Computer program, method for generating a learning model, information processing method, and information processing device

Abstract

The present application provides a computer program, a learning model generation method, an information processing method, and an information processing device that can reduce the effects of noise and the like in the state detection of substrate processing using a learning model. The computer program according to the present embodiment causes a computer to perform the following processing: acquire sensor data related to substrate processing; input the acquired sensor data to a learning model to acquire sensor data output by the learning model; calculate individual errors for each measured value of the sensor data input to the learning model and the sensor data output by the learning model; calculate a comprehensive error by weighting the plurality of individual errors calculated, respectively; update internal parameters of the learning model based on the comprehensive error calculated; and adjust the weight added to the individual error according to the convergence condition of the individual error.

Description

Technical Field

[0001] This invention relates to computer programs, methods for generating learning models, information processing methods, and information processing devices. Background Technology

[0002] Patent document 1 proposes an anomaly detection device in which the encoder compresses the input data using compression parameters adapted to normal data, and detects a first anomaly in the input data based on the log probability density of the compressed data on the pre-distributed data and the ratio of the log density of the compressed data distribution to the log density of the pre-distributed data; the decoder decodes the compressed data using decoding parameters adapted to normal data, and detects a second anomaly in the input data based on the difference between the reconstructed data after decoding and the input data.

[0003] Patent Document 1: Japanese Patent Application Publication No. 2018-49355 Summary of the Invention

[0004] This disclosure provides a computer program, a method for generating a learning model, an information processing method, and an information processing apparatus that are expected to reduce the effects of noise, etc., in state detection of substrate processing using a learning model.

[0005] One embodiment of the computer program causes a computer to perform the following processes: generating a learning model through machine learning, the learning model taking sensor data including multiple measurements from multiple sensors involved in substrate processing as input, and outputting corresponding sensor data including multiple measurements; causing the computer to perform the following processes: acquiring the sensor data involved in the substrate processing; inputting the acquired sensor data into the learning model, and acquiring the sensor data output by the learning model; calculating the individual error of each measurement between the sensor data input to the learning model and the sensor data output by the learning model; weighting the calculated individual errors respectively, and calculating a comprehensive error; updating the internal parameters of the learning model based on the calculated comprehensive error; and adjusting the weights attached to the individual errors according to the convergence status of the individual errors.

[0006] According to this disclosure, it is expected that the effects of noise and the like can be reduced in state detection of substrate processing using a learning model. Attached Figure Description

[0007] Figure 1 This is a schematic diagram used to illustrate the general outline of the information processing system involved in this embodiment.

[0008] Figure 2 This is a block diagram illustrating one configuration example of the information processing apparatus according to this embodiment.

[0009] Figure 3This is a schematic diagram illustrating one configuration example of the learning model involved in this embodiment.

[0010] Figure 4 This is a flowchart illustrating an example of the learning model generation process performed by the information processing apparatus according to this embodiment.

[0011] Figure 5 This is a flowchart illustrating an example of the anomaly detection process performed by the information processing apparatus according to this embodiment.

[0012] Figure 6 This is a schematic diagram showing an example of anomaly detection results.

[0013] Figure 7 This is a schematic diagram illustrating the data expansion processing performed by the information processing apparatus according to this embodiment.

[0014] Figure 8 This is a schematic diagram illustrating the preprocessing performed by the information processing apparatus involved in this embodiment. Detailed Implementation

[0015] Hereinafter, specific examples of the information processing system according to embodiments of the present disclosure will be described with reference to the accompanying drawings. Furthermore, the present disclosure is not limited to these examples, but is intended to include all modifications within the meaning and scope equivalent to the claims, as indicated by the claims.

[0016] Figure 1 This is a schematic diagram illustrating the general 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 apparatus 1 and a substrate processing apparatus 3. The substrate processing apparatus 3 is an apparatus for performing various processes on a substrate (wafer), such as CVD (chemical vapor deposition), sputtering, or etching. In this embodiment, the substrate processing apparatus 3 includes, for example, multiple sensors 4, such as a temperature sensor for measuring the temperature inside the chamber containing the substrate to be processed, and a pressure sensor for measuring pressure. During substrate processing such as etching, the substrate processing apparatus 3 repeatedly acquires (samples) data such as temperature and pressure measured by these multiple sensors 4 at predetermined intervals, and summarizes the data acquired from the multiple sensors 4 as sensor data and sends it to the information processing apparatus 1. Alternatively, the substrate processing apparatus 3 may store the data acquired from the multiple sensors 4 during substrate processing in a memory or the like, and summarize and send the stored data to the information processing apparatus 1 after the substrate processing is completed. Furthermore, the information exchange between the substrate processing apparatus 3 and the information processing apparatus 1 is not limited to wired or wireless communication; for example, it may also be performed via a recording medium such as a memory card or optical disc.

[0017] The information processing apparatus 1 is an apparatus for detecting substrate processing-related anomalies and the like performed in the substrate processing apparatus 3 based on sensor data obtained from multiple sensors 4 provided in the substrate processing apparatus 3. Multiple substrate processing apparatuses 3 may be the targets for anomaly detection. The information processing apparatus 1 according to this embodiment collects sensor data from the substrate processing apparatus 3 and records it in a database, and generates a learning model 2 for detecting anomalies by performing machine learning using the recorded sensor data.

[0018] The learning model 2 involved in this embodiment is a learning model that employs a so-called autoencoder structure. An autoencoder is a learning model that combines an encoder and a decoder. The encoder compresses and transforms the input data into feature data with reduced dimensionality, and the decoder restores the feature data to its original dimensionality. Machine learning using the autoencoder is performed through a so-called unsupervised machine learning method, which does not require teacher data with positive solution values ​​for the input data. The information processing device 1 can perform machine learning by updating the internal parameters of the autoencoder learning model 2 to ensure consistency between the input and output data.

