Computer program, information processing method, learning data generation method, learning model generation method, and information processing device

Abstract

Provided are a computer program, an information processing method, a learning data generation method, a learning model generation method, and an information processing device that can be expected to support measures against particles generated in a processing container of a substrate processing apparatus.　A computer program according to the present embodiment causes a computer to execute processes for: acquiring particle data including position information of particles generated in a processing container of a substrate processing apparatus; inputting the acquired particle data to a first learning model trained by machine learning to receive the particle data as input and output information about a particle recovery method; acquiring information about the recovery method output by the first learning model; and outputting the acquired information about the recovery method.

Description

[Supplementary based on Rule 26, October 1, 2025] Computer Program, Information Processing Method, Method for Generating Learning Data, Method for Generating Learning Model, and Information Processing Apparatus

[0001] The present disclosure relates to a computer program, an information processing method, a method for generating learning data, a method for generating a learning model, and an information processing apparatus.

[0002] In Patent Document 1, a computational defect analysis system for classifying defects on a substrate including electronic components has been proposed. In this system, a first-stage classification engine processes measurement data to generate an initial classification of the defect. Then, a second-stage classification engine outputs the probability that the defect is caused by one or more potential causes, using manufacturing information or previous defect knowledge in addition to the initial classification.

[0003] U.S. Patent Application Publication No. 2020 / 0226742

[0004] The present disclosure provides a computer program, an information processing method, a method for generating learning data, a method for generating a learning model, and an information processing apparatus that can be expected to assist in measures against particles generated in a processing chamber of a substrate processing apparatus.

[0005] A computer program according to an embodiment acquires particle data including position information of particles generated in a processing chamber of a substrate processing apparatus, inputs the particle data to a first learning model that has been machine-learned to output information regarding a particle recovery method when receiving the particle data as an input, acquires the information regarding the recovery method output by the first learning model, and causes a computer to execute a process of outputting the acquired information regarding the recovery method.

[0006] According to the present disclosure, it can be expected to assist in measures against particles generated in a processing chamber of a substrate processing apparatus.

[0007] This is a schematic diagram illustrating the outline of the information processing system according to this embodiment. This is a block diagram showing an example configuration of the information processing device according to this embodiment. This is a flowchart showing an example of the procedure for data collection processing performed by the information processing device according to this embodiment. This is a schematic diagram showing an example of displaying clustering results. This is a schematic diagram showing an example of displaying experimental conditions. This is a schematic diagram showing an example configuration of a dust source prediction model. This is a schematic diagram showing an example configuration of a recovery method prediction model. This is a schematic diagram illustrating an example of data augmentation. This is a schematic diagram illustrating an example of a method for generating particle map data using a generative model. This is a schematic diagram illustrating another example of a method for generating particle map data using a generative model. This is a flowchart showing an example of the procedure for learning model generation processing performed by the information processing device according to this embodiment. This is a flowchart showing an example of the procedure for prediction processing performed by the information processing device according to this embodiment. This is a schematic diagram showing an example configuration of a recovery method prediction model according to a modification.

[0008] Specific examples of information processing systems according to the embodiments of this disclosure will be described below with reference to the drawings. However, this disclosure is not limited to these examples and is intended to include all changes within the meaning and scope of the claims as indicated by the claims.

[0009] <System Configuration> Figure 1 is a schematic diagram illustrating the outline of the information processing system according to this embodiment. The information processing system according to this embodiment is configured to include an information processing device 1 and a substrate processing device 3. The substrate processing device 3 is a device that performs various substrate processing on semiconductor substrates (wafers), such as CVD (Chemical Vapor Deposition), sputtering, etching, CMP (Chemical Mechanical Polishing), ashing, or cleaning, and has a chamber (processing container) for housing the substrate to be processed. Alternatively, the substrate processing device 3 may be a device for manufacturing FPDs (Flat Panel Displays) such as liquid crystal display panels or organic EL (Electro-Luminescence) panels. In addition to semiconductor substrates (wafers), the substrate may also be a glass substrate or a substrate for flat panel displays, etc.

[0010] In the substrate processing apparatus 3, particles generated inside the processing container may adhere to the surface of the substrate being processed, potentially causing a decrease in yield. Particles are minute particles, some brought in from outside the substrate processing apparatus 3, and others generated inside the substrate processing apparatus 3. Particles consist of various inorganic or organic materials, such as metals, rubber, plastics, dust, or dirt. In this embodiment, particles may include not only solids but also liquids. If a decrease in yield or other issues occur due to particles, the user is required to identify the source of the particles (dust source) and remove it. The information processing system according to this embodiment is a system that supports the user by identifying the dust source and proposing recovery methods for particles generated inside the processing container of the substrate processing apparatus 3. In order to identify the dust source of the particles and propose recovery methods as described above, the information processing system according to this embodiment uses a learning model, so-called AI (Artificial Intelligence), that has been generated in advance by machine learning.

[0011] The processing performed by the information processing system according to this embodiment is divided into a preparation stage, which involves data collection and the generation of a learning model, and an implementation stage, which involves identifying dust sources and proposing recovery methods using the generated learning model. Figure 1 shows the preparation stage in the upper section and the implementation stage in the lower section. Furthermore, the information processing device 1 and the substrate processing device 3 of the information processing system according to this embodiment are distinguished as follows: the devices that perform processing in the preparation stage are designated as information processing device 1A and substrate processing device 3A, and the devices that perform processing in the implementation stage are designated as information processing device 1B and substrate processing device 3B, as necessary. However, information processing device 1A and information processing device 1B may be the same device or different devices. Similarly, substrate processing device 3A and substrate processing device 3B may be the same device or different devices.

[0012] In the preparation phase, the information processing system according to this embodiment conducts various experiments in the substrate processing apparatus 3A, such as how particles generated at a dust source diffuse within the processing container, and measures are taken to determine the position and size of the diffused particles. In the experiments, for example, multiple particles that will become particles are placed in the processing container of the substrate processing apparatus 3A, the substrate processing apparatus 3A performs substrate processing under predetermined conditions (recipe), and after the substrate processing is completed, the position and size of the particles attached to the surface of the substrate are measured under various conditions. As a result of these experiments, the information processing apparatus 1A acquires particle data containing information such as the position and size measured for each particle from the substrate processing apparatus 3A, and stores the acquired particle data in the process DB (database) 4. The particle data may also include information such as the conditions under which the substrate processing apparatus 3A performed the substrate processing.

[0013] Furthermore, the information processing device 1A according to this embodiment performs a process to propose experimental conditions in order to efficiently collect more particle data. Based on the particle data already obtained, the information processing device 1A generates a model (particle prediction model) that predicts the characteristics of particles (such as the position and distribution of particles) that can be obtained for substrate processing conditions. The information processing device 1A determines the substrate processing conditions to be experimented with using the generated particle prediction model. The information processing device 1A can determine the experimental conditions using methods such as Bayesian optimization and experimental design.

[0014] The information processing device 1A also performs the process of labeling the particle data acquired as experimental results, so-called annotation. In this embodiment, the information processing device 1A labels each of the multiple particles in the particle data that contain information, indicating which is the dust source. The information processing device 1A also labels each of the multiple particles for each dust source that indicates which is the appropriate recovery method. User intervention is required for these labeling processes, and the information processing device 1A provides support to reduce the user's operational burden for annotation. In this embodiment, the information processing device 1A classifies the multiple particles contained in the particle data into multiple clusters using a so-called unsupervised learning clustering method, and displays the classification results to the user. The information processing device 1A accepts corrections to the clustering results from the user and also accepts input of labels for each cluster from the user. The information processing device 1A stores information indicating which label corresponds to each particle contained in the particle data in the process DB 4.

