Information processing method, information processing device, and computer program

Abstract

Provided are an information processing method, an information processing device, and a computer program.　This computer executes processing of: receiving a query of a user related to substrate processing; generating an answer sentence to the received query of the user by using a language model referring to process data and document data related to the substrate processing; and outputting the generated answer sentence.

Description

Information Processing Method, Information Processing Apparatus, and Computer Program

[0001] The present invention relates to an information processing method, an information processing apparatus, and a computer program.

[0002] In a substrate processing apparatus, processing on a substrate is performed based on a process recipe. The recipe is composed of a plurality of steps, and for example, by controlling various parameters such as pressure and temperature for each step, an optimal processing result can be obtained. Since the set values of various parameters may differ for each step, measurement data of a plurality of sensors provided in the substrate processing apparatus are managed for each substrate.

[0003] In conventional process development, developers design various parameters for each step by referring to past measurement data or researching documents.

[0004] Japanese Patent Application Laid-Open No. 2023-000106

[0005] An object of the present disclosure is to provide an information processing method, an information processing apparatus, and a computer program that can generate a response sentence to a query for a user in consideration of process data and document data related to substrate processing.

[0006] The information processing method of the present disclosure receives a user's query related to substrate processing, generates a response sentence to the received user's query using a language model that refers to process data and document data related to substrate processing, and causes a computer to execute a process of outputting the generated response sentence.

[0007] According to the present disclosure, it is possible to generate a response sentence to a query for a user in consideration of process data and document data related to substrate processing.

[0008] This is a schematic diagram showing the configuration of an information processing system according to an embodiment. This is a block diagram showing the internal configuration of a developer terminal. This is an explanatory diagram illustrating an example of a response output to a developer's query. This is an explanatory diagram illustrating an example of a response output to a developer's query. This is an explanatory diagram illustrating an example of a response output to a developer's query. This is an explanatory diagram illustrating an example of a response output to a developer's query. This is a flowchart illustrating the processing procedure executed by the developer terminal according to Embodiment 1. This is an explanatory diagram illustrating the information extraction procedure in Embodiment 2. This is an explanatory diagram illustrating the information extraction procedure in Embodiment 2. This is a graph showing the incidence angle dependence of sputtering yield. This is a schematic diagram showing an example of the display of a response. This is a flowchart illustrating the processing procedure executed by the developer terminal according to Embodiment 2. This is a schematic diagram showing the configuration of an information processing system according to Embodiment 3. This is a block diagram showing the internal configuration of a customer terminal. This is an explanatory diagram illustrating an example of a response output to a customer's query.

[0009] Hereinafter, one embodiment will be described with reference to the drawings. (Embodiment 1) Figure 1 is a schematic diagram showing the configuration of an information processing system according to the embodiment. The information processing system according to the embodiment is a system for receiving queries from developers who are engaged in process development for substrate processing, generating response text using a language model that refers to prodata and literature data related to substrate processing, and providing the generated response text to the developers.

[0010] The information processing system includes a developer terminal 100, a circuit board processing unit 120, a device data infrastructure 140, storage 160, and the like. These developer terminal 100, circuit board processing unit 120, device data infrastructure 140, and storage 160 are connected to each other via a local communication network NW1, such as a LAN (Local Area Network) or dedicated line, of the company, university, or research institution to which the developer belongs.

[0011] The developer terminal 100 is a terminal device such as a personal computer used by the developer. The developer uses the developer terminal 100 to perform process development by referring to various data obtained from the board processing device 120, various data stored in the device data base 140 and storage 160, and various data obtained from the external communication network NW2.

[0012] The substrate processing apparatus 120 is a device for performing substrate processing on a substrate (wafer) to be processed, such as etching, film deposition, CMP (Chemical Mechanical Polishing), ashing, and cleaning. Alternatively, the substrate processing apparatus 120 may be a device for manufacturing FPDs (Flat Panel Displays) such as liquid crystal display panels and organic EL (Electro-Luminescence) panels. In the example in Figure 1, only one substrate processing apparatus 120 is shown, but the substrate processing system may include multiple types of substrate processing apparatuses, or multiple substrate processing apparatuses of the same type.

[0013] The substrate processing apparatus 120 is equipped with devices such as a heater, gas supply source, flow rate controller, and high-frequency power supply, and stores information about these devices as device information. When substrate processing is performed in the substrate processing apparatus 120, recipe information that defines the setting values ​​for the above devices, processing procedures, etc., is input. While substrate processing is being performed in the substrate processing apparatus 120 according to the recipe information, the temperature of the substrate, the pressure and gas flow rate in the chamber, the voltage applied to the upper and lower electrodes, the plasma emission intensity, etc. are measured and output as measurement data. In addition, after the completion of substrate processing, various measurements may be performed on the substrate obtained as the output.

[0014] The device data infrastructure 140 is a group of systems for collecting, managing, and analyzing data such as device information of the substrate processing apparatus 120, recipe information used for substrate processing, measurement data obtained during the execution of substrate processing, and measurement data obtained about the output product after substrate processing.

[0015] The device information stored in the device data board 140 includes information for identifying the substrate processing device 120 (such as the device name and model number), and information on the equipment mounted on each substrate processing device 120 (such as the chamber, heater, gas supply source, flow controller, and high-frequency power supply). The equipment information includes information such as the type and material of the components that make up the equipment. The device information stored in the device data board 140 may further include information on various measuring devices or sensors provided by the substrate processing device 120.