[0019] Furthermore, when performing anomaly detection using an autoencoder, the machine learning of the autoencoder uses normal-time data and does not require anomaly-time data. An autoencoder that has undergone machine learning using normal-time data will output normal-time data when input with normal-time data, but will not output anomaly-time data even when input with anomaly-time data. The information processing device 1 inputs normal or anomaly-unknown data into the learning model 2 of the autoencoder, obtains the data output by the learning model 2, and can determine that the input data is anomaly when the difference between the input data and the output data exceeds a predetermined threshold. Moreover, the learning method of the autoencoder's learning model and the anomaly detection method using the autoencoder's learning model are prior art, and therefore detailed descriptions are omitted.

[0020] The information processing apparatus 1 of this embodiment records sensor data obtained from the substrate processing apparatus 3 into a database, and performs machine learning on the learning model 2 of the autoencoder using the sensor data during normal substrate processing. Then, the information processing apparatus 1 uses the machined learning model 2 to detect anomalies in the substrate processing performed by the substrate processing apparatus 3. The information processing apparatus 1 acquires sensor data measured by multiple sensors 4 accompanying substrate processing from the substrate processing apparatus 3, inputs the acquired sensor data into the learning model 2, and calculates the difference between the data output by the learning model 2 and the input sensor data. When the calculated difference exceeds a predetermined threshold, the information processing apparatus 1 determines that there is an anomaly in the substrate processing, and notifies users such as the administrator of this information processing system of the anomaly.

[0021] In such an information processing system that uses an autoencoder to learn model 2 for anomaly detection, the electrical signals output by the multiple sensors 4 of the substrate processing device 3 as measurement results such as temperature or pressure sometimes contain noise unrelated to substrate processing. In this case, the information processing device 1 may falsely detect the noise contained in the acquired sensor data as an anomaly. In the information processing system of this embodiment, machine learning of model 2 is performed to suppress false detections caused by noise contained in the sensor data.

[0022] <Device Structure>

[0023] Figure 2 This is a block diagram illustrating one configuration example of the information processing apparatus 1 according to this embodiment. The information processing apparatus 1 according to this embodiment can be implemented, for example, by installing a specified application program on a general-purpose information processing device such as a personal computer or server computer. The information processing apparatus 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. Furthermore, in this embodiment, processing is described as being performed by a single information processing apparatus 1, but processing can also be performed by multiple devices distributed among the information processing apparatus 1.

[0024] The processing unit 11 is constructed using computing devices such as CPU (Central Processing Unit), MPU (Micro-Processing Unit), GPU (Graphics Processing Unit) or quantum processor, ROM (Read Only Memory), and RAM (Random Access Memory). The processing unit 11 reads and executes the program 12a stored in the storage unit 12, thereby performing various processes such as generating a learning model 2 based on sensor data obtained from the sensor 4 of the substrate processing device 3, and detecting anomalies in substrate processing using the generated learning model 2.

[0025] The storage unit 12 is constructed using a high-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 required for processing by the processing unit 11. In this embodiment, the storage unit 12 stores program 12a executed by the processing unit 11. Furthermore, the storage unit 12 includes a sensor data storage unit 12b that stores sensor data acquired from the substrate processing device 3, and a model information storage unit 12c that stores information related to the learning model 2 generated by the information processing device 1.

[0026] In this embodiment, the program (computer program, program product) 12a is provided in the form of 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, for example, during the manufacturing stage of the information processing device 1. In addition, for example, the program 12a may be obtained by the information processing device 1 from content distributed by a remote server device or the like via communication. For example, the program 12a may also 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 either by distribution via a network or by recording on the recording medium 99.

[0027] The sensor data storage unit 12b stores sensor data acquired from the substrate processing apparatus 3, along with information such as the date and time of acquisition, identification information of the substrate processing apparatus 3, and identification information of the substrate processed in the substrate processing apparatus 3. The substrate processing apparatus 3 includes multiple sensors 4, and sensor data is obtained by repeatedly sampling the signals output by the multiple sensors 4 at predetermined intervals during substrate processing. For example, if the substrate processing apparatus 3 has M sensors and samples are taken N times during substrate processing, the sensor data acquired by the information processing apparatus 1 contains M×N values. The sensor data storage unit 12b stores the date, time, and identification information in correspondence with the sensor data containing the M×N values.

[0028] The model information storage unit 12c stores information related to the learning model that has undergone machine learning. This information may include, for example, information representing the structure of the learning model and the values ​​of internal parameters determined through machine learning. In this embodiment, the model information storage unit 12c stores information related to the learning model 2 of the autoencoder used for anomaly detection in the substrate processing apparatus 3. The learning model 2 in this embodiment is, for example, a learning model obtained by performing machine learning on sensor data containing the aforementioned M×N values ​​as input and outputting predicted values ​​of normal sensor data containing that sensor data as M×N data outputs. The information processing apparatus 1 generates the learning model 2 by performing machine learning on normal sensor data acquired from the substrate processing apparatus 3 and stores information related to the generated learning model 2 in the model information storage unit 12c.

[0029] Furthermore, in this embodiment, the information processing device 1 is configured to store information related to the learning model and perform model-based processing, but this is not a limitation. The information related to the learning model may also be stored in a device different from the information processing device 1, and the model-based processing may be performed by that device, from which the information processing device 1 obtains the processing results. Additionally, the machine learning processing of the learning model can be performed either in the information processing device 1 or in a device different from the information processing device 1.

[0030] The communication unit 13 is connected to the substrate processing device 3, for example, via a communication line, and performs data transmission and reception with the substrate processing device 3. In this embodiment, the communication unit 13 receives sensor data sent from the substrate processing device 3 and sends it to the processing unit 11. In addition, the communication unit 13 sends data such as control commands from the processing unit 11 to the substrate processing device 3.

[0031] The display unit 14 is constructed using a liquid crystal display or the like, and displays various images and text based on the processing of the processing unit 11. The display unit 14 displays, for example, various information such as sensor data acquired from the substrate processing device 3, information related to the generated learning model 2, and the operating status of the controlled substrate processing device 3.

