condition of the optical system after the proposed change with the performance condition of the optical system before the proposed change, wherein the decision module is configured to output the change command if the comparison result indicates that the performance condition is improved by a predetermined amount.

15. The apparatus of claim 1 , wherein the proposed changes to the optical system comprise: One or more of a proposed refilling of a gas mixture within the optical system and a proposed change of a configuration of the optical system.

16. The apparatus of claim 15, wherein the decision module is configured to estimate the effectiveness of the proposed refilling of the gas mixture within the optical system further based on the detected change to the configuration of the optical system.

17. A device for controlling an optical system, comprising: a decision module in communication with the optical system and the interface module, the decision module including a predetermined learning model and configured to: receiving, from the interface module, a performance metric related to a performance condition of the optical system that transmits the light beam; estimating the effectiveness of a proposed change to the optical system based on the performance metric and the predetermined learning model, wherein the estimation of the effectiveness of the proposed change indicates whether a performance condition of the optical system after the proposed change is improved relative to a condition of the optical system before the proposed change; as well as If the proposed change to the optical system is estimated to be effective, a change command is output to the optical system.

18. A method for controlling an optical system, comprising: A training data set based on a plurality of test optical systems is received, the training data set comprising, for each of a plurality of variations of each test optical system: a plurality of performance condition values ​​associated with the test optical system prior to the changing; and a plurality of performance condition values ​​associated with the test optical system after the changing; and A predictive model is generated based on the training data set, the predictive model estimating the effectiveness of proposed changes to the optical system based on a performance metric related to a performance condition of the optical system.

19. The method of claim 18, wherein generating the prediction model comprises: For each change, the plurality of performance condition values ​​associated with the test optical system after the change are compared with the plurality of performance condition values ​​associated with the test optical system before the change, the comparison result indicating the effectiveness of the change.

20. The method of claim 18, wherein the predictive model is a learned model.

21. The method of claim 20, wherein the learning model comprises a support vector machine.

22. The method of claim 18, wherein generating the prediction model comprises: The performance metric is mapped to one of a maintain command or a change command, the maintain command and the change command being based on the estimated effectiveness of the proposed change to the optical system.

23. The method of claim 18, wherein the training data set comprises: At least several thousand variations from the plurality of test optical systems.

24. The method of claim 18, wherein the performance condition comprises one or more of: a beam quality error rate of the optical beam generated from the optical system, a type of beam quality error of the optical beam generated from the optical system, a number of discharge events occurring in a gas mixture of the optical system within a certain period of time, one or more faults associated with errors in beam quality of the optical beam generated from the optical system, an anomaly in the operating efficiency of the optical system, errors in one or more spectral characteristics of the optical beam generated from the optical system, error events associated with operating parameters or characteristics of the optical beam, analysis of each error event with respect to a fault signature, and a change in the configuration of the optical system.

25. The method of claim 18, further comprising: Before applying the performance metric related to the performance condition of the optical system to the predictive model, testing the predictive model, wherein testing the predictive model comprises: using a test data set, for each change in a plurality of changes and each test optical system in a plurality of test optical systems, the test data set including: a plurality of performance status values ​​associated with the test optical system before the change and a plurality of performance status values ​​associated with the test optical system after the change, the test data set being excluded from the training data set; and The plurality of performance condition values ​​associated with the test optical system before the change are applied to the prediction model, and each actual output of the prediction model is compared with the associated performance condition value from the test data set associated with the test optical system after the change.

26. The method of claim 18, wherein the estimate of the effectiveness of the proposed change indicates whether a performance condition of the optical system after the proposed change is improved relative to a condition of the optical system before the proposed change.

27. The method of claim 18, further comprising: The predictive model is adjusted to reduce the likelihood that the predictive model estimates that the proposed change is ineffective.

28. The method of claim 18, wherein generating the predictive model is further based on the type, configuration, and / or age of the optical system.

29. The method of claim 18, wherein the predictive model is configured to estimate the effectiveness of the proposed changes to the optical system.

30. The method of claim 18, wherein the proposed changes to the optical system comprise: One or more of a proposed refilling of a gas mixture within the optical system and a proposed change of a configuration of the optical system.

31. The method of claim 18, wherein the optical system is different from the test optical system.

32. A method for controlling an optical system, comprising: receiving a performance metric related to a performance condition of the optical system that transmits the light beam; Based on the performance metric and a predetermined learning model, evaluating the effectiveness of a proposed change to the optical system, the evaluating comprising: determining whether the performance condition of the light beam is improved; as well as If the proposed change is estimated to be effective, changes to the optical system are directed.

33. The method of claim 32, wherein directing changes to the optical system comprises: A change command is output to the optical system.

34. The method of claim 32, further comprising: If the proposed change is estimated to be ineffective, the change is delayed.

35. The method of claim 34, wherein delaying the change comprises: A maintenance command is output to the optical system.

