Laser management server and laser management method

Abstract

The invention provides a laser management server and a laser management method. The laser management server includes: a transmission / reception processor that receives an inquiry and transmits an inquiry response; an inquiry input processor that receives and decomposes the inquiry, and generates a first inquiry item requiring external information and a second inquiry item not requiring external information; an agent behavior processor that receives the first inquiry item, acquires required external information from the unstructured data of the laser device and the structured data of the laser device, and generates an agent response to the first inquiry item; a response output processor that receives the second inquiry item, receives an agent response, and generates an inquiry item cue related to the second inquiry item; and the large-scale language model processor receives the inquiry item cues from the response output processor, generates inquiry item responses, receives the inquiry item responses by the response output processor, forms inquiry responses and sends the inquiry responses to the transceiving processor.

Description

Technical Field

[0001] The present disclosure relates to a laser management server and a laser management method. Background Art

[0002] In recent years, semiconductor exposure equipment has been required to achieve higher resolution as semiconductor integrated circuits become increasingly miniaturized and highly integrated. Consequently, there has been a trend toward shorter wavelengths of light emitted from exposure light sources. For example, gas lasers used for exposure include KrF excimer lasers, which output laser light with a wavelength of approximately 248 nm, and ArF excimer lasers, which output laser light with a wavelength of approximately 193 nm.

[0003] The spectral line width of the natural oscillation light of KrF excimer laser devices and ArF excimer laser devices is relatively wide, ranging from 350 to 400 pm. Therefore, when a projection lens is constructed using a material that transmits ultraviolet light such as KrF and ArF laser light, chromatic aberration may sometimes occur. As a result, the resolution may be reduced. Therefore, it is necessary to narrow the spectral line width of the laser light output from the gas laser device to a level where chromatic aberration is invisible. Therefore, in order to narrow the spectral line width, a narrowing module (Line Narrowing Module: LNM) containing narrowing elements (etalon, grating, etc.) is sometimes included in the laser resonator of the gas laser device. Hereinafter, a gas laser device with a narrowed spectral line width is referred to as a narrowed gas laser device.

[0004] Patent Document 1: Japanese Patent Application Laid-Open No. 2022-19441

[0005] Patent Document 2: Japanese Patent Application Laid-Open No. 1-216299

[0006] Patent Document 3: International Publication No. 2020 / 161865 Summary of the Invention

[0007] A laser management server for a laser device according to one aspect of the present disclosure comprises: a transceiver processor that receives a query from the outside and sends a query response as an answer to the query to the outside; a query input processor that receives a query from the transceiver processor, decomposes the query, and generates a first query item requiring external information and a second query item not requiring external information; an agent behavior processor that receives the first query item from the query input processor, obtains the required external information from unstructured data including manuals and maintenance reports of the laser device and structured data including working data of the laser device, and generates an agent response as a response to the first query item; a response output processor that receives the second query item from the query input processor, receives the agent response from the agent behavior processor, and generates a query item prompt word related to the second query item; and a large-scale language model processor that receives the query item prompt word from the response output processor, generates a query item response as a response to the query item prompt word, the response output processor receives the query item response from the large-scale language model processor, constructs a query response based on the agent response and the query item response, and sends the query response to the transceiver processor.

[0008] A laser management method for a laser device according to another aspect of the present disclosure comprises the following steps: a first step in which a transceiver processor receives an inquiry from the outside; a second step in which a query input processor decomposes the inquiry received from the transceiver processor to generate a first query item requiring external information and a second query item not requiring external information; a third step in which an agent behavior processor receives the first query item from the query input processor, obtains required external information from unstructured data including manuals and maintenance reports of the laser device and structured data including working data of the laser device, and generates an agent response as a response to the first query item; a fourth step in which a response output processor receives the second query item from the query input processor, receives the agent response from the agent behavior processor, and generates an inquiry item prompt word related to the second query item; a fifth step in which a large-scale language model processor receives the inquiry item prompt word from the response output processor, and generates an inquiry item response as a response to the inquiry item prompt word; a sixth step in which the response output processor receives the inquiry item response from the large-scale language model processor, and constructs an inquiry response based on the agent response and the inquiry item response; and a seventh step in which the transceiver processor receives the inquiry response from the response output processor and sends the inquiry response to the outside. BRIEF DESCRIPTION OF THE DRAWINGS

[0009] Several embodiments of the present disclosure are described below by way of example only with reference to the accompanying drawings.

[0010] Figure 1 is a diagram showing the structure of an exemplary laser device.

[0011] Figure 2 1 is a diagram showing the configuration of a laser management system according to a comparative example.

[0012] Figure 3 This is a diagram showing the laser management system according to the first embodiment.

[0013] Figure 4 This is a diagram showing the operation flow of the laser management server in the first embodiment.

[0014] Figure 5 This is a diagram showing an example of inquiry items generated in response to an inquiry from a user according to the first embodiment.

[0015] Figure 6 This is a diagram showing an example of the content of a reply to a user's inquiry according to the first embodiment.