[0016] The recipe information stored in the device data board 140 is information about procedures designed in advance by developers, etc., to achieve a specific substrate processing. Each recipe consists of multiple steps. Each step represents the smallest processing unit that changes the state (attributes of the object to be processed or the state of the substrate) in the substrate processing process.

[0017] The measurement data stored in the device data board 140 is the measurement result data obtained when actually processing a substrate using the substrate processing apparatus 120. Such measurement data is obtained by various measuring devices or sensors mounted on the substrate processing apparatus 120. Alternatively, the measurement data may be obtained from measuring devices or sensors located outside the substrate processing apparatus 120. Such measurement data may include, for example, data such as substrate temperature, high-frequency voltage, gas flow rate, emission intensity by emission spectroscopy, reflected wave power, plasma density, ion energy, and ion flow rate. The measurement data may also include initial data related to the object to be processed (substrate) (initial limit dimensions, material, thickness, aspect ratio, mask coverage, etc.). The measurement data may also include data set for the substrate processing apparatus 120 (pressure in the chamber, power of the high-frequency power supply, gas flow rate, temperature in the chamber, and surface temperature of the object to be processed, etc.). The measurement data may also include data related to the results obtained in each process (etching rate, film deposition rate, etc.). The measurement data may also include data on the substrate measured before and / or after processing (shape, film thickness, composition, etc.).

[0018] The device data infrastructure 140 collects and stores recipe information, measurement data during the substrate processing, and measurement data obtained for the output product each time substrate processing is performed by the substrate processing apparatus 120. If the collected measurement data is analyzed by the developer, the data of that analysis may also be stored in the device data infrastructure 140. Hereinafter, the data stored in the device data infrastructure 140 will be referred to as process data.

[0019] Storage 160 is a device for storing data such as specialized literature and reports related to substrate processing. Here, the data stored in storage 160 is data from documents created by process development experts, such as reports, internal documents, research diaries, and academic papers. The data stored in storage 160 may also include data such as specialized literature collected from external sources by experts. Here, experts may be the developers themselves, or engineers, analysts, researchers, etc., other than developers. Documents created by experts may include not only text, but also mathematical formulas, chemical reaction equations, figures, tables, graphs, flowcharts, scripts to be input into simulators, program source code, etc. The data stored in storage 160 is local data that can be accessed via the communication network NW1, and access from outside the communication network NW1 is restricted. In the following description, the document data stored in storage 160 is referred to as internal literature data.

[0020] The information processing system further includes an LLM server 200, storage 220, etc. These LLM server 200 and storage 220 are connected via a communication network NW2, such as the Internet.

[0021] The LLM server 200 is a server computer that provides services using a Large Language Model (LLM). The LLM server 200 is equipped with a language model MD that generates response sentences to input queries. The language model MD is an existing large language model such as GPT-4 (Generative Pretrained Transformer 4), LLaMA (Large Language Model Meta AI), or BERT (Bidirectional Encoder Representations from Transformers). Alternatively, the language model MD may be a proprietary language model that has been tuned from an existing large language model. The language model MD may be a unimodal language model that outputs a response sentence in text data in response to a question sentence in text data, or it may be a multimodal language model that can accept inputs other than text data, such as image data or audio data.

[0022] Storage 220 is a device that stores data such as specialized literature related to substrate processing. The data stored in storage 220 is the same as the literature data stored in storage 160, except that it is made publicly available via the communication network NW2. The data stored in storage 220 is data from documents created by process development experts, and includes data such as documents, specialized books, and academic papers made public on the communication network NW2 by companies, universities, research institutions, etc. In the following description, the data stored in storage 220 will be referred to as external literature data. Also, when it is not necessary to distinguish between internal and external literature data, both will simply be referred to as literature data.

[0023] The developer terminal 100 receives a developer query regarding board processing, and sends the received query to the LLM server 200. The LLM server 200 then generates a response using the language model MD, which references process data and literature data, and outputs the generated response.

[0024] Figure 2 is a block diagram showing the internal configuration of the developer terminal 100. The developer terminal 100 is a dedicated or general-purpose computer comprising, for example, a control unit 101, a storage unit 102, a communication unit 103, an operation unit 104, and a display unit 105.

[0025] The control unit 101 includes a CPU (Central Processing Unit), ROM (Read Only Memory), RAM (Random Access Memory), and the like. The ROM in the control unit 101 stores control programs that control the operation of each hardware component of the developer terminal 100. The CPU in the control unit 101 reads and executes the control programs stored in the ROM and the computer programs described later stored in the memory unit 102, and controls the operation of each hardware component, thereby making the entire device function as an information processing device according to the present disclosure. The RAM in the control unit 101 temporarily stores data used during the execution of calculations.

[0026] In this embodiment, the control unit 101 is configured to include a CPU, ROM, and RAM, but the configuration of the control unit 101 is not limited to the above. The control unit 101 may be one or more control circuits or arithmetic circuits equipped with, for example, a GPU (Graphics Processing Unit), FPGA (Field Programmable Gate Array), DSP (Digital Signal Processor), quantum processor, volatile or non-volatile memory, etc. Furthermore, the control unit 101 may include functions such as a clock that outputs date and time information, a timer that measures the elapsed time from the time a measurement start instruction is given to the time a measurement end instruction is given, and a counter that counts numbers.