36. The method of claim 34, further comprising: After delaying the changes: receiving a performance metric related to a performance condition of the optical system; as well as Based on the performance metric and the predetermined learning model, the effectiveness of the proposed changes to the optical system is evaluated.

37. The method of claim 32, wherein determining whether the performance condition of the light beam is improved comprises: It is determined whether the error rate of the beam quality is reduced.

38. The method of claim 32, wherein the performance condition comprises one or more of: a type of beam quality error of the optical beam, a number of discharge events occurring in the gas mixture within a time period, one or more faults associated with an error in the beam quality of the optical beam, the beam quality of the optical beam, an error in the beam quality of the optical beam, error events associated with an operating parameter or characteristic of the optical beam, an analysis of each error event relative to a fault signature, and a change in the configuration of the optical system.

39. The method of claim 32, wherein the predetermined learning model is configured to estimate the effectiveness of the proposed change to the optical system before implementing the change.

40. The method of claim 32, wherein the performance condition comprises: A count of discharge events in a gas mixture over a period of time.

41. The method of claim 32, wherein the predetermined learning model receives the performance metric as input and outputs the estimate.

42. The method of claim 32, wherein the predetermined learning model is a support vector machine.

43. The method of claim 42, wherein the predetermined learning model comprises a separating hyperplane that classifies the performance metric as yes or no, wherein a yes classification indicates that the proposed change is effective and a no classification indicates that the proposed change is not effective.

44. The method of claim 42, wherein the estimate of the effectiveness of the proposed change indicates whether a condition of the optical system after the proposed change is improved relative to a condition of the optical system before the proposed change.

45. The method of claim 32, wherein the proposed changes to the optical system comprise: One or more of a proposed refilling of a gas mixture within the optical system and a proposed change of a configuration of the optical system.

46. ​​The method of claim 45, wherein estimating the effectiveness of the proposed refilling of the gas mixture within the optical system is further based on possible changes to the configuration of the optical system.

47. A non-transitory computer-readable medium storing instructions that, when executed by a computer, cause it to perform the following method, the method comprising: A training data set based on a plurality of test optical systems is received, the training data set comprising, for each of a plurality of variations of each test optical system: a plurality of performance condition values ​​associated with the test optical system prior to the changing; and a plurality of performance condition values ​​associated with the test optical system after the changing; and A predictive model is generated based on the training data set, the predictive model estimating the effectiveness of proposed changes to the optical system based on a performance metric related to a performance condition of the optical system.

48. A non-transitory computer-readable medium storing instructions that, when executed by a computer, perform the following method, the method comprising: receiving a performance metric related to a performance condition of an optical system that transmits the light beam; estimating the effectiveness of a proposed change to the optical system based on the performance metric and a predetermined learning model; as well as directing a change to said optical system if said proposed change is estimated to be effective; Wherein the predetermined learning model receives the performance metric as input and outputs the estimate.

49. A device for controlling an optical system, comprising: A decision module is in communication with the optical system and the interface module, and is configured to: receiving, from the interface module, a performance metric related to a performance condition of the optical system that transmits the light beam; estimating effectiveness of the proposed refilling of the optical system with a gas mixture based on the performance metric and a predetermined learning model; as well as If the proposed refilling of the gas mixture of the optical system is estimated to be effective, a refilling command is output to the optical system.

50. A method for controlling an optical system, comprising: A training data set based on a plurality of test optical systems is received, for each gas refill of a plurality of gas refills for each test optical system, the training data set comprising: a plurality of performance condition values ​​associated with the test optical system prior to the gas refill; and a plurality of performance condition values ​​associated with the test optical system after the gas refill; and A prediction model is generated based on the training data set, the prediction model estimating effectiveness of gas refilling of a proposed gas mixture in the optical system based on a performance metric related to a performance condition of the optical system.

51. A method for controlling an optical system, comprising: receiving a performance metric related to a performance condition of the optical system that transmits the light beam; estimating effectiveness of gas refilling of the proposed gas mixture of the optical system based on the performance metric and a predetermined learning model; as well as If the proposed gas refill is estimated to be effective, gas refill of the optical system is directed.

52. A non-transitory computer-readable medium storing instructions that, when executed by a computer, cause it to perform the following method, the method comprising: A training data set based on a plurality of test optical systems is received, for each gas refill of a plurality of gas refills for each test optical system, the training data set comprising: a plurality of performance condition values ​​associated with the test optical system prior to the gas refill; and a plurality of performance condition values ​​associated with the test optical system after the gas refill; and A prediction model is generated based on the training data set, the prediction model estimating the effectiveness of a proposed gas refill in the optical system based on a performance metric related to a performance condition of the optical system.

53. A non-transitory computer-readable medium storing instructions that, when executed by a computer, cause it to perform the following method, the method comprising: receiving a performance metric related to a performance condition of an optical system that transmits the light beam; estimating effectiveness of the proposed gas refilling in the optical system based on the performance metric and a predetermined learning model; as well as If the proposed gas refill is estimated to be effective, gas refill of the optical system is directed.

Images