[0016] Figure 7 This is a diagram showing a laser management system according to a modified example of the first embodiment.

[0017] Figure 8 This is a diagram showing the operation flow of the laser management server according to the modified example of the first embodiment.

[0018] Figure 9 This is a diagram showing a laser management system according to a second embodiment.

[0019] Figure 10 This is a diagram showing an example of the content of a reply to a user's inquiry according to the second embodiment. DETAILED DESCRIPTION

[0020] -Table of contents-

[0021] 1. Explanation of terms

[0022] 2. Comparative Example

[0023] 2.1 Laser device

[0024] 2.1.1 Structure

[0025] 2.1.2 Action

[0026] 2.2 Laser Management System

[0027] 2.2.1 Structure

[0028] 2.2.2 Action

[0029] 2.3 Topics

[0030] 3. Implementation Method 1

[0031] 3.1 Structure

[0032] 3.2 Action

[0033] 3.3 Function / Effect

[0034] 4. Modification of Implementation 1

[0035] 4.1 Structure

[0036] 4.2 Action

[0037] 4.3 Action / Effect

[0038] 5. Implementation Method 2

[0039] 5.1 Structure

[0040] 5.2 Action

[0041] 5.3 Action / Effect

[0042] 6. Others

[0043] Below, embodiments of the present disclosure are described in detail with reference to the accompanying drawings. The embodiments described below illustrate several examples of the present disclosure and do not limit the content of the present disclosure. In addition, the structures and actions described in each embodiment are not necessarily all required structures and actions of the present disclosure. In addition, the same reference numerals are given to the same structural elements, and repeated descriptions are omitted.

[0044] 1. Explanation of terms

[0045] The terms used in this specification are defined as follows.

[0046] “Standardized software” is application software that performs at least one of the tasks of managing, monitoring, and analyzing a specific device according to predefined functions and display specifications.

[0047] A "web application" is software that can be used on the internet using a web browser. Web applications can be used on various devices, such as personal computers, smartphones, and tablets, without installing the application software, as long as they are connected to the internet.

[0048] The "standard operation screen" is a pre-designed standard user interface (UI) in application software.

[0049] The “interactive operation screen” is an operation screen in which a user asks questions and issues commands using natural language, and the application software responds to the questions and commands from the user.

[0050] Unstructured data is data that lacks regularity and continuity and is not easily managed using conventional databases and tables. Examples of unstructured data include text documents, images, and audio files.

[0051] "Operation data" is the operational information generated by the laser device.

[0052] "Professional Information Processors" are chat agents with highly specialized knowledge in a specific field. Unlike general chat agents, these agents can accurately handle conversations using terms and concepts related to that field.

[0053] A "query" is a question or instruction received by the system from the user. It is a natural language instruction string that specifies the desired information and the format in which it is provided, or specifies an operation desired by the laser device. An example of a query might be "Report detailed operational diagnostics of laser 65400011 in a standard format."

[0054] A query response is a natural language response from the system in response to a specific query. Depending on the content, a query response may include not only strings but also graphs, images, and more.

[0055] The Large Language Model Processor (LLM) performs natural language processing based on AI (Artificial Intelligence) models trained using large amounts of text data. The LLM can understand and execute complex language tasks.

[0056] "Prompt words" are natural language text instructions or questions sent to the LLM. Based on the prompt words, the LLM understands what to respond to (answer) or what task to perform.

[0057] The "Query Input Processor (QIP)" analyzes the user's request (query) and executes a process of replacing the query with input prompt words suitable for the LLM.

[0058] The "Response Output Processor (ROP)" executes processing for composing the content returned as a reply response to the LLM into a format and content that conforms to the content specified by the user.

[0059] The Agent Action Processor (AAP) performs the following processing: it analyzes the output of the LLM, obtains the information required to form an appropriate response from an external information source, and forms an appropriate response that matches the user's query.

[0060] 2. Comparative Example

[0061] 2.1 Laser device

[0062] 2.1.1 Structure

[0063] Figure 1 1 is a diagram illustrating the structure of an exemplary laser device 10. Laser device 10 is a discharge-excited gas laser device and includes an oscillator (OSC) 20, an amplifier (AMP) 50, a monitor module 70, and a laser processor 80. The processor disclosed herein is a processing device comprising a storage device storing a control program and a CPU that executes the control program. The processor is specifically configured or programmed to perform various processes.

[0064] The OSC 20 includes a narrowbanding module (LNM) 22 , a cavity 24 , an output coupler (OC) 26 , a pulsed power module (PPM) 28 , and a charger 32 .

[0065] The LNM 22 includes prisms 36 and 38, a grating 42, and a rotation stage 44 for rotating the prism 38. The LNM 22 rotates the prism 38 to change the incident angle with respect to the grating 42, thereby controlling the center wavelength of the pulsed laser light.

[0066] The cavity 24 includes a pair of discharge electrodes 46 and 47, and two windows 48 and 49 through which the laser light passes. Excimer laser gas is introduced into the cavity 24. The excimer laser gas includes, for example, a rare gas (Ar gas or Kr gas), a halogen gas (F2 gas), and a buffer gas (Ne gas).