[0027] The storage unit 102 includes storage devices such as an HDD (Hard Disk Drive), an SSD (Solid State Drive), and an EEPROM (Electronically Erasable Programmable Read Only Memory). The storage unit 102 stores various computer programs executed by the control unit 101 and various data used by the control unit 101.

[0028] The computer program (program product) stored in the memory unit 102 includes a data extraction program PG that causes the computer to execute a process to receive user queries regarding substrate processing, generate a response statement to the received user query using a language model that references process data and literature data related to substrate processing, and output the generated response statement. The data extraction program PG may be a single computer program or may consist of multiple computer programs. Furthermore, the data extraction program PG may be executed collaboratively by multiple computers. In addition, the data extraction program PG may partially utilize existing libraries.

[0029] The computer program, including the data extraction program PG, is provided on a non-temporary recording medium RM on which the computer program is recorded in a readable format. The recording medium RM is a portable memory such as a CD-ROM, USB memory, SD (Secure Digital) card, microSD card, or CompactFlash®. The control unit 101 reads various computer programs from the recording medium RM using a reading device (not shown in the figure) and stores the read computer programs in the storage unit 102. The computer programs stored in the storage unit 102 may also be provided via communication. In this case, the control unit 101 can acquire the computer programs via communication through the communication unit 103 and store the acquired computer programs in the storage unit 102.

[0030] The communication unit 103 is equipped with a communication interface for sending and receiving various types of data with external devices. A communication interface compliant with a communication standard such as LAN can be used as the communication interface of the communication unit 103. The external devices include a board processing device 120, a device data board 140, and a storage device 160 connected to the communication network NW1, as well as an LLM server 200 and a storage device 220 connected to the communication network NW2. When data to be transmitted is input from the control unit 101, the communication unit 103 transmits the data to the destination external device, and when data is received from the external device, it outputs the received data to the control unit 101.

[0031] The operation unit 104 is equipped with operating devices such as a touch panel, keyboard, and switches, and accepts various operations and settings from developers and others. The control unit 101 performs appropriate control based on the various operation information provided by the operation unit 104, and stores setting information in the storage unit 102 as needed.

[0032] The display unit 105 is equipped with a display device such as an LCD monitor or an OLED monitor, and displays information that should be notified to developers, etc., in response to instructions from the control unit 101.

[0033] The developer terminal 100 in this embodiment may be a single computer, or it may be a computer system composed of multiple computers and peripheral devices. Furthermore, the developer terminal 100 may be a virtualized virtual machine, or it may be a cloud.

[0034] The following describes an overview of the processes performed in the information processing system. Figures 3 to 6 are explanatory diagrams illustrating examples of response texts output to a developer's query. The developer terminal 100 receives the developer's query through a dedicated or general-purpose application program. The application program is, for example, the data extraction program PG mentioned above. Alternatively, a general-purpose web browser may be used. The developer terminal 100 can accept queries by text input through the operation unit 104. If the developer terminal 100 is equipped with a voice input unit, it may also accept queries by voice.

[0035] Figure 3 shows an example of a query received from a developer: "I want to create a plasma simulation model for the Yyy process of apparatus XXX, so please provide a list of relevant literature." In this example, "apparatus XXX" represents the name of the substrate processing apparatus 120, and "Yyy process" represents the process within the recipe. Such terms are used within the developer's organization (e.g., internal jargon) but may not be commonly used. Therefore, even if a general search engine is used with terms like "apparatus XXX" and "Yyy process" as search keywords, sufficient information may not be obtained. Similarly, even if a general search engine is used to search for information in a specific specialized field such as plasma simulation model generation, the developer may not find the information they are looking for.

[0036] Therefore, in the information processing system according to this embodiment, a language model MD that references process data and literature data is used to generate a response to a developer's query. That is, when a developer's query is received, the developer terminal 100 generates input data (prompts) for the language model MD based on the received query, and sends the generated prompts to the LLM server 200 to generate a response.

[0037] For example, the developer terminal 100 generates a prompt specifying that its role is that of a process development specialist, that it should refer to literature data and process data as reference information, and that it should answer the developer's query. The prompt may also define the output format of the response (in the example in Figure 3, the format of the literature list). The developer terminal 100 may generate the prompt through internal processing, or it may display the generated prompt on the display unit 105 and accept modifications and approvals from the developer. Regarding reference information, it may be specified to refer only to internal literature data (or external literature data), or to refer only to process data.

[0038] The developer terminal 100 sends the generated prompt to the LLM server 200 via the communication networks NW1 and NW2. The LLM server 200 inputs the prompt received from the developer terminal 100 into the language model MD. The language model MD outputs a response sentence according to the input prompt. Since the prompt specifies references to process data and literature data, the likelihood of obtaining a list of literature desired by the developer increases by searching for process data and literature data using search keys such as "device xxx" and "Yyy process". The LLM server 200 returns the response sentence data output from the language model MD to the developer terminal 100.

[0039] The developer terminal 100 receives the response data sent back from the LLM server 200 via the communication networks NW2 and NW1. Based on the received data, the developer terminal 100 displays the response on the display unit 105. In the example in Figure 3, the developer's query requests a list of relevant literature for creating a plasma simulation model, so the response includes a list of relevant literature. The list of relevant literature shown in Figure 3 is indicated by links to each document (document file). Each document may be a document file accessible via the internal communication network NW1, or a document file accessible via the external communication network NW2. The list of relevant literature is not limited to links to document files; it may also be a list of document names (a list of text information).