[0015] After storing a sufficient amount of particle data in the process DB4, the information processing device 1A performs a process to generate two learning models using machine learning with this particle data: a dust source prediction model (second learning model) 5 that identifies (classifies) dust sources, and a recovery method prediction model 6 (first learning model) that determines (classifies) recovery methods. For example, the information processing device 1A generates a dust source prediction model 5 that accepts particle data as input and classifies which particle is the dust source by performing supervised machine learning using training data (training data) that associates particle data with labels indicating the dust source of each particle. The information processing device 1A also generates a recovery method prediction model 6 that accepts particle data for each dust source as input and classifies which recovery method is appropriate for the particle by performing supervised machine learning using training data that associates particle data divided by dust source with labels indicating the recovery method. Information regarding the dust source prediction model 5 and the recovery method prediction model 6 generated by the information processing device 1A is stored in the model DB 7 of the information processing device 1B that performs substrate processing.

[0016] In the implementation phase, the information processing system according to this embodiment causes the substrate processing apparatus 3B to perform various substrate processing operations based on substrate processing conditions set in advance by the user. The substrate processing apparatus 3B measures particles attached to the substrate to be processed at appropriate timings, such as during or after the substrate processing. The substrate processing apparatus 3B may also measure the results of the substrate processing, such as the amount removed by etching, using sensors or the like. The substrate processing apparatus 3B outputs the data obtained from these measurements as the substrate processing results to the information processing apparatus 1B.

[0017] The information processing device 1B, having acquired the substrate processing results from the substrate processing device 3B, generates map data for each dust source for particles generated within the chamber of the substrate processing device 3B based on the particle data included in the substrate processing results, and provides the user with a recovery method for each generated dust source. The information processing device 1B uses a dust source prediction model 5, which is pre-stored in the model DB 7, to predict the dust source of each particle included in the particle data acquired from the substrate processing device 3B, and generates map data for each dust source by dividing the particle data according to the dust source. The information processing device 1B also uses a recovery method prediction model 6, which is pre-stored in the model DB 7, to predict a recovery method according to the map data for each dust source.

[0018] The exchange of information between the substrate processing device 3 and the information processing device 1 may be carried out, for example, via wired or wireless communication, or via a recording medium such as a memory card or optical disc.

[0019] <Device Configuration> Figure 2 is a block diagram showing an example configuration of the information processing device 1 according to this embodiment. The information processing device 1 according to this embodiment can be realized, for example, by installing a predetermined application program on a general-purpose information processing device such as a personal computer or a server computer. The information processing device 1 according to this embodiment is configured to include a processing unit 11, a storage unit 12, a communication unit 13, a display unit 14, and an operation unit 15, etc. In Figure 2, the information processing device 1 is shown as having the functions of both the information processing devices 1A and 1B shown in Figure 1, but it is not limited to this, and an information processing system may be configured by separately providing a device that has the functions of only the information processing device 1A and a device that has the functions of only the information processing device 1B. Furthermore, the processing performed by the information processing device 1 may be distributed and performed by three or more devices.

[0020] The processing unit 11 is composed of a CPU (Central Processing Unit), MPU (Micro-Processing Unit), GPU (Graphics Processing Unit), or quantum processor, as well as ROM (Read Only Memory) and RAM (Random Access Memory). The processing unit 11 reads and executes a program 12a stored in the memory unit 12, thereby performing various processes such as collecting necessary data from the substrate processing device 3, generating a dust source prediction model 5 and a recovery method prediction model 6 using machine learning based on the data obtained from the substrate processing device 3, and generating map data for each dust source and proposing recovery methods based on the substrate processing results of the substrate processing device 3.

[0021] The storage unit 12 is configured using a large-capacity storage device such as a hard disk or an SSD (Solid State Drive). The storage unit 12 stores various programs executed by the processing unit 11, and various data necessary for the processing of the processing unit 11. In this embodiment, the storage unit 12 stores the program 12a executed by the processing unit 11. The storage unit 12 is also provided with a process DB 4 for storing various data acquired from the substrate processing device 3, and a model DB 7 for storing information related to the learning model generated by the information processing device 1.

[0022] In this embodiment, the program (computer program, program product) 12a is provided in a form recorded on a recording medium 99 such as a memory card or optical disc, and the information processing device 1 reads the program 12a from the recording medium 99 and stores it in the storage unit 12. However, the program 12a may also be written to the storage unit 12 during the manufacturing stage of the information processing device 1, for example. Alternatively, the program 12a may be distributed by a remote server device or the like and acquired by the information processing device 1 via communication. For example, the program 12a may be read from the recording medium 99 by a writing device and written to the storage unit 12 of the information processing device 1. The program 12a may be provided by distribution via a network, or it may be provided in a form recorded on the recording medium 99.

[0023] Process DB4 stores particle data measured by the substrate processing apparatus 3, along with various information such as the date and time the data was acquired, identification information of the substrate processing apparatus 3, identification information of the substrate that underwent processing, the conditions for substrate processing, the results of substrate processing, and log information output by the substrate processing apparatus 3. In this embodiment, particle data is a collection of information about multiple particles that are generated in the chamber of the substrate processing apparatus 3 and adhere to the substrate to be processed. For each particle, the x and y coordinate values ​​indicating the attachment position on the substrate are associated with the particle's size, composition, or other related information, and this multidimensional information is collected for each particle that has adhered to the substrate, forming a point cloud data which is treated as particle data in this embodiment. In the following description, as an example, particle data will be described as a collection of three-dimensional information that associates the x, y, and size values ​​of each particle.

[0024] In this embodiment, "particle data" includes at least this point cloud data, and may or may not include further information such as the substrate processing conditions and the results of the substrate processing. Also, in this embodiment, when distinguishing only the point cloud data included in the particle data, this point cloud data alone is called "particle map data" or "map data". If the particle data does not include other information such as the substrate processing conditions and the results of the substrate processing, the particle data and particle map data are substantially the same. The map data for each dust source generated by the information processing device 1 is obtained by extracting point cloud data relating to particles from the same dust source.

[0025] Model DB7 stores information about the machine learning model. This information may include, for example, information indicating the configuration of the model and information such as the values ​​of internal parameters determined by machine learning. In this embodiment, Model DB7 stores information about a dust source prediction model 5 that predicts the dust source for multiple particles included in particle data (particle map data), and a recovery method prediction model 6 that predicts the recovery method for particles based on the map data for each dust source. In this embodiment, the dust source prediction model 5 accepts the particle map included in the particle data as input and outputs information classifying which dust source each particle is. The recovery method prediction model 6 accepts the map data for each dust source and information such as the substrate processing conditions and substrate processing results included in the process data as input, and outputs information classifying which of a predetermined number of recovery methods is appropriate.

[0026] The communication unit 13 is connected to the substrate processing device 3, for example, via a communication line, and exchanges data with the substrate processing device 3. In this embodiment, the communication unit 13 receives particle data transmitted from the substrate processing device 3 and provides it to the processing unit 11. The communication unit 13 also transmits data such as control commands provided by the processing unit 11 to the substrate processing device 3.

[0027] The display unit 14 is configured using a liquid crystal display or the like, and displays various images and characters based on the processing of the processing unit 11. The display unit 14 displays various information such as particle data acquired from the substrate processing device 3, information about the generated learning model, map data for each dust source, and recovery methods for particles.

[0028] The operation unit 15 receives user input and notifies the processing unit 11 of the received input. For example, the operation unit 15 receives user input via a mechanical button or an input device such as a touch panel provided on the surface of the display unit 14. Alternatively, the operation unit 15 may be an input device such as a mouse and a keyboard, and these input devices may be configured to be detachable from the information processing device 1.