[0032] The operation unit 15 accepts user operations and notifies the processing unit 11 of the accepted operations. For example, the operation unit 15 accepts user operations via input devices such as mechanical buttons or a touch panel provided on the surface of the display unit 14. Alternatively, the operation unit 15 may also be an input device such as a mouse and keyboard, and these input devices may be detachable from the information processing device 1.

[0033] Alternatively, the storage unit 12 may be an external storage device connected to the information processing device 1. Furthermore, the information processing device 1 may be a multi-computer system comprising multiple computers, or a virtual machine constructed using software. Additionally, the information processing device 1 is not limited to the above configuration; for example, it may not include a display unit 14 and an operation unit 15.

[0034] Furthermore, in the information processing apparatus 1 according to this embodiment, the processing unit 11 reads and executes the program 12a stored in the storage unit 12, thereby implementing the sensor data acquisition unit 11a, the learning processing unit 11b, the anomaly detection unit 11c, and the display processing unit 11d as software functional units in the processing unit 11. In addition, in this figure, only functional units related to the processing of the substrate processing apparatus 3, such as sensor data collection, learning model 2 generation, and anomaly detection of the substrate processing apparatus 3, are shown as functional units of the processing unit 11; functional units related to other processing are omitted from the figure.

[0035] The sensor data acquisition unit 11a communicates with the substrate processing apparatus 3 via the communication unit 13 to acquire sensor data, which includes multiple measured values ​​such as temperature and pressure measured by multiple sensors 4 during substrate processing by the substrate processing apparatus 3. During substrate processing, the substrate processing apparatus 3 repeatedly samples the signals output by the multiple sensors 4 at a predetermined period, stores the sensor data in a memory, and, for example, sends the stored sensor data to the information processing apparatus 1 at an appropriate time after substrate processing is completed. The sensor data acquisition unit 11a of the information processing apparatus 1 acquires the sensor data sent from the substrate processing apparatus 3 after substrate processing is completed, adds information such as the acquisition date and time and the identification information of the substrate processing apparatus 3 to the acquired sensor data, and stores it in the sensor data storage unit 12b.

[0036] Alternatively, the sensor data acquisition unit 11a may communicate with the substrate processing apparatus 3 at predetermined intervals during substrate processing to repeatedly acquire sensor data. Thus, the sensor data acquisition unit 11a can acquire sensor data in real time as the substrate processing apparatus 3 performs substrate processing. Furthermore, the period and timing at which the sensor data acquisition unit 11a acquires sensor data from the substrate processing apparatus 3 can be appropriately specified by the designer of this information processing system.

[0037] The learning processing unit 11b uses the sensor data stored in the sensor data storage unit 12b to perform machine learning processing to generate the learning model 2. As described above, the learning model 2 is an autoencoder structure, which, for example, accepts sensor data containing M×N values ​​as input and outputs data containing M×N values. The learning processing unit 11b uses the sensor data stored in the sensor data storage unit 12b, specifically the sensor data during normal substrate processing, to update the internal parameters of the learning model 2 so that the sensor data input to the learning model 2 matches the data output by the learning model 2. Thus, the learning processing unit 11b generates a learning model 2 that predicts normal sensor data based on the sensor data input. The learning processing unit 11b stores the internal parameters and other information of the learning model 2 obtained through machine learning in the model information storage unit 12c.

[0038] Furthermore, in this embodiment, among the multiple sensors 4 included in the substrate processing apparatus 3, the sensors 4 that are difficult to predict in the learning model 2 are those whose sensor data contains noise unrelated to substrate processing and are considered to have little impact on subsequent anomaly detection, etc. During the machine learning process of generating the learning model 2, the learning processing unit 11b calculates the prediction error for each sensor 4, reduces the learning importance of sensors 4 whose errors are difficult to converge, thereby advancing the machine learning process and preventing the learning model 2 from outputting noisy sensor data as correct data. Additionally, the learning processing unit 11b stores information about which sensor 4 is determined to be difficult to predict and whose learning importance has been reduced, along with information such as the internal parameters of the learning model 2, in the model information storage unit 12c.

[0039] The anomaly detection unit 11c uses a learning model 2 pre-generated through machine learning to detect anomalies related to substrate processing based on sensor data acquired from the substrate processing apparatus 3. The anomaly detection unit 11c reconstructs and uses the learning model 2 based on information stored in the model information storage unit 12c. For example, the anomaly detection unit 11c acquires sensor data measured by multiple sensors 4 during or after substrate processing in the substrate processing apparatus 3. The anomaly detection unit 11c inputs the sensor data acquired from the substrate processing apparatus 3 into the learning model 2 and acquires data output by the learning model 2 in response. Based on the difference between the data input to the learning model 2 and the data output by the learning model 2, the anomaly detection unit 11c determines whether there is an anomaly related to substrate processing.

[0040] Furthermore, the anomaly detection unit 11c can individually detect anomalies for each of the multiple sensors 4 included in the substrate processing apparatus 3. For example, the anomaly detection unit 11c calculates errors such as mean square error for each sensor 4 based on the input data and output data of the learning model 2. For a sensor 4 whose calculated error exceeds a threshold, it can determine that an anomaly has occurred in the substrate processing event measured by that sensor 4. In addition, at this time, the anomaly detection unit 11c can exclude sensors 4 that are judged to be difficult to predict in the machine learning of the learning model 2 and thus have reduced learning importance from being considered as anomaly detection objects, without performing error calculation or comparison with the threshold.

[0041] The display processing unit 11d performs processing to display various texts and images on the display unit 14. In this embodiment, the display processing unit 11d, for example, graphically displays sensor data acquired from the substrate processing apparatus 3 on the display unit 14. Furthermore, the display processing unit 11d displays on the display unit 14 the progress of the machine learning processing performed by the learning processing unit 11b, for example, a graph showing the change in the prediction error of the learning model 2. Additionally, the display processing unit 11d displays on the display unit 14 information related to the results of anomaly detection in the substrate processing using the learning model 2, such as whether there are any anomalies in the substrate processing, and which sensor 4 detected the anomaly if one exists. Furthermore, the display processing unit 11d can display various other information on the display unit 14 besides the information described above.