[0040] Figure 4 shows an example of receiving a developer query, "Please provide a list of reaction equations for related gases." This query represents a new query entered by the developer after the answer in Figure 3 was obtained (i.e., a query entered while the dialogue between the developer and the LLM server 200 is ongoing). Based on the received query, the developer terminal 100 generates input data (prompts) for the language model MD. As with the above, literature data and process data are specified as reference information in the prompt. The output format of the answer (the reaction equation format in the example of Figure 4) may also be defined in the prompt. The developer terminal 100 sends the generated prompts to the LLM server 200.

[0041] The LLM server 200 inputs the prompt received from the developer terminal 100 into the language model MD. While the session with the developer terminal 100 is maintained, the LLM server 200 is configured to maintain the context (the history of the interaction with the developer). Therefore, even if the above query does not explicitly state the words "device XXX", "Yyy process", and "plasma simulation", the language model MD can output a response sentence that includes a list of responses for "plasma simulation of the Yyy process of device XXX". Since the prompt specifies references to process data and literature data, the likelihood of obtaining the desired response list by searching the process data and literature data using "device XXX", "Yyy process", and "plasma simulation" as search keys increases. The LLM server 200 returns the data of the response sentence output from the language model MD to the developer terminal 100.

[0042] The developer terminal 100 receives the response data sent back from the LLM server 200 via the communication networks NW2 and NW1. Based on the received data, the developer terminal 100 displays the response on the display unit 105. An example of the response is shown in Figure 4 and includes a list of chemical reaction equations related to "plasma simulation of the Yyy process of apparatus XXX".

[0043] Figure 5 shows an example of receiving a developer's query "Please create a script for running calculations on Simulator Xxx using the above reaction list." This query represents a newly entered developer's query after the response text in Figure 4 was obtained (i.e., a query newly entered during the ongoing interaction between the developer and the LLM server 200). Based on the received query, the developer terminal 100 generates input data (prompt) for the language model MD. Similar to the above, literature data and process data are specified as reference information in the prompt. Also, in the prompt, the output format of the response text (in the example of Figure 5, a format suitable for input to the simulator) may be defined. The developer terminal 100 transmits the generated prompt to the LLM server 200.

[0044] The LLM server 200 inputs the prompt received from the developer terminal 100 into the language model MD. As described above, while the session with the developer terminal 100 is maintained, the LLM server 200 maintains the context. Therefore, when the above prompt is input into the language model MD, the language model MD can output a response text including a script for running calculations on Simulator Xxx using the previous reaction list. Since the reference to literature data is specified in the prompt, the process data and literature data are searched using the above reaction list as a search key, increasing the likelihood of obtaining the script desired by the developer. The LLM server 200 returns the data of the response text output from the language model MD to the developer terminal 100. Also, the LLM server 200 may extract only the script part from the response text output from the language model MD to create a script file, and store it in a storage unit (not shown) in the server as a file for download. Further, the LLM server 200 may transmit the created script file to the developer terminal 100.

[0045] The developer terminal 100 receives the response data sent back from the LLM server 200 via the communication networks NW2 and NW1. Based on the received data, the developer terminal 100 displays the response on the display unit 105. An example of the response is shown in Figure 5. The response includes, for example, a link to download a script created by the language model MD.

[0046] Figure 6 shows an example of receiving a developer query, "Please create an internal evaluation report related to the Yyy process of device XXX." This query represents a new query entered by the developer after the response in Figure 5 was obtained (i.e., a query entered while the dialogue between the developer and the LLM server 200 is ongoing). Based on the received query, the developer terminal 100 generates input data (prompts) for the language model MD. As above, literature data and process data are specified as reference information in the prompt. The output format of the response (the report format in the example of Figure 6) may also be defined in the prompt. The developer terminal 100 sends the generated prompts to the LLM server 200.

[0047] The LLM server 200 inputs the prompt received from the developer terminal 100 into the language model MD. As described above, the LLM server 200 maintains the context while the session with the developer terminal 100 is maintained, so when the above prompt is input into the language model MD, the language model MD can output a response containing information such as a process overview, relevant responses, and process considerations. Since the prompt specifies references to process data and literature data, the likelihood of obtaining the desired report by searching for process data and literature data using search keys such as "device xxx", "Yyy process", and "plasma simulation" increases. The LLM server 200 returns the data of the response output from the language model MD to the developer terminal 100.

[0048] The developer terminal 100 receives the data of the answer text returned from the LLM server 200 via the communication networks NW2 and NW1. Based on the received data, the developer terminal 100 causes the display unit 105 to display the answer text. An example of the answer text is as shown in FIG. 6. The answer text includes, for example, an overview of the process, highly relevant reactions, considerations of the process, and the content of the report to the company.

[0049] In FIGS. 3 to 6, the developer's query and the answer from the LLM server 200 server that occurred during a series of conversations between the developer and the LLM server 200 are shown, but it is not essential to utilize the conversation history. That is, the LLM server 200 may generate an answer text for the query from the developer without referring to the conversation history and transmit the data of the generated answer text to the developer terminal 100.