[0029] The memory unit 12 may be an external storage device connected to the information processing device 1. The information processing device 1 may be a multicomputer comprising multiple computers, or it may be a virtual machine virtually constructed by software. Furthermore, the information processing device 1 is not limited to the above configuration, and may not include, for example, a display unit 14 and an operation unit 15.

[0030] Furthermore, in the information processing device 1 according to this embodiment, the processing unit 11 reads and executes the program 12a stored in the storage unit 12, thereby realizing the data acquisition unit 11a, model generation unit 11b, dust source separation unit 11c, recovery method prediction unit 11d, and display processing unit 11e, etc., as software-based functional units in the processing unit 11. In this figure, the functional units of the processing unit 11 that perform processing related to particles generated in the chamber of the substrate processing device 3 are shown, and functional units related to other processing are omitted from the illustration.

[0031] The data acquisition unit 11a communicates with the substrate processing apparatus 3 via the communication unit 13 to acquire particle data (particle map data and substrate processing conditions, etc.) obtained by measuring particles generated in the chamber and adhering to the substrate during substrate processing, and stores the acquired particle data in the process DB 4. In the information processing system according to this embodiment, the user experimentally performs substrate processing with the substrate processing apparatus 3 under various conditions and obtains particle map data by measuring particles with sensors etc. provided by the substrate processing apparatus 3. For example, the user can place nanoparticles that become particles or pieces of apparatus components that can cause particle generation at predetermined positions in the chamber of the substrate processing apparatus 3, or intentionally damage components inside the chamber in advance and subject them to wear processing to create conditions that make it easier for particles to be generated during substrate processing, and then perform experimental substrate processing with the substrate processing apparatus 3 under various combinations of conditions such as pressure or temperature to obtain particle map data under various conditions.

[0032] In this embodiment, the data acquisition unit 11a uses the acquired particle data to receive information such as substrate processing conditions as input, and generates a map prediction model by machine learning that predicts particle map data or its features (mean, variance, or standard deviation, etc.) corresponding to this information. In the information processing system according to this embodiment, the data acquisition unit 11a generates a Bayesian model as the map prediction model. The data acquisition unit 11a uses the generated map prediction model to determine the conditions for the next experiment to be performed, for example, from the confidence interval of the prediction of this map prediction model, and presents the information regarding the determined experimental conditions to the user by displaying it on the display unit 14. In this embodiment, the data acquisition unit 11a determines the experimental conditions using a Bayesian optimization method, but is not limited to this, and any method may be used to determine the experimental conditions. By the user performing experiments under the presented experimental conditions, the information processing device 1 can collect more useful particle data into the process DB 4.

[0033] Furthermore, the data acquisition unit 11a accepts labeling operations, or so-called annotation operations, from the user for the particle data collected in the process DB 4, and generates training data (training data) that associates particle data with label information. The data acquisition unit 11a reads the particle map data of the particle data stored in the process DB 4 and classifies the multiple particles contained in the read particle map data into multiple clusters based on information such as position and size (clustering). Various methods of so-called unsupervised learning, such as the shortest distance method, longest distance method, group average method, k-means, or DBSCAN (Density Based Spatial Clustering of Applications with Noise), can be employed for the clustering process performed by the data acquisition unit 11a.

[0034] The data acquisition unit 11a, for example, graphs the particle map data and displays it on the display unit 14, and also distinguishes and displays multiple particles based on the clustering results by color-coding according to the cluster. The data acquisition unit 11a may accept correction operations on the displayed clustering results from the user. The data acquisition unit 11a accepts operations from the user to label the dust source for each classified cluster, assigns one label to each particle included in the particle data, and stores the data of the dust source label for each particle in the process DB 4 in association with the particle data. This data stored in the process DB 4 is training data used in machine learning to generate the dust source prediction model 5.

[0035] The data acquisition unit 11a also divides the particle map data into map data for each dust source based on the dust source label, and displays the divided map data for each dust source on the display unit 14. The data acquisition unit 11a receives a labeling operation from the user for the recovery method for the map data for each dust source, and stores the information such as substrate processing conditions and results included in the particle data, the map data for each dust source, and the label for the recovery method in the process DB 4. This data stored in the process DB 4 is the training data used in machine learning to generate the recovery method prediction model 6.

[0036] The model generation unit 11b generates a dust source prediction model 5 and a recovery method prediction model 6 by performing machine learning processing using the training data stored in the process DB 4 through processing by the data acquisition unit 11a. The dust source prediction model 5 and the recovery method prediction model 6 in this embodiment may employ various configurations, such as SVM or neural networks. The model generation unit 11b can generate the dust source prediction model 5 by performing so-called supervised machine learning processing using training data in which particle map data included in the particle data is associated with labels related to the dust source of each particle. The model generation unit 11b can also generate the recovery method prediction model 6 by performing so-called supervised machine learning processing using training data in which map data for each dust source is associated with information such as substrate processing conditions and results and labels related to the recovery method. The model generation unit 11b stores information such as the internal parameters of the generated dust source prediction model 5 and recovery method prediction model 6 in the model DB 7.

[0037] The dust source separation unit 11c performs a process to separate multiple particles generated in the chamber of the substrate processing apparatus 3 according to their dust source, based on the particle data obtained when the substrate processing apparatus 3 performs substrate processing. The dust source separation unit 11c constructs a dust source prediction model 5 based on the information stored in the model DB 7 and inputs the particle map data included in the particle data acquired from the substrate processing apparatus 3 to the dust source prediction model 5. In response to this input, the dust source prediction model 5 outputs the classification result of the dust source for each particle, and the dust source separation unit 11c generates multiple map data for each dust source by separating the multiple particles included in the particle map data according to their dust source based on the classification result of the dust source output by the dust source prediction model 5.

[0038] The recovery method prediction unit 11d performs a process to predict a recovery method for each dust source for particles generated in the chamber of the substrate processing apparatus 3, based on the map data for each dust source obtained as a result of the separation by the dust source separation unit 11c. The recovery method prediction unit 11d inputs one of the map data for each of the multiple dust sources separated by the dust source separation unit 11c, along with information such as the substrate processing conditions and results included in the original particle data, to the recovery method prediction model 6. The recovery method prediction unit 11d predicts a recovery method according to the dust source by acquiring information on the recovery method output by the recovery method prediction model 6 in response to this input.

[0039] The display processing unit 11e performs the processing of displaying various characters and images on the display unit 14. In this embodiment, for example, the display processing unit 11e graphs the particle map data included in the particle data acquired from the substrate processing apparatus 3 and displays it on the display unit 14. The display processing unit 11e also displays information on the experimental conditions for substrate processing determined by the data acquisition unit 11a on the display unit 14. The display processing unit 11e also displays the progress of the machine learning processing performed by the model generation unit 11b, for example, a graph showing the change in the prediction error of the learning model on the display unit 14. The display processing unit 11e also graphs the map data for each dust source generated by the dust source separation unit 11c and displays it on the display unit 14. The display processing unit 11e also displays information on the recovery method for each dust source predicted by the recovery method prediction unit 11d on the display unit 14. The display processing unit 11e may also display various other information on the display unit 14.

[0040] <Data Acquisition Processing> In the information processing system according to this embodiment, for example, in the preparation stage before the substrate processing apparatus 3 is put into operation in a factory, experiments are conducted under various conditions to generate particles inside the chamber of the substrate processing apparatus 3. In these experiments, the user places nanoparticles that will become particles in predetermined locations inside the chamber of the substrate processing apparatus 3, and performs substrate processing on the substrate processing apparatus 3 under predetermined substrate processing conditions. The user can appropriately determine the size and placement of the nanoparticles by assuming, for example, components inside the chamber that could become particle sources. Furthermore, even if the size and placement of the nanoparticles are the same, if the substrate processing conditions are different, the position and amount of particles adhering to the substrate after substrate processing may differ. For this reason, it is preferable to conduct experiments with various combinations of nanoparticle size and placement locations and substrate processing conditions.