[0042] <Learning Model Generation Processing>

[0043] Figure 3 This is a schematic diagram illustrating one configuration example of the learning model 2 according to this embodiment. The learning model 2 according to this embodiment is a learning model with an autoencoder structure having an encoder and decoder (not shown). The encoder and decoder of the learning model 2 may, for example, employ a neural network structure. The encoder of the learning model 2 transforms the input data into features with fewer dimensions, and the decoder inversely transforms these features back into data of the original dimensions. The learning model 2 performs machine learning by accepting sensor data measured by multiple sensors 4 provided by the substrate processing device 3 as input and outputting predicted values ​​of normal sensor data for the input sensor data.

[0044] The information processing device 1 acquires sensor data during substrate processing by the substrate processing device 3, and stores the acquired normal sensor data in the sensor data storage unit 12b. Information indicating whether the substrate processing was performed normally is added to the sensor data stored in the sensor data storage unit 12b, for example, by a user who has confirmed the results of the substrate processing. In this embodiment, the user determines whether the sensor data is normal, but this is not a limitation. For example, the information processing device 1 may measure the substrate processing result using a measuring instrument and determine whether the measured value is within a specified range. Based on this determination, the information processing device 1 may also add normal or abnormal information to the sensor data and store it in the sensor data storage unit 12b. Based on this information, the information processing device 1 can extract normal sensor data from the sensor data stored in the sensor data storage unit 12b. The information processing device 1 uses the sensor data stored in the sensor data storage unit 12b that indicates normal substrate processing to perform machine learning processing for generating the learning model 2.

[0045] In this figure, the sensor data input and output to the learning model 2 are represented as waveforms. However, in practice, for example, if the substrate processing apparatus 3 has M sensors and N samples are taken during substrate processing, the input and output data to the learning model 2 can both be set as M×N matrices. The information processing apparatus 1 reads the sensor data stored in the sensor data storage unit 12b, and preprocesses the read sensor data as needed, such as by normalization or standardization, to obtain data for machine learning of the learning model 2.

[0046] The information processing device 1 inputs learning sensor data read from the sensor data storage unit 12b into the learning model 2, whose internal parameters have been pre-set with appropriate initial values, and obtains the data output by the learning model 2. As described above, the sensor data input to the learning model 2 is M×N matrix data, and the data output by the learning model 2 is also M×N matrix data. Based on the input data to the learning model 2 and the output data of the learning model 2, the information processing device 1 in this embodiment calculates the squared error of each component of the M×N matrix as the individual error L using the following formula (1). t current Through this calculation, information processing device 1 obtains M×N individual errors. Furthermore, in equation (1), x... t input It is a component of the matrix of input data, x t pred It is a component of the matrix corresponding to the output data. In addition, in equation (1) and the following equations, the superscript t represents the number of repetitions in machine learning, that is, the so-called epoch number.

[0047] [Number 1]

[0048]

[0049] Next, the information processing device 1 addresses individual errors L. t current The exponential smoothed moving average error L is calculated based on the following equation (2). t ema Through this calculation, information processing device 1 obtains M×N exponentially smoothed moving average errors L. t emaFurthermore, in equation (2) and the following equations, the coefficient α is a hyperparameter specified by the designer of the information processing system involved in this embodiment. The information processing device 1 calculates the current exponential smoothed moving average error based on the previous exponential smoothed moving average error in the repetition of machine learning and the current individual error. For this purpose, the information processing device 1 stores the calculation result of the current exponential smoothed moving average error in the storage unit 12 at least until the next exponential smoothed moving average error is calculated.

[0050] [Number 2]

[0051]

[0052] Next, the information processing device 1 calculates the machine learning weights *coef.* for each component of the M×N matrix of sensor data based on the following equation (3). Through this operation, the information processing device 1 obtains M×N weights *coef.*. Furthermore, in equation (3), the coefficient β is a hyperparameter specified by the designer of the information processing system according to this embodiment. For components of sensor 4 whose errors are difficult to converge through repeated machine learning, the value of the weight *coef.* decreases. The information processing device 1 can add weights in such a way that the smaller the individual error at a certain time point is compared to previous individual errors, the greater the weight added to that individual error.

[0053] [Number 3]

[0054]

[0055] Next, the information processing device 1 uses the M×N individual errors L calculated based on equations (1) to (3) above. t current And its corresponding M×N weights, the comprehensive error L is calculated based on the following equation (4). t Overall error L t It is to divide M×N individual errors L t current Multiply by the corresponding weights, calculate the sum of the M×N values ​​obtained, and thus obtain a numerical value.

[0056] [Number 4]

[0057]

[0058] Information processing device 1 uses the calculated comprehensive error L t With this comprehensive error L tThe internal parameters of learning model 2 are updated in a smaller manner, thereby performing machine learning on learning model 2. Information processing device 1, for example, can update the internal parameters of learning model 2 using the backpropagation method. The update processing of the internal parameters of learning model 2 based on the backpropagation method is prior art, so detailed explanation is omitted. The comprehensive error L calculated based on equations (1) to (4) above... t The weight of the value associated with sensor 4, such as sensor 4 with high noise and whose error is difficult to converge, becomes smaller, thus reducing the importance or influence of such sensor 4 in machine learning. This is achieved by performing a comprehensive error L... t The machine learning model 2 does not learn its features for a sensor 4 that is noisy and difficult to predict, but learns its features for a sensor 4 that is quiet and easy to predict.