[0050] FIG. 7 is a flowchart for explaining the procedure of the process executed by the developer terminal 100 according to the first embodiment. When the developer makes a query regarding process development, the developer terminal 100 starts a dedicated application (for example, the data extraction program PG) or a web browser and accesses the LLM server 200. The developer terminal 100 receives the developer's query through the operation unit 104 (step S101). The control unit 101 of the developer terminal 100 generates input data (prompt) to be input to the language model MD based on the query received in step S101 (step S102).

[0051] At this time, the control unit 101 generates a prompt designating that the role is an expert in process development, referring to literature data and process data as reference information, and answering the developer's query. The prompt may be generated using template data. For example, the template data is stored in the storage unit 102 in advance. When the developer's query is received, the control unit 101 may generate the prompt by embedding the developer's query into the template data read from the storage unit 102.

[0052] The control unit 101 transmits the generated prompt to the LLM server 200 via the communication unit 103 (step S103). The transmitted prompt reaches the LLM server 200 via the communication networks NW1 and NW2.

[0053] Upon receiving a prompt, the LLM server 200 inputs the received prompt to the language model MD. The language model MD outputs a response sentence according to the input prompt. At this time, process data stored in the device data base 140, internal literature data stored in the storage 160, and external literature data stored in the storage 220 are referenced as needed. The LLM server 200 returns the response sentence data output from the language model MD to the developer terminal 100.

[0054] The control unit 101 of the developer terminal 100 receives the response data sent back from the LLM server 200 via the communication unit 103 (step S104). Based on the received response data, the control unit 101 displays the response on the display unit 105 (step S105). If the query received from the developer is, for example, a request to create a literature list, a response list, a script, or a report, a response text like those shown in Figures 3 to 6 will be obtained, and the corresponding response text will be displayed on the display unit 105 of the developer terminal 100. The same applies even if the query received from the developer is not a request to create the above; a response text corresponding to the developer's query will be displayed on the display unit 105 of the developer terminal 100.

[0055] As described above, when the developer terminal 100 according to Embodiment 1 receives a query from a developer regarding substrate processing, it can generate a response to the query by using the language model MD, which refers to process data and literature data related to substrate processing, and provide it to the developer.

[0056] (Embodiment 2) Embodiment 2 describes a configuration that generates a response statement containing physical property information related to substrate processing based on a developer's query.

[0057] Traditionally, when developers seek specific physical property information related to substrate processing, they have manually searched academic papers, documents on communication networks, and internal company reports, and then formatted the data according to their purpose. However, in specialized fields such as substrate processing, the accuracy of representation prediction is expected to be low. Furthermore, while techniques for individually extracting data from images and tables within text have been reported, it is difficult to associate specific technical terms with graphs or tables. For this reason, even if developers search for specific physical property information related to substrate processing on general search engines, there is a high probability that they will not obtain the information they desire.

[0058] Therefore, in Embodiment 2, a configuration is described in which a developer's query for specific physical property information related to substrate processing is received, a response to the received query is generated using a language model MD that references process data and literature data, and the generated response is provided to the developer.

[0059] In the embodiment, the physical property information includes not only information indicating physical properties such as thermal properties, electrical properties, magnetic properties, optical properties, and mechanical properties, but also information indicating chemical properties such as chemical reactions. For example, the physical property information may include characteristic values ​​in gas-phase reactions (excitation, ionization, elastic scattering, etc.), characteristic values ​​in surface reactions (etching, deposition, sputtering, etc.), and compound-specific information such as chemical formula, chemical structure, polarizability, and dipole moment in chemical reactions.

[0060] Figures 8 and 9 are explanatory diagrams illustrating the information extraction procedure in Embodiment 2. Figure 8 shows the procedure for generating a search query for process data, and Figure 9 shows the procedure for generating a search query for literature data. The developer terminal 100 receives queries from the developer to inquire about physical property information related to substrate processing, for example, through the operation unit 104. In the example in Figure 8, a query such as "Please tell me the generation rate of desorbed material that is scraped off when ions generated in apparatus A collide with the wall or wafer" is shown. In this example, "generation rate of desorbed material" is the physical property information requested by the developer.

[0061] The developer terminal 100 generates a search query for searching process data based on the query received from the developer, using the language model MD. For example, the developer terminal 100 generates a prompt requesting the generation of a search query to search process data in relation to the developer's query (consultation content). In the prompt, the developer terminal 100 may specify that the role is that of a process development expert and that process data and literature data should be referenced as reference information. The developer terminal 100 generates the search query by sending the generated prompt to the LLM server 200.

[0062] In the example in Figure 8, in order to obtain physical property information such as the rate of desorbed material formation, it is necessary to understand, for example, the type of chemical reaction occurring in apparatus A, the type of ions generated in apparatus A, and the materials of the chamber wall and wafer. For this reason, when a prompt is sent to the LLM server 200, it is expected that the search query returned by the LLM server 200 will include, for example, search queries to search for the type of chemical reaction occurring in apparatus A, the type of ions generated in apparatus A, and the materials of the chamber wall and wafer. The example in Figure 8 shows that such search queries were generated: search query 1 (chemical reaction occurring in apparatus A), search query 2 (generated ions), search query 3 (material of the chamber wall), and search query 4 (material of the wafer).