[0041] The substrate processing apparatus 3 performs experimental substrate processing according to the substrate processing conditions set by the user, and then measures the position and size of particles attached to the substrate. This measurement may be performed by a device other than the substrate processing apparatus 3. In this embodiment, the substrate processing apparatus 3 measures the position and size of each particle attached to the substrate, and generates three-dimensional data having three values: x-coordinate, y-coordinate, and size for each particle. The substrate processing apparatus 3 generates this three-dimensional data for multiple particles attached to the substrate, and combines the point cloud data of the multiple three-dimensional data to create particle map data. In this embodiment, three-dimensional data having three values: x-coordinate, y-coordinate, and size of the particle is generated to create particle map data, but this is not limited to this, and multi-dimensional data that associates the x-coordinate and y-coordinate of the particle with size, components, or other related information may be generated to create particle map data.

[0042] Furthermore, the substrate processing apparatus 3 can obtain information regarding the results of the substrate processing by measuring, for example, the amount of excavation (etching amount) of the substrate when etching is performed as a substrate processing. The substrate processing apparatus 3 also records time-series data (log data) of numerical values ​​such as temperature or pressure measured by various internal sensors during the substrate processing as information regarding the results of the substrate processing. The substrate processing apparatus 3 generates particle data that combines the above particle map data, information regarding the results of the substrate processing, and information regarding the conditions of the performed substrate processing, and transmits the generated particle data to the information processing apparatus 1. Regarding information regarding the conditions of the substrate processing, if the information processing apparatus 1 already possesses this information, it does not need to be included in the particle data transmitted from the substrate processing apparatus 3 to the information processing apparatus 1.

[0043] The information processing apparatus 1 according to the present embodiment performs data collection processing by acquiring particle data obtained by substrate processing experimentally performed by the substrate processing apparatus 3 and storing it in the process DB 4. FIG. 3 is a flowchart showing an example of the procedure of data collection processing performed by the information processing apparatus 1 according to the present embodiment. The data collection unit 11a of the processing unit 11 of the information processing apparatus 1 according to the present embodiment acquires the conditions of the substrate processing experimentally performed by the substrate processing apparatus 3, for example, by receiving an input from the user (step S1). The data collection unit 11a transmits the conditions of the substrate processing acquired in step S1 to the substrate processing apparatus 3, thereby causing the substrate processing apparatus 3 to perform substrate processing according to these conditions (step S2). In this example, it is assumed that the information processing apparatus 1 receives an input of the conditions of the substrate processing from the user and controls the substrate processing by the substrate processing apparatus 3. However, the present invention is not limited to this. For example, a configuration in which the user inputs the conditions of the substrate processing to the substrate processing apparatus 3 may be employed. In this case, the information processing apparatus 1 may not perform the processing of steps S1 and S2.

[0044] The data collection unit 11a acquires the particle data transmitted from the substrate processing apparatus 3 that has performed the substrate processing by receiving it with the communication unit 13 (step S3). The data collection unit 11a performs clustering processing on the particle map data included in the particle data acquired in step S3 (step S4). The data collection unit 11a performs clustering of the particle map data using an existing clustering algorithm, whereby a plurality of particles included in the particle map data can be classified into a plurality of clusters. The data collection unit 11a displays the result of the clustering in step S4 on the display unit 14 (step S5).

[0045] Figure 4 is a schematic diagram showing an example of how clustering results are displayed. In this embodiment, the data acquisition unit 11a generates a graph on a two-dimensional plane in which multiple points corresponding to particles are drawn based on the x and y coordinates of each particle included in the particle map data, and displays it on the display unit 14. In this graph, the data acquisition unit 11a provides the clustering results to the user by drawing the points of each particle in a color corresponding to the classified cluster (in this figure, the color coding is represented by making the shape of each point different). In this example, it is shown that multiple particles included in the particle map data are classified into three clusters, clusters 1 to 3.

[0046] The data acquisition unit 11a accepts user input to the operation unit 15 while displaying the clustering results, thereby accepting modifications to the clusters of each particle in the clustering results (step S6). Here, the data acquisition unit 11a accepts the selection of one or more particles from among the displayed particles, and accepts changes to the clusters related to the selected particles. The data acquisition unit 11a may also accept modifications such as merging multiple clusters or splitting one cluster into multiple clusters. The data acquisition unit 11a modifies the clustering results according to the accepted modifications.

[0047] Next, the data acquisition unit 11a accepts the user's operation of the operation unit 15 and accepts the setting of labels for each cluster obtained by the clustering process (step S7). In this embodiment, the data acquisition unit 11a accepts the setting of two types of labels: a label indicating which is the dust source and a label indicating which is the appropriate recovery method. For example, the label indicating which is the dust source may be parts A, B, C, etc. of the substrate processing apparatus 3. For example, the label indicating which is the recovery method may be processing condition adjustment, cleaning timing / time adjustment, replacement of specific parts, and cleaning / coating. The above labels are just examples and are not limited to these.

[0048] The data collection unit 11a stores, in the process DB 4, data in which the particle data acquired in step S3 is associated with information regarding the label whose setting was received in step S7 (step S8). These data stored in the process DB 4 are used as learning data (teacher data) for performing supervised machine learning.

[0049] Next, the data collection unit 11a generates a map prediction model that predicts the feature amount of the particle map data based on the substrate processing conditions, based on the particle data obtained so far (step S9). In the present embodiment, the map prediction model receives, as inputs, for example, the substrate processing conditions (recipe) and information such as the size and position of the nanoparticles arranged in the chamber, and outputs information such as the number and distribution tendency of the particles adhering to the substrate in the chamber when substrate processing is performed under these conditions. The data collection unit 11a generates the map prediction model as a Bayesian model in step S9 based on the method of Bayesian optimization, and evaluates the number and distribution tendency of the particles, which are the output information of the model, as the confidence interval of the map prediction model, thereby determining the conditions of the experiment to be additionally performed (step S10).

[0050] The data collection unit 11a determines whether or not the condition for ending the experiment is satisfied, for example, when there are no conditions for the experiment to be additionally performed in the process of step S10 (step S11). For example, when there are conditions for the experiment to be additionally performed and the condition for ending the experiment is not satisfied (S11: NO), the data collection unit 11a displays the experiment conditions determined in step S10 on the display unit 14 (step S12), and returns the process to step S1. The user confirms the experiment conditions displayed on the display unit 14 and performs a new experiment with the substrate processing apparatus 3, and the data collection unit 11a can repeatedly perform the above-described processing to collect data in the process DB 4 by acquiring the substrate processing result of the performed experiment. When the condition for ending the experiment is satisfied (S11: YES), the data collection unit 11a ends the data collection process.

[0051] Figure 5 is a schematic diagram showing an example of how experimental conditions are displayed. In step S12 of the flowchart described above, the information processing device 1 according to this embodiment displays the experimental condition suggestion screen shown on the display unit 14. In the experimental condition suggestion screen in this example, the title string "Suggested Experimental Conditions" is displayed at the top, and below it, multiple experimental conditions are listed in a table format. In this example, the experimental conditions include multiple condition items such as "particle size," "low RF (Radio Frequency) power," "high RF power," and "shutter temperature." In this example, the numerical values ​​of multiple sets of experimental conditions determined by the data acquisition unit 11a in step S10 are shown in correspondence with these condition items. The user can select one of the multiple suggested experimental conditions and have the substrate processing device 3 perform experimental substrate processing with the selected conditions.