[0059] Figure 4 This is a flowchart illustrating an example of the learning model generation processing steps performed by the information processing apparatus 1 according to this embodiment. The processing unit 11 of the information processing apparatus 1 according to this embodiment initializes the value of the variable t used in this processing to 1 (step S1). Furthermore, the variable t is a variable used to count the number of machine learning repetitions, and is implemented, for example, using a register within the processing unit 11 or a storage area of ​​the storage unit 12. The learning processing unit 11b of the processing unit 11 acquires sensor data measured by the sensor 4 during normal substrate processing from the sensor data stored in the sensor data storage unit 12b of the storage unit 12, and uses this as sensor data for machine learning (step S2).

[0060] The learning processing unit 11b inputs the sensor data acquired in step S2 into the learning model 2 (step S3). The learning processing unit 11b acquires the data output by the learning model 2 in response to the sensor data input in step S3 (step S4). The learning processing unit 11b calculates the individual error based on the above equation (1) for each component of the sensor data input into the learning model 2 in step S3 and the output data of the learning model 2 acquired in step S4 (step S5). Based on the individual error calculated in step S5 and the previous exponential smoothed moving average error, the learning processing unit 11b calculates the weight of each component of the sensor data based on the above equations (2) and (3) (step S6). Based on the individual error calculated in step S5 and the weight calculated in step S6, the learning processing unit 11b calculates the comprehensive error based on the above equation (4) (step S7). Based on the comprehensive error calculated in step S7, the learning processing unit 11b updates the internal parameters of the learning model 2 in a way that reduces the comprehensive error using methods such as the backpropagation method (step S8).

[0061] The learning processing unit 11b determines whether to end the machine learning process (step S9) based on conditions such as the number of repetitions of machine learning reaching a predetermined number, or the overall error calculated in step S7 being lower than a predetermined target value. If the machine learning process is not to be ended (S9: No), the learning processing unit 11b increments the variable t by 1 (step S10), returning the process to step S3. If the machine learning process is ended (S9: Yes), the learning processing unit 11b stores the internal parameters of the final learning model 2, as well as the weights of each component of the sensor data obtained during the machine learning process, in the model information storage unit 12c (step S11), and ends the process. Furthermore, the weights of each component stored in step S11 can be representative values ​​such as the minimum or average value among the weights obtained through multiple repetitions of machine learning.

[0062] <Anomaly Detection Processing> The information processing apparatus 1 according to this embodiment uses the learning model 2 generated by the above-described machine learning to process anomalies related to the substrate processing of the substrate processing apparatus 3. For example, during or after substrate processing by the substrate processing apparatus 3, the information processing apparatus 1 acquires sensor data including various measurement values ​​measured by multiple sensors 4 during the substrate processing process. The information processing apparatus 1 inputs the acquired sensor data (or data obtained by appropriately preprocessing the sensor data) into the learned learning model 2, and acquires data output by the learning model 2 in response.

[0063] The information processing device 1 compares the data input to the learning model 2 with each component of the data output by the learning model 2. When the difference (error) between each component exceeds a predetermined threshold, it determines that the sensor 4 involved in the component whose difference exceeds the threshold is abnormal. The threshold used for anomaly detection can be a common value set for multiple components, or an individual value set for each component. These thresholds are predetermined by the designer of the information processing system according to this embodiment and are stored in the storage unit 12 of the information processing device 1. In addition, the difference between each component can be the difference between the two values ​​of the input and output or their absolute values, or the individual error of the above formula (1), or a value obtained by calculation other than these.

[0064] Furthermore, the information processing device 1 can also determine whether there is an anomaly by each sensor 4, instead of judging the anomaly by each component of the input and output data. For example, if sensor data is obtained by sampling N times from M sensors 4, the input and output data to the learning model 2 is an M×N matrix. When judging the anomaly by each component, the information processing device 1 calculates the difference between the M×N values ​​and compares it with a threshold. When judging the anomaly by each sensor 4, the information processing device 1 processes the N values ​​of each sensor 4 as an N-dimensional vector, calculates the similarity between the two N-dimensional vectors of the input and output, and can judge sensor 4s with similarity below a threshold as an anomaly.

[0065] Furthermore, as described above, during the machine learning process of generating the learning model 2, the information processing apparatus 1 in this embodiment sets a small weight for sensors 4 that are noisy and have difficulty converging errors, and does not learn their features. Therefore, the learned learning model 2 has low prediction accuracy for sensors 4 with small weights, and may not be able to perform anomaly detection accurately. Therefore, the information processing apparatus 1 stores the weights calculated during machine learning along with information such as the internal parameters of the learning model 2 obtained through machine learning, and excludes sensors 4 with weights below a predetermined threshold from anomaly detection. Alternatively, the information processing apparatus 1 may not store all weights obtained during machine learning, but instead, for example, store the minimum or average weight of each sensor 4. Alternatively, it may not store weights at all, but instead store identification information of sensors 4 excluded from anomaly detection based on the weights.

[0066] The information processing device 1 displays information related to the results of anomaly detection on the display unit 14. For example, the information processing device 1 displays a list of identification information of the multiple sensors 4 included in the substrate processing device 3, along with information on whether each sensor 4 has an anomaly. At this time, the information processing device 1 can also simultaneously display information about which sensor 4, other than the one being detected for anomaly detection, is associated with. Furthermore, for sensors 4 that have detected anomalies, the information processing device 1 can, for example, display more detailed information such as graphical representation of the sensor data.

[0067] Figure 5 This is a flowchart illustrating an example of the anomaly detection processing steps performed by the information processing apparatus 1 according to this embodiment. The sensor data acquisition unit 11a of the processing unit 11 of the information processing apparatus 1 according to this embodiment acquires sensor data (step S31) including the measurement values ​​of multiple sensors 4 provided by the substrate processing apparatus 3 by communicating with the substrate processing apparatus 3 via the communication unit 13.