[0067] OC26 is a partial reflection mirror that reflects part of the pulsed laser light and transmits the other part.

[0068] The LNM 22 and the OC 26 together constitute an optical resonator, and the cavity 24 is arranged on the optical path of the optical resonator.

[0069] AMP 50 includes a rear mirror (RM) 52 , a cavity 54 , an output coupler (OC) 56 , a pulsed power module (PPM) 58 , and a charger 62 .

[0070] RM52 is a partial reflector that reflects part of the pulsed laser and transmits the other part. The reflectivity of RM52 can be 80% to 90%.

[0071] The cavity 54 includes a pair of discharge electrodes 64 and 65 and two windows 66 and 67 for transmitting laser light. Excimer laser gas is introduced into the cavity 54.

[0072] OC56 is a partial reflector that reflects a portion of the pulsed laser and transmits it. The reflectivity of OC56 can be 10% to 30%.

[0073] RM52 and OC56 together form an optical resonator, and cavity 54 is arranged on the optical path of the optical resonator. The optical resonator may be a Fabry-Perot type optical resonator.

[0074] Monitor module 70 includes beam splitters 72 and 74, a spectrum detector 76 for measuring the wavelength and spectral line width of the pulsed laser light, and a photosensor 78 for detecting the pulse energy of the pulsed laser light. Spectrum detector 76 may be an etalon spectrometer, and photosensor 78 may be a photodiode.

[0075] 2.1.2 Action

[0076] Laser processor 80 receives target central wavelength λt and target pulse energy Et from an external device such as an exposure device (not shown). Laser processor 80 then sets charging voltages V1 and V2 of charger 32 and charger 62 to obtain pulsed laser light of target pulse energy Et.

[0077] A first charging capacitor (not shown) in the PPM 28 is charged at a charging voltage V1, and a second charging capacitor (not shown) in the PPM 58 is charged at a charging voltage V2.

[0078] Upon receiving a light emission trigger Trt from an external device such as an exposure device, the laser processor 80 transmits a light emission trigger Tr1 to the switch 33 within the PPM 28. When the switch 33 is actuated, the charge stored in the first charging capacitor is converted within the PPM 28 into a high-voltage pulse corresponding to the charging voltage V1, which is then applied between the discharge electrodes 46 and 47 within the cavity 24.

[0079] As a result, discharge occurs between discharge electrodes 46 and 47 within cavity 24, exciting the laser gas. Then, the optical resonator formed by OC 26 and LNM 22 generates narrowband seed light with an ultraviolet wavelength of 150 nm to 380 nm, which is then output from OSC 20. The wavelength of the seed light can be the oscillation wavelength of an ArF excimer laser or a KrF excimer laser.

[0080] Furthermore, upon receiving the emission trigger Trt, the laser processor 80 sends the emission trigger Tr2 to the switch 59 of the PPM 58 , so that discharge occurs between the discharge electrodes 64 and 65 when the seed light output from the OSC 20 enters the discharge space of the cavity 54 of the AMP 50 .

[0081] When the switch 59 is operated, the charge stored in the second charging capacitor is converted into a high voltage pulse corresponding to the charging voltage V2 in the PPM 58 and applied between the discharge electrodes 64 and 65 in the cavity 54 .

[0082] As a result, discharge occurs between discharge electrodes 64 and 65 in cavity 54, exciting the laser gas. At this point, seed light output from OSC 20 passes through RM 52 and enters the discharge space in cavity 54. The incident seed light is amplified by the optical resonator formed by RM 52 and OC 56 and output from AMP 50.

[0083] The pulsed laser light output from AMP 50 enters monitor module 70. Part of the pulsed laser light entering monitor module 70 is reflected by beam splitter 72, and part of it is reflected by beam splitter 74 and enters spectrum detector 76. The pulsed laser light that has passed through beam splitter 74 enters optical sensor 78.

[0084] The spectrum detector 76 measures the central wavelength of the pulsed laser light, and the optical sensor 78 measures the pulse energy of the pulsed laser light.

[0085] The laser processor 80 can control the rotating stage 44 in the LNM 22 so that the central wavelength measured by the spectrum detector 76 becomes the target central wavelength λt. The rotating stage 44 can be a rotating stage including a piezoelectric element.

[0086] The laser processor 80 can control the charging voltage V2 output from the charger 62 so that the pulse energy measured by the optical sensor 78 becomes the target pulse energy Et.

[0087] 2.2 Laser Management System

[0088] 2.2.1 Structure

[0089] Figure 2 This diagram shows the structure of a comparative example laser management system 100. The comparative example disclosed herein is a method known only to the applicant and is not a publicly known example acknowledged by the applicant. The laser management system 100 performs at least one of management, monitoring, and analysis of the laser device 10.

[0090] The laser management system 100 includes a laser device 10 , a work data server 104 , a laser management server 110 , and a terminal 150 .

[0091] The operation data server 104 is a data server that stores the operation data of the laser device 10 .