[0052] <Model Generation Process> After collecting data in the process DB 4, the information processing device 1 performs a process to generate a dust source prediction model 5 and a recovery method prediction model 6 using machine learning with the collected data. Figure 6 is a schematic diagram showing one example configuration of the dust source prediction model 5. The dust source prediction model 5 according to this embodiment is a learning model that has been pre-generated by machine learning to accept particle map data as input and output the classification result of the dust source for each particle. In this embodiment, one particle is represented as individual information of three-dimensional data having three values: coordinates x, y of the attachment position on the substrate and size s, and the particle map data is a collection of three-dimensional data for multiple particles. The number of particles included in one particle map data is variable and is the number of particles generated in the chamber of the substrate processing device 3. That is, the data length of the particle map data is variable. The dust source prediction model 5 accepts three-dimensional data of N particles (N is a natural number) included in the particle map data as input and outputs labels indicating N dust sources corresponding to each particle. The number of 3D data inputs (N) that the dust source prediction model 5 accepts is set to a sufficiently large number for the particles that may be generated in the chamber of the substrate processing device 3. However, if the number of particles included in the particle map data is greater than N, the information processing device 1 may appropriately decimate the number of particles by removing 3D data of particles from the particle map data to reduce the number of particles to N or less. The dust source prediction model 5 is a learning model in which the order of the corresponding classification result labels is changed when the order of inputs of multiple 3D data is changed. In this embodiment, the dust source prediction model 5 accepts 3D data having three values: x-coordinate, y-coordinate, and size of the particle as input, but it is not limited to this, and the dust source prediction model 5 may also accept multidimensional data as input that associates the x-coordinate and y-coordinate of the particle with size, components, or other related information.

[0053] The model generation unit 11b of the information processing device 1 reads particle map data and the corresponding dust source labels stored in the process DB 4, and generates a dust source prediction model 5 by performing so-called supervised machine learning processing using training data (teaching data) that associates the read particle map data and dust source labels. For example, the model generation unit 11b inputs the particle map data of the training data to the dust source prediction model 5, which has appropriate initial parameters set, calculates the error between the dust source prediction result output by the dust source prediction model 5 and the dust source labels of the training data, and performs machine learning by updating the internal parameters of the dust source prediction model 5 using a method such as backpropagation based on the calculated error. For example, the model generation unit 11b repeatedly updates the internal parameters until the error falls below a threshold, and finally stores the obtained internal parameters in the model DB 7 as information about the dust source prediction model 5.

[0054] Figure 7 is a schematic diagram showing one example configuration of the recovery method prediction model 6. The recovery method prediction model 6 according to this embodiment is a learning model generated in advance by machine learning that accepts map data for each dust source, substrate processing conditions, and substrate processing results as input, and outputs a classification result of the recovery method according to these. The map data for each dust source is obtained by dividing the particle map data into multiple map data for each dust source label. The model generation unit 11b of the information processing device 1 reads the particle map data, substrate processing conditions, and substrate processing result information contained in the particle data stored in the process DB 4, and the corresponding recovery method labels for each dust source, and generates the recovery method prediction model 6 by performing so-called supervised machine learning processing using the learning data (training data) that associates these.

[0055] Furthermore, the information processing device 1 according to this embodiment increases the amount of data by performing data augmentation on the training data stored in the process DB 4, thereby improving the accuracy of the learning model generated by machine learning using this data. Figure 8 is a schematic diagram illustrating an example of data augmentation. In the example shown in the upper part of Figure 8, the model generation unit 11b of the information processing device 1 generates new particle map data (augmented data) by, for example, randomly extracting (sampling) multiple particles from the particle map data (original data) stored in the process DB 4. The model generation unit 11b can generate multiple particle map data from one particle map data by appropriately changing the number of particles extracted from the original particle map data.

[0056] In the example shown in the lower part of Figure 8, the model generation unit 11b of the information processing device 1 generates new particle map data (extended data) by rotating the coordinates of each particle in the particle map data (original data) stored in the process DB 4 by approximately 90 degrees clockwise with respect to the center of the substrate. The model generation unit 11b can generate multiple particle map data from one particle map data by appropriately changing the angle at which the coordinates are rotated relative to the original particle map data.

[0057] Although not shown in the diagram, the model generation unit 11b can perform data augmentation in a manner different from the sampling and rotation described above. For example, the model generation unit 11b can generate new particle map data by combining multiple particle map data stored in the process DB 4 into a single particle map data. Alternatively, the model generation unit 11b can generate new particle map data by perturbing the original particle map data, i.e., by adding minute noise to the position or size of each particle. The model generation unit 11b may perform data augmentation in any manner different from the above, and may perform data augmentation by combining multiple methods.

[0058] Furthermore, the information processing device 1 according to this embodiment can perform data augmentation using a learning model (generative model) that generates particle map data. Figure 9 is a schematic diagram illustrating an example of a method for generating particle map data using a generative model. The generative model 8a shown in Figure 9 is a learning model that has been pre-generated by machine learning to accept information on particle map data, substrate processing conditions, and substrate processing results as input and output distribution parameters of particle map data. The generative model 8a is pre-generated by supervised machine learning using, for example, learning data (training data) that associates particle data (particle map data, substrate processing conditions, and substrate processing results) obtained from the substrate processing device 3 with distribution parameters of particle map data predetermined based on this particle data.

[0059] The information processing device 1 pre-stores information about the generation model 8a in the model DB 7. The model generation unit 11b inputs particle map data, substrate processing conditions, and substrate processing results information contained in the particle data stored in the process DB 4 to the generation model 8a and obtains the distribution parameters output by the generation model 8a. The distribution parameters are, for example, the coefficients of a predetermined distribution function. The model generation unit 11b generates particle map data by performing appropriate probability calculations based on the distribution function represented by the distribution parameters obtained from the generation model 8a, thereby generating data for multiple discrete particles that follow this distribution. The model generation unit 11b can generate multiple particle map data based on one (or one set) of distribution parameters generated by the generation model 8a, and can expand the data used for machine learning.

[0060] Figure 10 is a schematic diagram illustrating another example of a method for generating particle map data using a generative model. The generative model 8b shown in Figure 10 is a pre-trained model generated by machine learning to accept particle map data, substrate processing conditions, and substrate processing results as input and output particle map data. The generative model 8b is pre-trained by machine learning, for example, by creating data by appropriately thinning out several particles from the particle map data included in the particle data obtained from the substrate processing apparatus 3, and then reconstructing the original particle map data when this thinned-out data is input along with the substrate processing conditions and substrate processing results. As a result, the trained generative model 8b can generate particle map data with an increased number of particles compared to the input particle map data.

[0061] The information processing device 1 pre-stores information about the generated model 8b in the model DB 7. The model generation unit 11b inputs particle map data, substrate processing conditions, and substrate processing results information contained in the particle data stored in the process DB 4 to the generated model 8b and acquires the particle map data output by the generated model 8b. As a result, the model generation unit 11b can generate particle map data with an increased number of particles compared to the particle map data input to the generated model 8b, thereby expanding the data used for machine learning.

[0062] Figure 11 is a flowchart showing an example of the procedure for generating a learning model performed by the information processing device 1 according to this embodiment. The model generation unit 11b of the processing unit 11 of the information processing device 1 according to this embodiment reads the learning data stored in the process DB 4 (step S21). In this embodiment, the learning data for generating the dust source prediction model 5 is data in which particle map data included in particle data and labels indicating which dust source each particle included in the particle map data is associated. In this embodiment, the learning data for generating the recovery method prediction model 6 is data in which particle map data included in particle data, substrate processing conditions and substrate processing results information and labels indicating which recovery method is appropriate are associated.