[0068] The anomaly detection unit 11c of the processing unit 11 constructs a learning model 2 based on the information stored in the model information storage unit 12c, and inputs the sensor data obtained in step S31 into the learning model 2 (step S32). The anomaly detection unit 11c obtains the data output by the learning model 2 in response to the sensor data input in step S32 (step S33). The anomaly detection unit 11c compares the sensor data input into the learning model 2 in step S32 with the output data of the learning model 2 obtained in step S33, and calculates the error of each component of the input and output data based on the above equation (1) (step S34).

[0069] The anomaly detection unit 11c acquires the weight information of each component stored in the model information storage unit 12c during machine learning of the learning model 2. For sensors 4 containing components with weights below a predetermined threshold, they are excluded from the anomaly detection list (step S35). For sensors 4 not excluded in step S35, the anomaly detection unit 11c determines whether the error calculated in step S34 exceeds a predetermined threshold (step S36). If the error exceeds the threshold (S36: YES), the anomaly detection unit 11c notifies the sensor 4 with the error exceeding the threshold that an anomaly has been detected (step S37) and terminates the process. If the error does not exceed the threshold (S36: No), the anomaly detection unit 11c terminates the process.

[0070] Figure 6 This is a schematic diagram showing an example of anomaly detection results. The information processing apparatus 1 according to this embodiment displays information related to the sensor 4, which is the object of anomaly detection, and the degree of anomaly (error) of each sensor on the display unit 14, as the result of anomaly detection based on substrate processing of sensor data acquired from the substrate processing apparatus 3. Figure 6 In the example shown, the information processing device 1 displays the multiple sensors 4 included in the board processing device 3 in descending order of anomaly severity (from largest to smallest error) based on the errors of each sensor 4 obtained through anomaly detection processing. This allows the information related to the sensor 4, which is the object of anomaly detection, to be displayed in a correspondence with the degree of anomaly. In this example, the information processing device 1 represents the errors of each sensor 4 as a bar graph at the top of the screen, arranging the sensors 4 with the largest errors from left to right.

[0071] Furthermore, the display of the anomaly detection results of the information processing device 1 is not limited to... Figure 6The example shown can employ various display methods. The information processing device 1 can use various charts, such as line graphs or pie charts, instead of bar charts to display the degree of anomaly (error) of each sensor 4. The information processing device 1 can also display information related to each sensor 4 in tabular form instead of charts; for example, it can display the information related to the sensor 4 in a table format, arranged from highest to lowest anomaly degree. Alternatively, the information processing device 1 can also display the information related to the sensor 4 in an appropriate order, such as according to the sensor 4's identification number, instead of in anomaly degree order, and emphasize information with high anomaly degrees, thereby establishing a correspondence between the information related to the sensor 4, the object of anomaly detection, and the degree of anomaly. Emphasis can be achieved by, for example, changing the color, size, or font of the displayed text according to the degree of anomaly, or by flashing information about the sensor 4 with high anomaly degrees.

[0072] Furthermore, the information processing device 1 can display sensor 4, for which the weight of individual errors is set low during machine learning, as an object outside the display scope, or in a way that can be distinguished as an object outside the anomaly detection scope. For example, by setting differences in the display method, the information processing device 1 can display information related to sensor 4 as the object of anomaly detection and information related to sensor 4 outside the object in a way that can be distinguished. In addition, the information processing device 1 can also add an icon or other symbol indicating that the information related to sensor 4 outside the object is an anomaly detection object to the information displayed.

[0073] In this figure, the information processing device 1 displays, side-by-side, the graph of the time-series sensor data used for anomaly detection and the graph of the time-series error calculated from the sensor data for the sensor with the largest error (sensor K in the figure) at the bottom of the screen for the sensor 4 with the largest error. Furthermore, while this example displays a graph of sensor data and error for the sensor 4 with the largest error, it is not limited to this. For instance, a user could select the names of multiple sensors 4 arranged in descending order of error at the top of the screen, and the information processing device 1, upon receiving this selection, would display a graph of sensor data and error related to the selected sensor 4 at the bottom of the screen.

[0074] <Additional Feature 1: Data Expansion>

[0075] The information processing apparatus 1 uses multiple sensor data acquired from the substrate processing apparatus 3 and stored in the sensor data storage unit 12b to perform machine learning to generate a learning model 2. The more learning data used in machine learning, the more likely it is to improve the prediction accuracy of the generated learning model 2. Therefore, the information processing apparatus 1 according to this embodiment generates new sensor data that can be used for machine learning based on the sensor data acquired from the substrate processing apparatus 3, thereby enabling processing to increase the learning data, a process known as data expansion.

[0076] Figure 7 This is a schematic diagram illustrating the data expansion processing performed by the information processing apparatus 1 according to this embodiment. The information processing apparatus 1 according to this embodiment uses the time series measurement values ​​of each sensor 4 included in the sensor data previously acquired from the substrate processing apparatus 3 as raw data, and generates one or more expanded data by performing appropriate calculations on the raw data. The information processing apparatus 1 determines the range of the expanded data as the range after expanding the raw data by a predetermined amount in each direction (vertical axis (measurement value) and horizontal axis (time). Figure 7 The upper section shows a step waveform as an example of the original data, and the middle section shows an example of an extended range for that waveform.

[0077] In this embodiment, the information processing device 1 determines the expansion range for the original data based on information about the resolution of the sensor 4 that outputs the original data (e.g., the vertical axis is the minimum value that the sensor 4 can measure, and the horizontal axis is determined by the sampling period). For example, the information processing device 1 can use the range in which the original data can be shifted by positive and negative resolutions on the vertical and horizontal axes, respectively, as the expansion range.

[0078] Information processing device 1 generates one or more extended data by appropriately performing operations such as shifting, scaling up, or reducing the original data to converge to a determined range of extension. Figure 7 The lower section shows waveforms of three expanded data points generated in a manner that converges to the expanded range. The expanded data shown on the left is the original data scaled down along the vertical axis. The expanded data shown in the center is the original data scaled up along the vertical axis. The expanded data shown on the right is the original data shifted horizontally along the vertical axis. The information processing device 1, for example, randomly selects various pre-defined operations such as shifting, scaling up, or scaling down, and performs the selected operations on the original data, thereby expanding the data within the expanded range.