[0092] The terminal 150 is, for example, a personal computer, a smartphone, or a tablet.

[0093] The laser management server 110 includes a standard software processor (SSP) 128 storing the standard software 122 .

[0094] The SSP 128 obtains the operating data of the laser device 10 from the operating data server 104 via the network.

[0095] The standardization software 122 is provided in the form of a web application that operates via a network.

[0096] The standardization operation screen 124 of the standardization software 122 is displayed on the terminal 150 and is generated in a manner that allows user operation.

[0097] 2.2.2 Action

[0098] The laser device 10 transmits the operation data to the operation data server 104 , and the operation data server 104 accumulates the operation data.

[0099] The laser management server 110 receives a request for browsing predetermined information from the user through the standard operation screen 124 .

[0100] The calibration software 122 of the laser management server 110 uses the stored operation data to perform analysis of a specific laser device 10. The calibration software 122 provides the user with at least one of the predetermined functions of management, monitoring, and analysis of the laser device 10.

[0101] The calibration software 122 of the laser management server 110 displays the requested information and the like on the calibration operation screen 124 .

[0102] The user browses information on the setting operation screen 124 .

[0103] Different standardized software 122 or standardized operation screens 124 are provided according to the needs of different users such as field service engineers (FSE), device owners, and researchers.

[0104] 2.3 Topics

[0105] At a semiconductor manufacturing site, standardized software 122 is used to manage, monitor, and maintain the operation of the laser device 10. The standardized software 122 can only display and operate based on all predetermined operating specifications.

[0106] In recent semiconductor manufacturing sites, as manufacturing processes become more complex and diverse, there is a growing demand for data analysis, monitoring, and retrieval using application software. However, it is difficult to meet diverse user needs using only predefined operations and screens in the standard software 122.

[0107] Furthermore, the data that application software can currently process is limited to structured data with clear numerical values, such as device parameters and sensor data. Furthermore, human-generated, unstructured reports, manuals, and images also contain information useful for advanced data analysis. However, this information is unstructured data with neither regularity nor continuity, making it difficult to process as analytical data.

[0108] 3. Implementation Method 1

[0109] 3.1 Structure

[0110] Figure 3 : is a diagram showing a laser management system 100A according to the first embodiment. Figure 3 The structure shown is Figure 2 The differences are explained.

[0111] Compared with the laser management system 100, the laser management system 100A is different in the structure within the laser management server 110A, the laser management server 110A and the document server 106, and the screen that the user actually processes through the terminal 150 is changed from the standard operation screen 124 to the interactive operation screen 126.

[0112] The laser management server 110A includes a sending and receiving processor (SRP) 130, a query input processor (QIP) 132, a large-scale language model processor (LLM) 134, an agent behavior processor (AAP) 136 and a response output processor (ROP) 138.

[0113] SRP130, QIP132, LLM134, AAP136, and ROP138 can be application software or hardware such as a CPU. Individual application software can also be installed as a single processor.

[0114] The user's input and output to the laser management system 100A is performed on the interactive operation screen 126 of the terminal 150 via the SRP 130 .

[0115] The document server 106 includes a laser device technical document database (Database: DB) that stores unstructured data such as technical documents such as manuals and maintenance reports of the laser device 10 .

[0116] At least one of the work data server 104 and the document server 106 may also be installed inside the laser management server 110A.

[0117] 3.2 Action

[0118] Figure 4 This is a diagram showing the operation flow of the laser management server 110A according to the first embodiment.

[0119] [Step 1] SRP 130 receives a query W0 from terminal 150. Step 1 is an example of "Step 1" in this disclosure. The method by which a user inputs query W0 into terminal 150 is not limited to inputting via a physical keyboard or an application software keyboard; voice input is also possible. Furthermore, query W0 may be input via at least one of email, text message, and voicemail, for example.

[0120] [Step 2] The QIP 132 decomposes the query W0 received from the SRP 130 to generate a first query item W1 requiring external information and a second query item W2 not requiring external information. Step 2 is an example of the "second step" in the present disclosure.

[0121] Figure 5 : is a diagram showing an example of the first inquiry item W1 and the second inquiry item W2 generated based on the inquiry W0. Figure 5 1 shows an example of query item generation when the query W0 received by the QIP 132 is "report the operation diagnosis of the laser 65400011 in a standard format." Note that "laser 65400011" is a name that identifies the model of the laser.

[0122] In this case, QIP 132 decomposes the received query W0 into four query items. Query item 1 is "Information about the model of laser 65400011," query item 2 is "Information about the standard format of the diagnostic report," query item 3 is "Analysis of the operating data of laser 65400011," and query item 4 is "Generation of a diagnostic report based on this information."

[0123] Of these four inquiry items, item numbers 1 and 3 are difficult to answer appropriately using only LLM 134 and require external information to obtain an appropriate answer. Therefore, they are classified as first inquiry item W1. Item numbers 2 and 4 do not require external information to answer and are classified as second inquiry item W2.