[0063] The model generation unit 11b performs data augmentation processing on the training data read in step S21 (step S22). The model generation unit 11b generates new particle map data and label sets by combining, for example, the particle map data contained in the training data read in step S21 and the labels attached to each particle contained in this particle map data, and performing appropriate processing such as sampling, rotation, joining, or perturbation.

[0064] The model generation unit 11b may also generate new particle map data using the generation model 8a shown in Figure 9 or the generation model 8b shown in Figure 10. In this case, the model generation unit 11b can, for example, divide the particle map data included in the training data by label and input it into the generation model 8a or 8b, assign the same label to the new particle map data generated based on the output of the generation model 8a or 8b, and combine the new particle map data generated for each label to form a single set of particle map data and label.

[0065] The model generation unit 11b inputs the input data included in the training data obtained in steps S21 and S22 to the target training model (step S23). Specifically, the model generation unit 11b inputs the particle map data included in the training data to the dust source prediction model 5, or inputs the particle map data, substrate processing conditions, and substrate processing result information included in the training data to the recovery method prediction model 6. Next, the model generation unit 11b acquires the data output by the training model in response to the data input in step S23 (step S24). Specifically, the model generation unit 11b acquires the classification result of the dust source output by the dust source prediction model 5, or the classification result of the recovery method output by the recovery method prediction model 6.

[0066] The model generation unit 11b calculates the error between the output data of the learning model acquired in step S24 and the labels of the correct values ​​included in the training data (step S25). Based on the error value calculated in step S25, the model generation unit 11b updates the internal parameters of the learning model, for example, based on backpropagation (step S26). The model generation unit 11b determines whether a termination condition has been met, for example, that the error calculated in step S25 is below a threshold or that the number of training iterations has reached a predetermined number (step S27). If the termination condition has not been met (S27: NO), the model generation unit 11b returns to step S23 and repeatedly updates the internal parameters of the learning model. If the termination condition has been met (S27: YES), the model generation unit 11b stores the internal parameters of the learning model at this point in the model DB 7 (step S28) and terminates the process.

[0067] <Particle Prediction Processing> In the information processing system according to this embodiment, information regarding the dust source prediction model 5 and the recovery method prediction model 6, which are generated in advance during the preparation stage, is stored in the model DB 7 of the information processing device 1. In the implementation stage, when the substrate processing device 3 is actually operated to perform mass production, the information processing device 1 analyzes the particles generated in the chamber of the substrate processing device 3 based on the particle data obtained from the substrate processing device 3 that has performed substrate processing, using the dust source prediction model 5 and the recovery method prediction model 6 stored in the model DB 7, and provides the results to the user.

[0068] Figure 12 is a flowchart showing an example of the procedure for prediction processing performed by the information processing device 1 according to this embodiment. The processing unit 11 of the information processing device 1 according to this embodiment transmits information on substrate processing conditions (recipe) set by the user to the substrate processing device 3 and gives an instruction to perform the substrate processing, thereby enabling the substrate processing device 3 to perform the desired substrate processing (step S41). After the substrate processing by the substrate processing device 3 is completed, the processing unit 11 acquires particle data including information such as particle map data, substrate processing conditions, and substrate processing results from the substrate processing device 3 (step S42).

[0069] The dust source separation unit 11c of the processing unit 11 inputs the particle map data included in the particle data acquired in step S42 to the dust source prediction model 5 stored in the model DB 7 (step S43). In response to this data input, the dust source prediction model 5 outputs a classification result indicating which dust source each particle included in the particle map data is. The dust source separation unit 11c acquires the classification result of the dust source output by the dust source prediction model 5 (step S44). Based on the classification result acquired in step S44, the dust source separation unit 11c separates the particles included in the particle map data according to their dust source and generates map data for each dust source (step S45).

[0070] The recovery method prediction unit 11d of the processing unit 11 inputs one of the map data for each dust source generated in step S45, and information such as substrate processing conditions and substrate processing results included in the particle data acquired in step S42, to the recovery method prediction model 6 stored in the model DB 7 (step S46). In response to this data input, the recovery method prediction model 6 outputs a classification result of a recovery method suitable for this map data for each dust source. The recovery method prediction unit 11d acquires the classification result of the recovery method output by the recovery method prediction model 6 (step S47).

[0071] The display processing unit 11e of the processing unit 11 displays the map data for each dust source generated in step S45, and information regarding the recovery method based on the classification result of the recovery method acquired in step S47, on the display unit 14 (step S48), and then terminates the process. The information processing device 1 has, for example, information to be presented to the user regarding the recovery method stored in the storage unit 12 in advance, and can retrieve the necessary information from the storage unit 12 based on the classification result acquired in step S47 and display the information.

[0072] <Modification> In the information processing system described above, the recovery method prediction model 6 is a classification model that classifies recovery methods based on the input data. In contrast, this modification describes a configuration in which the information processing device 1 provides information about recovery methods to the user using the recovery method prediction model 6, which outputs feature quantities corresponding to the input data.

[0073] Figure 13 is a schematic diagram showing one example configuration of the recovery method prediction model 6 according to the modified example. The recovery method prediction model 6 according to the modified example is a so-called encoder learning model that is pre-generated by machine learning to accept particle map data (or map data for each dust source) obtained from the substrate processing apparatus 3, substrate processing conditions, and the results of substrate processing as inputs, and to output vector information of feature quantities corresponding to this input information. Since the encoder learning model is an existing technology, a detailed explanation will be omitted, but for example, an encoder that converts input information into feature quantities and a decoder that converts these feature quantities back into input information can be combined, and a learning model of the encoder and decoder can be generated by performing machine learning so that the input information to the encoder and the output information to the decoder match. In the recovery method prediction model 6 according to the modified example, representation learning is performed so that the distance between feature quantity vectors becomes small when the dust source is the same, and the distance between feature quantity vectors becomes large when the dust source is different.

[0074] Furthermore, the modified information processing device 1 stores in a database the particle data (particle map data, substrate processing conditions, and substrate processing results, etc.) obtained by the substrate processing device 3 performing substrate processing, the feature quantities obtained by the recovery method prediction model 6 based on this particle data, and information regarding the recovery method performed on this particle data, in association with each other.

[0075] In the modified information processing device 1, when new particle data is obtained after substrate processing is performed by the substrate processing device 3, this particle data is converted into features using the recovery method prediction model 6, and the similarity (e.g., cosine similarity) with the features of each particle map data stored in the database is calculated. Based on the calculated similarity, the information processing device 1 selects a predetermined number of particle data from the database in descending order of similarity, and displays the selected particle data and the information on the recovery method associated with them on the display unit 14. At this time, the information processing device 1 may rank and display multiple pieces of information in descending order of similarity.

[0076] <Summary> In the information processing system according to this embodiment with the above configuration, the information processing device 1 acquires particle data (particle map data, substrate processing conditions, and substrate processing results, etc.) including positional information of particles generated in the chamber of the substrate processing device 3. The information processing device 1 receives the particle data as input and inputs the acquired particle data to a recovery method prediction model 6 that has been trained by machine learning to output information on particle recovery methods. The information processing device 1 acquires information on recovery methods output by the recovery method prediction model 6 in response to the input of particle data and displays (outputs) the acquired information on recovery methods on the display unit 14. As a result, the information processing system according to this embodiment can provide the user with information on recovery methods according to particle data and is expected to support measures against particles generated in the chamber of the substrate processing device 3.

[0077] Furthermore, in the information processing system according to this embodiment, the recovery method prediction model 6 is a classification model that classifies the input particle data into multiple recovery methods. As a result, the information processing system according to this embodiment can, for example, read information on recovery methods according to the classification results from a database and provide it to the user, and is expected to support measures against particles generated in the chamber of the substrate processing apparatus 3.