[0079] The information processing device 1 can generate one or more expanded sensor data from a single sensor data (normal sensor data) stored in the sensor data storage unit 12b, and increase the amount of sensor data available for machine learning of the learning model 2 by more than double. By using more learning data for machine learning, the information processing device 1 can expect to generate a learning model 2 with higher prediction accuracy. Furthermore, by determining the expansion range of the data based on the resolution of the sensor 4, it is expected to prevent false detections of anomalies caused by the resolution of the sensor 4.

[0080] <Additional Feature 2: Preprocessing>

[0081] The information processing device 1 can also perform appropriate preprocessing on the sensor data acquired from the substrate processing device 3, and input the preprocessed sensor data into the learning model 2. Preprocessing can, for example, employ standardization or normalization. Standardization transforms multiple values ​​based on their mean and standard deviation to a mean of 0 and a variance of 1. Normalization transforms multiple values ​​based on their minimum and maximum values ​​to a range of 0 to 1. Furthermore, standardization and normalization are prior art, and therefore detailed explanations are omitted.

[0082] Figure 8 This is a schematic diagram illustrating the preprocessing performed by the information processing apparatus 1 according to this embodiment. In this embodiment, the information processing apparatus 1 performs preprocessing using the summary values ​​of sensor data, and inputs the preprocessed sensor data into the learning model 2. When standardization is performed as preprocessing, the summary values ​​of the sensor data are the mean and standard deviation; when normalization is performed as preprocessing, they are the minimum and maximum values. Furthermore, the information processing apparatus 1 may also perform preprocessing different from standardization and normalization, in which case the summary values ​​of the sensor data can be various values ​​required for the preprocessing. The following describes the case where standardization is performed as preprocessing.

[0083] Information processing device 1 can obtain sensor data containing M×N values ​​by sampling N times for M sensors 4. By calculating the average and standard deviation of the N values ​​for each sensor 4, information processing device 1 can obtain two summary values, thus obtaining 2×M summary values ​​from one sensor data.

[0084] The information processing apparatus 1 according to this embodiment has a prediction model 5 for summary values ​​generated in advance through machine learning (i.e., information related to the prediction model 5 is stored in the model information storage unit 12c). The prediction model 5 for summary values ​​is a learning model obtained by machine learning in a manner that accepts summary values ​​as input and predicts normal summary values ​​for given summary values. In this embodiment, the information processing apparatus 1 uses a model that has undergone dimensionality reduction through principal component analysis as the prediction model 5 for summary values, transforms 2×M summary values ​​into principal component information of a number less than the given number, and then inversely transforms the transformed principal component information again into 2×M summary values. As a result, the information processing apparatus 1 can obtain a predicted value for a normal summary value for sensor data based on the summary value of sensor data obtained from the substrate processing apparatus 3.

[0085] Information processing device 1 uses the predicted values ​​of 2×M summary values ​​obtained from prediction model 5 to preprocess the sensor data containing the original M×N values. In this example, the sensor data contains values ​​associated with M sensors 4, and each sensor 4 contains N time series values. Information processing device 1 standardizes the N time series values ​​using the predicted mean and standard deviation of each sensor 4. Information processing device 1 can perform the same standardization for each of the M sensors 4, thereby standardizing all the M×N values ​​contained in the sensor data.

[0086] Before performing machine learning on generative learning model 2, information processing device 1 generates a prediction model 5 for preprocessing. Information processing device 1 acquires normal sensor data for machine learning of model 2, and calculates a summary value for each sensor 4 based on the acquired sensor data. Based on the calculated multiple summary values, information processing device 1 performs dimensionality reduction using principal component analysis or similar methods, and generates a prediction model 5 that restores the dimensionality-reduced information to its original dimensions. Furthermore, principal component analysis is prior art, so detailed explanation is omitted. Additionally, prediction model 5 is not limited to being based on principal component analysis; any learning model capable of reducing the dimensionality of input data and restoring it to its original dimensions can be used. Prediction model 5 can be any learning model as long as it is capable of predicting normal summary values ​​for the input summary values.

[0087] Furthermore, when the information processing device 1 generates the learning model 2 through machine learning and when it uses the learned learning model 2 to perform anomaly detection, it performs preprocessing using the pre-generated prediction model 5. The prediction model 5 used in either case is the same (the internal parameters determined by machine learning are the same). The information processing device 1 pre-generates the prediction model 5 and stores information such as the internal parameters of the prediction model 5 in the model information storage unit 12c.

[0088] The information processing apparatus 1 of this embodiment uses the prediction model 5 to predict normal summary values ​​from the summary values ​​of sensor data, and performs preprocessing such as standardization or normalization on the predicted normal summary values. As a result, the information processing apparatus 1 reduces or invalidates the changes in values ​​caused by factors such as individual differences of the substrate processing apparatus 3 or changes over time through preprocessing, thereby expecting to improve the prediction accuracy of the learning model 2.

[0089] <Summary>

[0090] In the information processing system according to this embodiment of the above structure, the information processing device 1 generates a learning model 2 through machine learning. This learning model 2 accepts sensor data containing multiple measured values ​​from multiple sensors 4 included in the substrate processing device 3 as input, and outputs corresponding sensor data containing multiple normal measured values. The information processing device 1 obtains normal sensor data related to substrate processing from the sensor data stored in the sensor data storage unit 12b, inputs the obtained sensor data into the learning model 2, and obtains the sensor data output by the learning model 2. The information processing device 1 calculates the individual error of each measured value contained in the sensor data input to the learning model 2 and the sensor data output by the learning model 2, and calculates a comprehensive error by weighting the calculated individual errors. The information processing device 1 performs machine learning by updating the internal parameters of the learning model 2 based on the calculated comprehensive error, thereby generating the learning model 2. Furthermore, the information processing device 1 adjusts the weights attached to the individual errors according to the convergence status of the individual errors. Therefore, the information processing system according to this embodiment can generate a learning model 2 by, for example, performing machine learning that takes into account sensor data with more noise and sensor data with less noise, and it is expected to reduce the influence of noise in the state detection of substrate processing using the generated learning model 2.