[0063] The developer terminal 100 uses a search query generated via the LLM server 200 to perform a search on the process data stored in the device data infrastructure 140. The developer terminal 100 can search for process data using the search function of the OS (Operation System) or an application program such as a web browser.

[0064] Furthermore, the search queries generated through the LLM server 200 are not limited to one, but may be multiple. In the latter case, the developer terminal 100 may execute the search process using each of the multiple search queries, or it may execute the search process by combining multiple search queries. Also, the search query may be a single word (search keyword), a natural sentence containing one or more search keywords, or a search operator such as wildcards, AND conditions, or OR conditions. In addition, the developer terminal 100 may display the search queries generated through the LLM server 200 on the display unit 105, accept additions and modifications by the developer, etc., and then execute the search process.

[0065] When a search is performed using search queries 1 to 4, for example, the search results for search query 1 may be "ion sputtering," for search query 2 "argon ions," for search query 3 "aluminum alloy," and for search query 4 "silicon." A single search query may yield more than one search result; multiple results may be obtained for a single query. Furthermore, the search results may not be limited to single words, but may include sentences, numerical data, charts, graphs, etc.

[0066] The developer terminal 100 generates a search query for searching literature data based on a query received from the developer, using the language model MD. For example, the developer terminal 100 generates a prompt requesting the generation of a search query to search literature data by referring to reference information regarding the developer's query (consultation content). The reference information can specify search results for process data (e.g., ion sputtering in the above example). The developer terminal 100 may also specify in the prompt that its role is that of a process development expert. The developer terminal 100 generates the search query by sending the generated prompt to the LLM server 200.

[0067] In the example in Figure 9, in order to obtain physical property information such as the rate of desorbed material formation from literature data, it is necessary to understand, for example, the sputtering yield when ions generated in apparatus A are incident on the target object (chamber wall or wafer). For this reason, it is expected that the search query returned by the LLM server 200 when a prompt is sent to the LLM server 200 will include a search query to search for, for example, the sputtering yield when ions generated in apparatus A are incident on the target object (chamber wall or wafer) from literature data. The example in Figure 9 shows that such search queries were generated: search query 1 (sputtering yield), search query 2 (incident angle dependence or incident energy dependence), search query 3 (argon ions), and search query 4 (silicon or aluminum alloy).

[0068] The developer terminal 100 performs a search process on internal or external literature data using a search query generated through the LLM server 200. The developer terminal 100 can search the literature data using the OS search function or an application program such as a web browser. Note that the search query generated through the LLM server 200 is not limited to one, but may be multiple. In the latter case, the developer terminal 100 may perform the search process using each of the multiple search queries, or it may perform the search process by combining multiple search queries. Furthermore, the search query may be a single word (search keyword), a natural sentence containing one or more search keywords, or a search operator such as wildcards, AND conditions, or OR conditions. In addition, the developer terminal 100 may display the search query generated through the LLM server 200 on the display unit 105 and perform the search process after accepting additions and modifications by the developer.

[0069] When searching literature data, the search results may include one or more documents containing the physical property information requested by the developer. The physical property information is included as text information in the documents. In addition, the physical property information may also be included as graphs or charts. For example, when performing a search using search queries 1 to 4 shown in Figure 9, a document containing a graph (Figure 10) showing the incidence angle dependence of the sputtering yield of silicon when argon ions are incident may be found. The search results are not limited to documents containing the above graph, but may also include documents containing numerical tables showing the above relationship, graphs or numerical tables showing the incidence angle dependence or energy dependence of the sputtering yield on the chamber wall (e.g., aluminum alloy), etc.

[0070] If the physical property information is presented as a graph or chart (table) in the literature obtained as a search result, the developer terminal 100 may extract the physical property information (characteristic values ​​related to physical properties) from the graph or chart. The characteristic value extracted from the graph or chart does not need to be a single value; it may consist of multiple values. Existing methods utilizing OCR (Optical Character Recognition) or the like can be used to extract characteristic values ​​from the graph or chart.

[0071] The developer terminal 100 generates a response text containing physical property information extracted from the literature and displays the generated response text on the display unit 105. Figure 11 is a schematic diagram showing an example of how the response text is displayed. Figure 11 shows an example in which a response text is generated that includes the names (links) of the two searched documents and the values ​​read from the graphs contained in these documents as physical property information, and the generated response text is displayed on the display unit 105.

[0072] Figure 11 shows an example of generating and displaying a response text that includes the literature title and physical property information. However, the response text may also include equipment data obtained from process data, recipe data, etc., in addition to physical property information. Furthermore, when receiving a developer's query, the developer may be identified, and the response text may be modified according to the developer's skill level. For example, if the developer's skill level is low, annotations for technical terms may be added. On the other hand, if the developer's skill level is high, instead of including numerical information in the response text, the page number in the literature where the graph is located may be included in the response text.

[0073] Furthermore, in the example shown in Figure 11, the system displays numerical values ​​read from the graph as physical property information. However, it is also possible to create a file to store the numerical values ​​read from the graph (for example, a CSV (Comma-Separated Values) file) and embed a link to the created file in the answer.

[0074] Figure 12 is a flowchart illustrating the procedure of processing performed by the developer terminal 100 according to this second embodiment. When a developer asks a question to inquire about specific physical property information related to substrate processing, the developer terminal 100 launches a dedicated application (e.g., data extraction program PG) or a web browser and accesses the LLM server 200. The developer terminal 100 receives the developer's query to inquire about specific physical property information related to substrate processing through the operation unit 104 (step S201). The developer's query includes information such as the equipment, process, gas, material information of constituent members, and physical property information for analysis purposes.