[0124] In this case, external information refers to information stored in the work data server 104 or the document server 106. The criterion for determining whether an inquiry item requires external information is, for example, whether LLM 134 can generate a highly accurate response independently. For inquiries classified as the first inquiry item W1, LLM 134 has difficulty generating a highly accurate response independently. For inquiries classified as the second inquiry item W2, LLM 134 can generate a highly accurate response independently. A highly accurate response is one that is based on facts and has fewer misidentifications or errors.

[0125] In addition, depending on the content of the inquiry W0 received from the SRP 130, one of the first inquiry item W1 and the second inquiry item W2 may be generated.

[0126] [Step 3] The QIP 132 sends the second inquiry item W2 to the ROP 138 .

[0127] [Step 4] The QIP 132 sends the first inquiry item W1 to the AAP 136. The first inquiry item W1 is used when confirming information, and therefore may also be sent to the ROP 138.

[0128] [Step 5] The AAP 136 assigns the corresponding external information processing program to the first query item W1. The external information processing program obtains necessary information from the outside (work data server 104, document server 106), analyzes it, and summarizes the results.

[0129] For example, for Figure 5 In the first inquiry item W1 of item number 3, AAP136 obtains the applied voltage (HV) between the discharge electrodes 46 and 47 in the cavity 24 of the OSC20 of the laser 65400011, the gas pressure in the cavity 24 of the OSC20 and the total number of device emissions from the work data server 104, analyzes these data and summarizes the results.

[0130] The information obtained by the external information processing program can include, for example, work diagnosis reports, parameter coordinate diagrams, component replacement procedures, light source technical explanations, past maintenance records, etc.

[0131] [Step 6] AAP136 executes all assigned external information processing programs. Then, AAP136 performs external information composition processing to collect all information related to the execution results, and generates an agent response W3 as a response to the first query item W1. Steps 5 and 6 are an example of "Step 3" in this disclosure. In addition, an example of the agent response W3 is described later. Figure 6 The AAP generation results 152 and 154 are not limited to character information, but may also include coordinate graphs, tables, etc.

[0132] [Step 7] AAP 136 sends the agent response W3 to ROP 138 .

[0133] [Step 8] The ROP 138 receives the second inquiry item W2 from the QIP 132 .

[0134] [Step 9] ROP 138 may also receive the agent response W3 from AAP 136 to confirm whether it contains the required information.

[0135] [Step 10] ROP 138 generates a query item prompt word W4 related to the response to the second query item W2. Steps 8 to 10 are an example of "Step 4" in this disclosure. For example, Figure 5 The question item prompt word W4 of the second question item W2 of the item number 2 is "Information of the standard format of the diagnosis report". Figure 5 The inquiry item prompt word W4 of the second inquiry item W2 of item number 4 shown includes the agent response W3 required for the response and the prompt word "generate a report on the agent response".

[0136] [Step 11] The LLM 134 receives the inquiry item presentation word W4 from the ROP 138 and generates an inquiry item response W5 as a response to the inquiry item presentation word W4. Step 11 is an example of the "fifth step" in the present disclosure.

[0137] For example, for Figure 5 The query item response W5 of the query item prompt word W4 of the second query item W2 of the item number 2 is "report the comprehensive diagnosis result and individual performance based on the operation data of the device". Figure 5 The question item response W5 of the question item presentation word W4 of the second question item W2 of the item number 4 shown is as follows, for example.

[0138] 1. Comprehensive diagnosis

[0139] Laser 65400011 is currently working in good condition.

[0140] Based on the trends of OSC cavity air pressure and HV over the past four weeks, a slight upward trend in air pressure was found.

[0141] It is premature to replace the part, however, if this trend continues, replacement may be necessary in up to three months.”

[0142] [Step 12] The ROP 138 receives the query item response W5 from the LLM 134. The ROP 138 then constructs a query response W6 based on the agent response W3 and the query item response W5. Step 12 is an example of "Step 6" in the present disclosure.

[0143] [Step 13] ROP 138 sends the constructed query response W6 to SRP 130.

[0144] [Step 14] The SRP 130 transmits the inquiry response W6 to the terminal 150. Step 14 is an example of the "seventh step" in the present disclosure.

[0145] [Step 15 ] The inquiry response W6 is displayed on the interactive operation screen 126 of the terminal 150 .

[0146] Figure 6 1 is a diagram showing an example of a query response W6 displayed on the terminal 150 . Figure 6 This is an example of the display of the inquiry response W6 when the inquiry W0 is "Report the operation diagnosis of laser 65400011 in a standard format". The laser management server 110A receives the inquiry W0 and responds to the user. Figure 6 The query response W6 of the content shown.

[0147] exist Figure 6 In the example, the query response W6 displayed on the interactive operation screen 126 includes LLM generation results 151 and 153 and AAP generation results 152 and 154. LLM generation results 151 and 153 are generated by the LLM 134. AAP generation results 152 and 154 are generated by the AAP 136. AAP generation results 152 and 154 may include graphs, tables, and the like. Furthermore, the creation of the work data graph can be performed by either the AAP 136 or the LLM 134.