[0078] Furthermore, in the information processing system according to this embodiment, the recovery method prediction model 6 is a learning model for an encoder that converts input particle data into features and outputs them. The information processing device 1 acquires the features output by the recovery method prediction model 6 according to the input particle data, and extracts information from a database that stores information on recovery methods based on the acquired features. The information processing device 1 also extracts multiple pieces of information from the database based on the similarity between the features acquired from the recovery method prediction model 6 and the features of the information stored in the database, and displays the extracted pieces of information in order. As a result, the information processing system according to this embodiment can provide the user with multiple pieces of information on recovery methods according to particle data, and is expected to support particle countermeasures generated in the chamber of the substrate processing device 3.

[0079] Furthermore, in the information processing system according to this embodiment, the information processing device 1 predicts the dust source for each particle based on the particle data acquired from the substrate processing device 3, and divides the particle data (particle map data) into multiple individual particle data (map data for each dust source) based on the prediction result. The information processing device 1 inputs the divided map data for each dust source into the recovery method prediction model 6, acquires information on the recovery method output by the recovery method prediction model 6, and displays the acquired information on the recovery method for each dust source. As a result, the information processing system according to this embodiment can provide the user with information on the recovery method that takes into account which dust source the particle is, and is expected to support measures against particles generated in the chamber of the substrate processing device 3.

[0080] Furthermore, in the information processing system according to this embodiment, the information processing device 1 receives particle data (particle map data) as input and inputs the particle data acquired from the substrate processing device 3 to a dust source prediction model 5 that has been trained by machine learning to output information regarding the dust source of each particle. The information processing device 1 predicts the dust source for each particle by acquiring the information output by the dust source prediction model 5 in response to the input of particle data. As a result, the information processing system according to this embodiment can predict the dust source of particles with greater accuracy and is expected to support measures against particles generated in the chamber of the substrate processing device 3.

[0081] Furthermore, in the information processing system according to this embodiment, the dust source prediction model 5 is generated by machine learning using training data that associates particle data (particle map data) with the labels of each particle contained in this particle data. The information processing device 1 acquires particle data from the substrate processing device 3 and performs clustering based on the acquired particle data to classify multiple particles into multiple clusters. The information processing device 1 accepts corrections to the cluster classification results for each particle from the user and also accepts the setting of labels for each cluster, and generates data that associates particle data with the labels of each particle contained in this particle data as training data. As a result, the information processing system according to this embodiment is expected to facilitate the input of the labels of each particle necessary for training data and is expected to support measures against particles generated in the chamber of the substrate processing device 3.

[0082] Furthermore, in the information processing system according to this embodiment, the information processing device 1 acquires particle data including positional information (particle map data) of particles generated in the chamber of the substrate processing device 3, and performs clustering based on the acquired particle data to classify multiple particles into multiple clusters. The information processing device 1 accepts corrections to the cluster classification results for each particle from the user and also accepts the setting of labels for each cluster, generates data that associates the particle data with the labels of each particle contained in this particle data as training data, and stores the generated training data in the process DB 4. As a result, the information processing system according to this embodiment can collect and store training data used for machine learning to generate a learning model in the process DB 4, and is expected to support countermeasures against particles generated in the chamber of the substrate processing device 3.

[0083] Furthermore, in the information processing system according to this embodiment, the information processing device 1 acquires particle data measured when the substrate processing device 3 is operated under predetermined substrate processing conditions with a particle source such as nanoparticles placed inside the chamber of the substrate processing device 3. As a result, the information processing system according to this embodiment can collect particle data under a wider variety of conditions and is expected to support measures against particles generated inside the chamber of the substrate processing device 3.

[0084] Furthermore, in the information processing system according to this embodiment, the information processing device 1 predicts the operating conditions for the substrate processing device 3, which should acquire more particle data, based on the acquired particle data, and displays (outputs) the prediction results. As a result, the information processing system according to this embodiment is expected to present the user with the conditions for substrate processing and enable efficient collection of particle data, and is expected to support measures against particles generated in the chamber of the substrate processing device 3.

[0085] Furthermore, in the information processing system according to this embodiment, the information processing device 1 uses the training data stored in the process DB 4 to generate a recovery method prediction model 6 that accepts particle data as input and outputs information regarding particle recovery methods, and a dust source prediction model 5 that accepts particle data as input and outputs information regarding the dust source of each particle, using machine learning. As a result, the information processing system according to this embodiment can pre-generate the recovery method prediction model 6 and the dust source prediction model 5 based on the collected particle data and store them in the model DB 7 of the information processing device 1, and can perform dust source prediction and recovery method prediction based on the particle data obtained in conjunction with the substrate processing performed by the substrate processing device 3.

[0086] Furthermore, in the information processing system according to this embodiment, the information processing device 1 acquires particle data including positional information (particle map data) of particles generated in the chamber of the substrate processing device 3, and inputs the acquired particle data to a dust source prediction model 5 that has been trained by machine learning to accept the particle data as input and output information regarding the dust source of each particle. The information processing device 1 acquires the information regarding the dust source output by the dust source prediction model 5 in response to the input of particle data, and displays (outputs) the acquired information regarding the dust source. As a result, the information processing system according to this embodiment can provide the user with information regarding the dust source of particles predicted based on the particle data acquired from the substrate processing device 3, and is expected to support measures against particles generated in the chamber of the substrate processing device 3.

[0087] Furthermore, in the information processing system according to this embodiment, the particle data (particle map data) handled by the information processing device 1 is data (point cloud data) collected for multiple particles generated in the chamber of the substrate processing device 3, which associate coordinate information indicating the position of the particle with values ​​indicating characteristics such as the size of the particle. As a result, the information processing system according to this embodiment can handle particle data of a variable size corresponding to the number of particles generated.

[0088] Furthermore, in the information processing system according to this embodiment, the dust source prediction model 5 receives multiple individual pieces of information about multiple particles as input and outputs multiple classification results of dust sources corresponding to the multiple individual pieces of information that have been input. As a result, the information processing system according to this embodiment is expected to maintain the dust source prediction results for each particle even if the order of the multiple individual pieces of information input to the dust source prediction model 5 is changed.

[0089] Furthermore, in the information processing system according to this embodiment, the information processing device 1 acquires particle data, which includes positional information (particle map data) of multiple particles generated in the chamber of the substrate processing device 3, and training data that associates information about the dust source of each particle. The acquired machine learning then generates a dust source prediction model 5. The dust source prediction model 5 accepts particle data as input and outputs information about the dust source of each particle. As a result, the information processing system according to this embodiment is expected to accurately predict the dust source of each particle based on the positional information of the particles generated in the chamber of the substrate processing device 3, and is expected to support measures against particles generated in the chamber of the substrate processing device 3.

[0090] Furthermore, in the information processing system according to this embodiment, the information processing device 1 performs data augmentation on the acquired training data and generates a dust source prediction model 5 by machine learning using the data augmented training data. Data augmentation may include processes such as particle sampling, particle movement, particle perturbation, and superposition of different particle data. As a result, the information processing system according to this embodiment can increase the amount of training data used for machine learning, and is expected to improve the prediction accuracy of the dust source prediction model 5 generated by machine learning.

[0091] Furthermore, in the information processing system according to this embodiment, the information processing device 1 acquires substrate processing information (particle data) including substrate processing conditions and substrate processing results by the substrate processing device 3, and inputs the acquired particle data into a generation model 8a that has been pre-trained to accept particle data as input and output particle distribution parameters. The information processing device 1 acquires the distribution parameters output by the generation model 8a in response to this data input, generates particle data (particle map data) using the acquired distribution parameters, and generates a dust source prediction model 5 by machine learning using training data including the generated particle data. As a result, the information processing system according to this embodiment can increase the amount of training data used for machine learning, and is expected to improve the prediction accuracy of the dust source prediction model 5 generated by machine learning.