[0075] The control unit 101 of the developer terminal 100 generates a search query for searching process data based on the query received in step S201 (step S202). Specifically, the control unit 101 generates a prompt requesting the generation of a search query to search process data regarding the developer's query (consultation content), and generates the search query by sending the generated prompt to the LLM server 200.

[0076] The control unit 101 executes a search process for process data using a search query generated through the LLM server 200 (step S203). The control unit 101 searches for process data stored in the device data base 140 using the OS search function or an application program such as a web browser. The process data obtained as a search result is stored in the storage unit 102. If the relevant process data does not exist in the device data base 140, the control unit 101 displays a message to that effect on the display unit 105.

[0077] Next, the control unit 101 generates a search query for searching for literature data based on the query received from the developer (step S204). Specifically, it generates a prompt requesting the generation of a search query to search for literature data by referring to reference information regarding the developer's query (consultation content). The reference information specifies the search results of the process data from step S203.

[0078] The control unit 101 executes a search process for the literature data using the search query generated through the LLM server 200 (step S205). The control unit 101 searches for the literature data stored in the storage 160 and 220 using the OS search function or an application program such as a web browser. The literature data obtained as a search result is stored in the storage unit 102. If the relevant literature data is not found in the storage 160 and 220, the control unit 101 displays a message to that effect on the display unit 105.

[0079] In step S205, if the relevant literature data is found, the control unit 101 extracts the desired physical property information from the literature data (step S206). If the desired physical property information is included in the literature data as text information, the control unit 101 only needs to extract the corresponding text information. If the desired physical property information is included in the literature data as a graph or chart, the control unit 101 can use existing methods such as OCR to extract characteristic values ​​(physical property values) related to the physical properties from the graph or chart.

[0080] The control unit 101 generates a response sentence containing the physical property information extracted in step S206 (step S207), and displays the generated response sentence on the display unit 105 (step S208).

[0081] As described above, in Embodiment 2, since the response is generated that includes the physical property information desired by the developer, it is possible to provide a response that is highly useful to the respondent.

[0082] (Embodiment 3) In Embodiments 1 and 2, the LLM server 200 is configured to generate response texts in response to queries from developers. However, the LLM server 200 may also generate response texts in response to queries from general users (customers) of the board processing device 120, or in response to queries from support personnel at customer sites. Embodiment 3 describes a configuration in which response texts are generated based on queries from customers or support personnel.

[0083] Figure 13 is a schematic diagram showing the configuration of the information processing system according to Embodiment 3. The information processing system according to Embodiment 3 includes a customer terminal 110A and a support terminal 110B. The customer terminal 110A and the support terminal 110B are connected to the LLM server 200 via the communication network NW2. The customer terminal 110A is a terminal used by general users (customers) of the substrate processing apparatus 120. In Embodiment 3, a general user refers to an end user who uses the substrate processing apparatus 120 to manufacture a desired substrate. The support terminal 110B is a terminal used by support personnel at the customer site. The customer site is a site established by the developer / manufacturer of the substrate processing apparatus 120, etc., to provide technical support to customers. Support personnel are operators who provide technical support to customers in response to various inquiries.

[0084] The LLM server 200, based on queries input from the customer terminal 110A and the support terminal 110B, refers to the data stored in the storage 220 and generates and outputs a response to the query. Furthermore, if connection to the communication network NW1 is permitted, the LLM server 200 may also refer to the data stored in the device data infrastructure 140 and / or the data stored in the storage 160 to generate and output the response to the query.

[0085] Figure 14 is a block diagram showing the internal configuration of the customer terminal 110A. The customer terminal 110A is a dedicated or general-purpose computer comprising, for example, a control unit 111, a storage unit 112, a communication unit 113, an operation unit 114, and a display unit 115. The functions of each hardware component of the customer terminal 110A are the same as those of the developer terminal 100. The control unit 111 reads and executes a computer program, such as a data extraction program PG, stored in the storage unit 112, thereby enabling the entire device to function as an information processing device according to the present disclosure.

[0086] Figure 14 shows the internal configuration of the customer terminal 110A. However, since the internal configuration of the support terminal 110B is the same as that of the customer terminal 110A, its explanation will be omitted.

[0087] Figure 15 is an explanatory diagram illustrating an example of a response text output to a customer query. In order to obtain a response using the information processing system, the customer enters authentication information such as a user ID and password and requests to log in to the information processing system via the customer terminal 110A. If the login to the information processing system is successful, the customer terminal 100A accepts the customer's query through the data extraction program PG. The customer terminal 110A can accept queries by text input through the operation unit 114. If the customer terminal 110A is equipped with a voice input unit, it may also accept queries by voice.

[0088] Figure 15 shows an example of a customer query received: "An alarm Zzz has occurred in the Yyy process of device XXX. Please tell me how to deal with it." In this example, "device XXX" is the name of the substrate processing device 120, "Yyy process" is the process in the recipe, and "Zzz" is the location where the alarm is set. Such terminology is used by the developers and manufacturers of the substrate processing device 120, but it may not be commonly used by the general public. Therefore, even if you use terms such as "device XXX," "Yyy process," and "Zzz" as search keywords to search general search sites, you may not be able to obtain sufficient information.