[0148] In the interactive operation screen 126, the AAP generation result 152 may be displayed immediately below the LLM generation result 151. Figure 6 In this case, AAP generation result 152 is information associated with LLM generation result 151, and AAP generation result 154 is information associated with LLM generation result 153. Therefore, in order to display these associated information in an easy-to-view form on the screen, it is preferred that the query response W6 be composed of a layout in which LLM generation result 151, AAP generation result 152, LLM generation result 153 and AAP generation result 154 are arranged alternately.

[0149] ROP 138 combines the LLM generation results 151, 153 generated by LLM 134 and the AAP generation results 152, 154 received from AAP 136 to generate an implementation Figure 6 Such a screen display responds to the inquiry W6.

[0150] When the SRP 130 receives the inquiry W0 again in response to the inquiry W6 displayed on the interactive operation screen 126, the laser management server 110A again performs the processes of steps 2 to 15. Steps 1 to 15 are an example of the "laser management method" in the present disclosure.

[0151] 3.3 Function / Effect

[0152] The laser management system 100A of the first embodiment includes the LLM 134 capable of language processing in the laser management server 110A. Therefore, there is no limitation on the format of the inquiry W0 or the display method of the inquiry response W6.

[0153] The AAP 136 obtains a first query item W1 requiring external information. The obtained external information is sent to the LLM 134 via the ROP 138. Therefore, the LLM 134 can also utilize information unique to the laser device 10, such as unstructured data such as technical documentation of the laser device 10 that has not yet been learned, and structured data such as operating data of the laser device 10.

[0154] Since the AAP 136 acquires external information every time, the LLM 134 can generate the query item response W5 in a manner that includes the latest external information.

[0155] The laser management server 110A of the first embodiment can generate a query response W6 in response to a query W0 from various users such as FSE, device owner, and research and development personnel.

[0156] The first query item W1 is used by the AAP 136 to obtain external information, analyze it, and generate an agent response W3. The first query item W1, which requires external information, is then sent to the LLM 134 in the form of the agent response W3. Therefore, compared to a case where all external information is sent to the LLM 134, the laser management server 110A of Embodiment 1 receives less information than would be possible if the LLM 134 were to receive all external information, thus preventing the LLM 134 from exceeding its capacity to receive prompt words.

[0157] Furthermore, the laser management server 110A of the first embodiment does not require various standard operation screens 124 , and therefore can reduce the time and cost required for the UI development process compared to the laser management server 110 of the comparative example.

[0158] 4. Modification of Implementation 1

[0159] 4.1 Structure

[0160] Figure 7 1 is a diagram showing a laser management system 100B according to a modified example of the first embodiment. Figure 7 The structure shown is Figure 3 The differences are explained. Figure 7 The laser management server 110B shown is connected to Figure 3 Compared to the laser management server 110A shown in FIG. 1 , the QIP 132 and the LLM 134 communicate with each other, and the AAP 136 and the LLM 134 communicate with each other. Other structures can be Figure 3 The structures shown are the same.

[0161] 4.2 Action

[0162] Figure 8 This is a diagram showing the operation flow of the laser management server 110B according to the modified example of the first embodiment. Figure 8 The operation flow of the laser management server 110B shown in FIG. Figure 4 The differences are explained.

[0163] The QIP 132 of the laser management server 110B may generate a sorting prompt word W7 when performing language processing in generating the first inquiry item W1 and the second inquiry item W2. The sorting prompt word W7 may be, for example, "decompose the inquiry and divide it into items."

[0164] LLM134 can receive the sorting prompt word W7 from QIP132 and generate a sorting response W8 as a response to the sorting prompt word W7. The sorting response W8 is, for example, Figure 5 The query items are divided into such items as shown.

[0165] QIP 132 may also receive a sort response W8 from LLM 134 .

[0166] When acquiring unstructured data and creating the agent response W3, the AAP 136 may generate an unstructured data prompt W9. The unstructured data prompt W9 may include, for example, the manual of the laser device 10 and the prompt "inform me of the location where the cavity replacement method is described."

[0167] LLM 134 may receive unstructured data prompt word W9 from AAP 136 and generate unstructured data response W10 as a response to unstructured data prompt word W9. Unstructured data response W10 may be, for example, "The cavity replacement method is described on line BB of page AA to line DD of page CC."

[0168] AAP 136 can also receive an unstructured data response W10 from LLM 134. AAP 136 parses the unstructured data response W10 received from LLM 134 and generates an agent response W3. For example, when receiving the unstructured data response W10 that states "The cavity replacement method is described on line BB of page AA to line DD of page CC," AAP 136 assigns an appropriate external information processing program and generates an agent response W3 based on the result. Other actions can be performed in conjunction with Figure 4 same.

[0169] 4.3 Action / Effect

[0170] The laser management server 110B can achieve the same effects as those of the laser management server 110A. In addition, the laser management server 110B has further improved language processing functions in the QIP 132 and the AAP 136.