[0092] Furthermore, in the information processing system according to this embodiment, the information processing device 1 acquires substrate processing information (particle data) including substrate processing conditions and substrate processing results by the substrate processing device 3, and inputs the acquired particle data to a generation model 8b that has been pre-trained by machine learning to accept particle data as input and output particle data (particle map data). In response to this data input, the generation model 8b generates particle data (particle map data) to be output, and a dust source prediction model 5 is generated by machine learning using training data including the generated particle data. As a result, the information processing system according to this embodiment can increase the amount of training data used for machine learning, and it is expected that the prediction accuracy of the dust source prediction model 5 generated by machine learning can be improved.

[0093] In this embodiment, the particle map data obtained by measuring particles generated in the chamber of the substrate processing apparatus 3 is a collection of three-dimensional data (point cloud data) that includes three pieces of information for each particle: x-coordinate, y-coordinate position information, and the size of each particle. However, it is not limited to this. The configuration may include four or more pieces of data for each particle. For example, the position information may be three-dimensional coordinates of x-coordinate, y-coordinate, and z-coordinate. In addition to size, various other pieces of information such as the type of material, color, or shape may be included as characteristics of each particle.

[0094] Furthermore, in this embodiment, point cloud data collected from three-dimensional data (multidimensional data) is used as particle map data, but this is not limited to this. For example, image data of the surface of a substrate to which particles are attached may be used as particle map data.

[0095] Furthermore, the configurations of the learning models, such as the dust source prediction model 5 shown in Figure 6, the recovery method prediction model 6 shown in Figure 7, the generation model 8a shown in Figure 9, and the generation model 8b shown in Figure 10, are examples only and are not limited thereto. For example, the dust source prediction model 5 may be configured to accept input information such as substrate processing conditions and / or substrate processing results in addition to particle map data. Also, for example, the recovery method prediction model 6 may not accept input information on substrate processing conditions and / or substrate processing results, but may be configured to classify recovery methods based on map data for each dust source. Also, for example, the generation models 8a and 8b may be configured to accept at least one piece of information from particle map data, substrate processing conditions, and substrate processing results as input.

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

[0097] The matters described in each embodiment can be combined with each other. Furthermore, the independent 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 use a form in which claims referencing two or more other claims (multi-claim form), but are not limited to this. A form in which multi-claims referencing at least one multi-claim (multi-multi-claim) may also be used.

[0098] 1, 1A, 1B Information processing device (computer) 3, 3A, 3B Substrate processing device 4 Process DB 5 Dust source prediction model (second learning model) 6 Recovery method prediction model (first learning model) 7 Model DB 8a, 8b Generation model 11 Processing unit 11a Data acquisition unit 11b Model generation unit 11c Dust source separation unit 11d Recovery method prediction unit 11e Display processing unit 12 Storage unit 12a Program (computer program) 13 Communication unit 14 Display unit 15 Operation unit 99 Recording medium N Network

Claims

1. A computer program that causes a computer to perform the following processes: acquire particle data including positional information of particles generated in the processing container of a substrate processing apparatus; input the acquired particle data into a first learning model that has been machine-trained to accept the particle data as input and output information regarding particle recovery methods; acquire the information regarding recovery methods output by the first learning model; and output the acquired information regarding recovery methods.

2. The computer program according to claim 1, wherein the first learning model is a classification model that classifies input particle data into a plurality of recovery methods.

3. The computer program according to claim 1 or 2, wherein the first learning model is a learning model that converts input particle data into features and outputs them, and the computer program acquires the features output by the first learning model and extracts information from a database storing information on a recovery method based on the acquired features.

4. A computer program according to claim 3, which extracts multiple pieces of information from the database based on the similarity between the acquired features and the features of the information stored in the database, and outputs the extracted pieces of information in a ranked order.

5. A computer program according to claim 1 or 2, which predicts the dust source for each particle based on the acquired particle data, divides the particle data into a plurality of individual particle data for each dust source based on the prediction result, inputs each of the divided individual particle data into the first learning model, acquires information on the recovery method output by the first learning model, and outputs the acquired information on the recovery method for each dust source.

6. The computer program according to claim 5, which inputs the acquired particle data into a second learning model that has been machine-trained to accept particle data as input and output information about the dust source of each particle, and obtains the information about the dust source output by the second learning model to predict the dust source for each particle.

7. The computer program according to claim 6, wherein the second learning model is generated by machine learning using training data which associates particle data with the labels of each particle contained in the particle data, the training data is generated by acquiring particle data which includes positional information of particles generated in the processing container of the substrate processing apparatus, performing clustering based on the acquired particle data to classify multiple particles into multiple clusters, accepting corrections to the cluster classification results for each particle, accepting the setting of labels for each cluster, and associating the particle data with the labels of each particle contained in the particle data.

8. A computer program that causes a computer to perform the following processes: acquire particle data including positional information of particles generated in the processing container of a substrate processing apparatus; perform clustering based on the acquired particle data to classify multiple particles into multiple clusters; accept the setting of labels for each classified cluster; and store training data in a database that associates the particle data with the labels of each particle contained in the particle data.

9. The computer program according to claim 8, which accepts a correction of the cluster classification result for each particle by the clustering and accepts the setting of a label for each corrected cluster.

10. A computer program according to claim 8 or 9, which acquires particle data measured when the substrate processing apparatus is operated under predetermined operating conditions with a particle source installed in the processing container.

11. A computer program according to claim 10, which predicts the operating conditions of a substrate processing apparatus 3 that should acquire further particle data based on the acquired particle data, and outputs the prediction result.

12. The computer program according to claim 8 or 9, wherein the label is a label for classifying particle recovery methods, and the program generates a first learning model by machine learning that accepts particle data as input and outputs information regarding particle recovery methods, using the learning data stored in the database.

13. The computer program according to claim 8 or 9, wherein the label is a label that classifies the dust source of each particle, and a second learning model is generated by machine learning using the training data stored in the database, which accepts particle data as input and outputs information regarding the dust source of each particle.

14. An information processing method comprising: an information processing device acquiring particle data including positional information of particles generated in the processing container of a substrate processing device; inputting the acquired particle data into a first learning model that has been machine-trained to accept the particle data as input and output information regarding a particle recovery method; acquiring the information regarding the recovery method output by the first learning model; and outputting the acquired information regarding the recovery method.

15. A method for generating training data, comprising: an information processing device acquiring particle data including positional information of particles generated in a processing container of a substrate processing device; performing clustering based on the acquired particle data to classify multiple particles into multiple clusters; accepting the setting of labels for each classified cluster; and storing training data in a database that associates the particle data with the labels of each particle contained in the particle data.

16. A method for generating a learning model, comprising: an information processing device acquiring particle data including positional information of particles generated in the processing container of a substrate processing device; clustering based on the acquired particle data to classify multiple particles into multiple clusters; accepting the setting of labels for each classified cluster; and generating a learning model by machine learning using training data that associates the particle data with the labels of each particle contained in the particle data, which accepts particle data as input and outputs information regarding the dust source of each particle.

17. An information processing device comprising a processing unit, wherein the processing unit acquires particle data including positional information of particles generated in the processing container of a substrate processing device, inputs the acquired particle data to a first learning model that has been machine-trained to accept the particle data as input and output information regarding a particle recovery method, acquires the information regarding the recovery method output by the first learning model, and outputs the acquired information regarding the recovery method.

18. An information processing device comprising a processing unit, wherein the processing unit acquires particle data including positional information of particles generated in the processing container of a substrate processing device, performs clustering based on the acquired particle data to classify multiple particles into multiple clusters, accepts the setting of labels for each classified cluster, and stores learning data in a database that associates the particle data with the labels of each particle contained in the particle data.

Images