[0089] Therefore, in the information processing system according to this embodiment, a language model MD that references process data and literature data is used to generate a response to a customer query. That is, when a customer query is received, the customer terminal 110A generates input data (prompt) for the language model MD based on the received query, and sends the generated prompt to the LLM server 200 to generate a response.

[0090] The customer terminal 110A sends the generated prompt to the LLM server 200 via the communication network NW2. The LLM server 200 inputs the prompt received from the customer terminal 110A into the language model MD. The language model MD outputs a response sentence according to the input prompt. Since the prompt specifies references to process data (process logs when an alarm occurs, status information, etc.) and literature data (trouble reports, knowledge put into language), the likelihood of obtaining the desired solution is increased by searching for process data and literature data using search keys such as "device xxx", "Yyy process", and "Zzz". The LLM server 200 returns the response sentence data output from the language model MD to the customer terminal 100.

[0091] The customer terminal 110A receives the response data sent back from the LLM server 200 via the communication network NW2. Based on the received data, the customer terminal 110A displays the response on the display unit 105. In the example in Figure 15, the customer is inquiring about how to deal with an alarm that occurred in a specific situation, so the response includes text indicating how to deal with the alarm.

[0092] Figure 15 shows an example of a response to a customer query, but the process is similar when a support staff member receives a query. Support staff members receive various inquiries (consultations) from customers via telephone, email, etc. Based on the content of the consultation from the customer, the support staff member inputs the query into the support terminal 110B. The support terminal 110B can receive queries via text input through the operation unit 114. If the support terminal 110B is equipped with a voice input unit, it may also accept queries via voice. For example, the support terminal 110B may receive a query such as "An alarm Zzz has occurred in the Yyy process of device XXX, please tell me how to deal with it," as shown in Figure 15.

[0093] The support terminal 110B sends the received query to the LLM server 200 and displays the data of the response text generated by the LLM server 200. The support terminal 110B responds to the customer's inquiry by displaying the response text obtained from the LLM server 200 on the display unit 115 of the customer terminal 110A.

[0094] As described above, in Embodiment 3, when a query is received from a customer or support staff other than the developer, the language model MD can be used to generate a response to the query and provide it to the customer or support staff.

[0095] The matters described in each embodiment can be combined with each other. Furthermore, the independent claims and dependent claims described in the claims can be combined with each other in any combination, regardless of the form of reference. In addition, the claims may be described in a form in which claims refer to two or more other claims (multi-claim form), or in a form in which multi-claims refer to at least one multi-claim (multi-multi-claim).

[0096] The embodiments disclosed herein should be considered in all respects to be illustrative and not restrictive. The scope of the invention is indicated by the claims, not in the sense described above, and all modifications within the sense and scope equivalent to the claims are intended.

[0097] For example, in Embodiment 1, a developer terminal 100 connected to the communication network NW1 receives a question from a developer, and the developer terminal 100 generates an answer using the LLM server 200. However, a server that generates an answer using the LLM server 200 may be provided separately within the communication network NW1. In this case, the developer's question received by the developer terminal 100 is sent to the server within the communication network NW1, the server generates an answer using the LLM server 200, and the generated answer is sent back to the developer terminal 100.

[0098] 100 Developer terminal 101 Control unit 102 Storage unit 103 Communication unit 104 Operation unit 105 Display unit PG Data extraction program RM Recording medium 200 LLM server

Claims

1. An information processing method that uses a computer to receive user queries regarding substrate processing, generate a response to the received user query using a language model that references process data and literature data related to substrate processing, and output the generated response.

2. The information processing method according to claim 1, wherein the computer generates a search query for searching the process data and the literature data based on the user's query, using the language model, and generates the answer text based on the search results from the process data and the literature data.

3. The information processing method according to claim 2, wherein the computer performs the process of searching for the process data using the search query and searching for the literature data using the search results of the process data.

4. The information processing method according to claim 1, wherein the process data includes equipment information of a substrate processing apparatus used for the substrate processing, recipe data used for the substrate processing, and measurement data measured during the execution of the substrate processing.

5. The information processing method according to claim 1, wherein the literature data includes data from specialized books on substrate processing and data from reports of users who have performed the substrate processing.

6. The information processing method according to any one of claims 1 to 5, wherein the computer performs a process of extracting physical property information related to the substrate processing from the aforementioned literature data and generating a response sentence containing the extracted physical property information.

7. The information processing method according to claim 6, wherein the physical property information includes characteristic values ​​related to physical properties extracted from a reaction equation included in the literature data, or from a graph or chart included in the literature data.

8. The information processing method according to claim 1, wherein the computer performs a process of generating a report relating to the substrate processing as the answer text.

9. The information processing method according to claim 1, wherein the computer performs a process of generating a script to be input into a simulator for substrate processing as the answer text.

10. The information processing method according to claim 1, wherein the substrate processing includes processing performed in a semiconductor manufacturing apparatus.

11. An information processing device comprising one or more processors, wherein the processors receive user queries regarding substrate processing, generate a response statement to the received user query using a language model that references process data and literature data regarding substrate processing, and output the generated response statement.

12. A computer program that receives user queries regarding substrate processing, generates response texts to the received user queries using a language model that references process data and literature data related to substrate processing, and outputs the generated response texts.

Images