[0171] 5. Implementation Method 2

[0172] 5.1 Structure

[0173] Figure 9 This is a diagram showing a laser management system 100C according to a second embodiment. Figure 9 The structure shown is Figure 7 The differences are explained.

[0174] The laser management system 100C differs from the laser management system 100B in that an AI prediction processing device 108 is connected to a laser management server 110C.

[0175] The AI ​​prediction processing device 108 includes a life prediction model for predicting the life of consumables of the laser device 10 and a laser performance prediction model for predicting the future laser performance of the laser device 10. The life prediction model and the laser performance prediction model are AI models that have been trained by machine learning in a manner that performs target tasks and are installed in the AI ​​prediction processing device 108. Other structures can be Figure 7 same.

[0176] 5.2 Action

[0177] The life prediction model may be, for example, a learned model created using the machine learning method described in Patent Document 3. The machine learning method described in Patent Document 3 creates a learning model for predicting the life of consumables of the laser device 10. The machine learning method includes the following steps: obtaining first life-related information, the first life-related information including data on life-related parameters of the consumables recorded corresponding to different numbers of oscillation pulses during the period from the start of use of the consumables to their replacement; dividing the first life-related information into multiple levels representing the degree of degradation of the consumables based on the number of oscillation pulses, and creating training data that associates the first life-related information with the levels representing the degree of degradation; performing machine learning using the training data to create a learning model that predicts the degree of degradation of the consumables based on the data on the life-related parameters; and storing the created learning model.

[0178] The learning model is, for example, a neural network model, and is actually a program that causes a computer to execute a process of predicting the degree of degradation of consumables of the laser device 10 .

[0179] The AI ​​prediction processing device 108 uses the corresponding life prediction model based on the life-related information of each consumable item scheduled for replacement in the laser device 10 to predict the life of each consumable item. The prediction result based on the life prediction model is transmitted from the AAP 136 to the ROP 138 as the agent response W3.

[0180] The laser performance prediction model is a learned model that can predict the future laser performance of laser device 10 in response to the number of pulses or time in any component replacement scenario. The prediction result based on the laser performance prediction model is sent from AAP 136 to ROP 138 as agent response W3.

[0181] In the ROP 138 , the inquiry response W6 is constructed based on the agent response W3 and the inquiry item response W5 , and the inquiry response W6 is displayed on the interactive operation screen 126 of the terminal 150 .

[0182] Figure 10 This is a diagram showing an example of the inquiry response W6 displayed on the terminal 150 in the second embodiment. Figure 10 This is an example of displaying the reply content to the user as the inquiry response W6 when the inquiry W0 is "Report the operating diagnosis of laser 65400011 in a standard format". Figure 10 , for Figure 6 The differences are explained.

[0183] exist Figure 10 In, with Figure 6Compared with the above, the prediction results of the bandwidth (line width) of the pulsed laser spectrum and the gas pressure in the cavity 24 of the OSC 20 are added. Figure 10 In the interactive operation screen 126, LLM generation result 159 and AAP generation result 160 are included. LLM generation result 159 is the result generated by LLM 134. AAP generation result 160 is the result of AAP 136 using the laser performance prediction model of AI prediction processing device 108. AAP generation result 160 may include a graph, a table, etc. Furthermore, the spectral bandwidth of a pulsed laser or the gas pressure within cavity 24 of OSC 20 are examples of "laser device performance" in this disclosure.

[0184] 5.3 Action / Effect

[0185] The laser management system 100C according to Embodiment 2 achieves the same effects as those of Embodiment 1. Furthermore, the laser management system 100C can generate an agent response W3 using information obtained from the AI ​​prediction processing device 108 regarding the lifespan of each consumable item scheduled for replacement and the laser performance of the laser device 10. This improves the quality of the query response W6 to the query W0 related to the prediction.

[0186] 6. Others

[0187] The above description is not limiting but merely illustrative. Therefore, those skilled in the art will appreciate that modifications can be made to the embodiments of the present disclosure without departing from the scope of the claims. Furthermore, those skilled in the art will appreciate that the embodiments of the present disclosure can be used in combination.

[0188] Unless otherwise expressly stated, the terms used in this specification and claims as a whole should be interpreted as “non-limiting” terms. For example, terms such as “including”, “having”, “having”, and “equipped” should be interpreted as “not excluding the presence of structural elements other than the structural elements described”. In addition, the modifier “one” should be interpreted as meaning “at least one” or “one or more”. In addition, terms such as “at least one of A, B, and C” should be interpreted as “A”, “B”, “C”, “A+B”, “A+C”, “B+C”, or “A+B+C”. Furthermore, it should be interpreted as also including combinations of these and parts other than “A”, “B”, and “C”.

Claims

1. A laser management server, which is a laser management server for a laser device, having: a transceiver processor configured to receive an inquiry from the outside and transmit an inquiry response as an answer to the inquiry to the outside; a query input processor that receives the query from the transceiver processor, decomposes the query, and generates a first query item that requires external information and a second query item that does not require external information; an agent behavior processor receiving the first inquiry item from the inquiry input processor, acquiring necessary external information from unstructured data including manuals and maintenance reports of the laser device and structured data including operating data of the laser device, and generating an agent response as a response to the first inquiry item; a response output processor that receives the second inquiry item from the inquiry input processor, receives the agent response from the agent behavior processor, and generates an inquiry item prompt word related to the second inquiry item; and a large-scale language model processor that receives the query item prompt word from the response output processor and generates a query item response as a response to the query item prompt word; The response output processor receives the query item response from the large-scale language model processor, constructs the query response based on the agent response and the query item response, and sends the query response to the transceiver processor.

2. The laser management server according to claim 1, wherein: The query item prompt word includes the agent response.

3. The laser management server according to claim 1, wherein: The agent behavior processor obtains the structured data including the prediction result of the life prediction model on the consumables of the laser device.

4. The laser management server according to claim 1, wherein: The agent behavior processor obtains the structured data including the prediction results of the laser performance prediction model on the performance of the laser device.

5. The laser management server according to claim 1, wherein: The query input processor generates sorting prompt words for decomposing the query when performing language processing in generating the first query item and the second query item. The large-scale language model processor receives the sorting prompt word from the query input processor, generates a sorting response as a response to the sorting prompt word, The query input processor receives the sorted response from the large-scale language model processor.

6. The laser management server according to claim 1, wherein: The agent behavior processor generates unstructured data prompt words when obtaining the unstructured data and making the agent response, The large-scale language model processor receives the unstructured data prompt word from the agent behavior processor, and generates an unstructured data response as an answer to the unstructured data prompt word. The agent behavior processor receives the unstructured data response from the large-scale language model processor.

7. The laser management server according to claim 6, wherein: The unstructured data prompt word includes the unstructured data.

8. The laser management server according to claim 1, wherein: The response output processor combines the information generated by the large-scale language model processor and the agent response received from the agent behavior processor to form the query response.

9. The laser management server according to claim 1, wherein: The transceiver processor receives the inquiry from a terminal connected via a network and transmits the inquiry response to the terminal.

10. The laser management server according to claim 9, wherein: Accepting the inquiry from the interactive operation screen of the terminal, The inquiry response is displayed on the interactive operation screen of the terminal.

11. A laser management method for a laser device, comprising the following steps: In step 1, the transceiver processor receives an inquiry from the outside; In the second step, the query input processor decomposes the query received from the transceiver processor to generate a first query item requiring external information and a second query item not requiring external information; In step 3, the agent behavior processor receives the first inquiry item from the inquiry input processor, obtains required external information from unstructured data including manuals and maintenance reports of the laser device and structured data including operating data of the laser device, and generates an agent response as a response to the first inquiry item. In step 4, the response output processor receives the second inquiry item from the inquiry input processor, receives the agent response from the agent behavior processor, and generates an inquiry item prompt word related to the second inquiry item; In step 5, the large-scale language model processor receives the query item prompt word from the response output processor and generates a query item response as a response to the query item prompt word; In step 6, the response output processor receives the query item response from the large-scale language model processor, and constructs a query response based on the agent response and the query item response; and In step 7, the transceiver processor receives the inquiry response from the response output processor and transmits the inquiry response to the outside.

12. The laser management method according to claim 11, wherein: The third step includes the following steps: the intelligent agent behavior processor obtains the structured data including the prediction result of the life prediction model on the consumables of the laser device.

13. The laser management method according to claim 11, wherein: The third step includes the following steps: the intelligent agent behavior processor obtains the structured data including the prediction results of the laser performance prediction model on the laser performance of the laser device.

14. The laser management method according to claim 11, wherein: The third step includes the following steps: when performing language processing in the generation of the first query item and the second query item, generating sorting prompt words that decompose the query into the first query item and the second query item, the large-scale language model processor receives the sorting prompt words from the query input processor, generates a sorting response as a response to the sorting prompt words, and the query input processor receives the sorting response from the large-scale language model processor.

15. The laser management method according to claim 11, wherein: The laser management method comprises the following steps: The agent behavior processor generates unstructured data prompt words when obtaining the unstructured data and making the agent response, The large-scale language model processor receives the unstructured data prompt word from the agent behavior processor, and generates an unstructured data response as a response to the unstructured data prompt word. The agent behavior processor receives the unstructured data response from the large-scale language model processor.

16. The laser management method according to claim 15, wherein: The unstructured data prompt word includes the unstructured data.

17. The laser management method according to claim 11, wherein: The query item prompt word includes the agent response.

18. The laser management method according to claim 11, wherein: The step 6 includes the following steps: the response output processor combines the information generated by the large-scale language model processor and the agent response received from the agent behavior processor to form the query response.

19. The laser management method according to claim 11, wherein: The first step includes the following steps: the transceiver processor receives the inquiry from a terminal connected via a network, The seventh step includes the following steps: sending the inquiry response to the terminal.

20. The laser management method according to claim 19, wherein: Accepting the inquiry from the interactive operation screen of the terminal, The inquiry response is displayed on the interactive operation screen of the terminal.

Images