[0091] Furthermore, in the information processing system according to this embodiment, the information processing device 1 calculates the weights based on equations (1) to (3) above. Therefore, the information processing device 1 adjusts the weights attached to each individual error in such a way that the smaller the individual error at a certain time point is compared to previous individual errors, the greater the weight attached to that individual error. That is, the information processing device 1 adjusts the weights attached to each individual error in such a way that the weight attached to individual errors that are difficult to converge decreases. Therefore, for the information processing system according to this embodiment, it is expected that, for example, for the sensor 4 with high noise and difficult-to-converge individual errors, the importance of machine learning can be reduced to generate the learning model 2.

[0092] Furthermore, in the information processing system of this embodiment, the learning model 2 generated by the information processing device 1 has the structure of an autoencoder. Therefore, the information processing system of this embodiment can perform machine learning to generate the learning model 2 by using learning data without correct answer labels, i.e., so-called unsupervised machine learning. By using normal sensor data for machine learning with the information processing device 1, it is expected to generate a learning model 2 that predicts normal sensor data.

[0093] Furthermore, in the information processing system according to this embodiment, the information processing device 1 acquires sensor data from the substrate processing device 3, including multiple measurement values ​​measured by multiple sensors 4 involved in substrate processing. The information processing device 1 inputs the acquired sensor data into the learning model 2 pre-generated through the aforementioned machine learning, and acquires the sensor data output by the learning model 2. Based on the comparison between the sensor data input to the learning model 2 and the sensor data output by the learning model 2, the information processing device 1 detects states such as anomalies related to substrate processing. Therefore, for the information processing system according to this embodiment, it is expected that the influence of noise and the like can be reduced in the state detection of substrate processing using the learning model 2.

[0094] Furthermore, in the information processing system of this embodiment, the information processing device 1 stores information related to the weight of each measured value of the sensor data calculated during the machine learning of the learning model 2 in the model information storage unit 12c. During state detection, the information processing device 1 determines which sensors are excluded from state detection based on the stored weights. Thus, the information processing system of this embodiment can exclude sensors 4, whose learning importance has decreased due to individual errors that are difficult to converge during machine learning, from state detection, and can be expected to prevent false detection of the state of sensors 4 with low prediction accuracy based on the learning model 2.

[0095] The embodiments disclosed herein should be considered illustrative in all respects and not restrictive. The scope of this disclosure is not defined by the foregoing meaning, but by the claims, and is intended to include all modifications within the meaning and scope equivalent to the claims.

[0096] The items 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 all possible combinations, regardless of their referencing form. Moreover, the claims use a form that refers to two or more other claims (multiple claim form), but are not limited to this. A multiple claim form that refers to at least one multiple claim (multiple-referencing-multiple-claims) can also be used.

[0097] Explanation of reference numerals in the attached figures

[0098] 1…Information processing device (computer); 2…Learning model; 3…Substrate processing device; 4…Sensor; 5…Prediction model; 11…Processing unit; 11a…Sensor data acquisition unit; 11b…Learning processing unit; 11c…Anomaly detection unit; 11d…Display processing unit; 12…Storage unit; 12a…Program (computer program); 12b…Sensor data storage unit; 12c…Model information storage unit; 13…Communication unit; 14…Display unit; 15…Operation unit.

Claims

1. A computer program that causes a computer to perform processing to generate a learning model through machine learning, the learning model taking sensor data including multiple measurements from multiple sensors involved in substrate processing as input, and outputting corresponding sensor data including multiple measurements, the computer program causing the computer to perform the following processing: Obtain the sensor data involved in the above substrate processing; The acquired sensor data is input into the learning model to obtain the sensor data output by the learning model. Calculate the individual error between each measurement of the sensor data input to the learning model and the sensor data output by the learning model. The calculated individual errors are weighted separately to calculate the comprehensive error; Based on the calculated comprehensive error, the internal parameters of the learning model are updated; and Based on the convergence status of the individual errors mentioned above, adjust the weights attached to those individual errors.

2. The computer program according to claim 1, wherein, The weights attached to individual errors are adjusted in such a way that the smaller the individual error at a given time point is compared to the individual errors at time points prior to that given time point, the greater the weight of the individual error at that given time point.

3. The computer program according to claim 1, wherein, The learning model described above is an autoencoder.

4. The computer program according to claim 1, wherein, Based on the acquired sensor data, extended sensor data is generated that allows the measured values ​​of the sensor data to vary within a specified range. The generated extended sensor data is used as sensor data for machine learning of the learning model described above.

5. The computer program according to claim 1, wherein, Calculate the summary value of the acquired sensor data. Based on the calculated summary values, a predictive model for predicting the summary values ​​of sensor data is generated. The summary values ​​predicted by the generated prediction model are used to preprocess the sensor data. The sensor data, after the above preprocessing, is input into the learning model.

6. A computer program that causes a computer to perform the following processes: Acquire sensor data, which includes multiple measurements from multiple sensors, involved in substrate processing; Input the acquired sensor data into the learning model, and obtain the sensor data output by the learning model, wherein, The aforementioned learning model is obtained through machine learning by accepting sensor data as input and outputting corresponding sensor data; and Based on the comparison between the sensor data input to the learning model and the sensor data output by the learning model, the state involved in the substrate processing is detected. The machine learning of the above learning model is performed as follows: The sensor data is input into the learning model to obtain the sensor data output by the learning model. Calculate the individual error between each measurement of the sensor data input to the learning model and the sensor data output by the learning model. The calculated individual errors are weighted separately to calculate the comprehensive error; Based on the calculated comprehensive error, the internal parameters of the learning model are updated; and Based on the convergence status of the individual errors mentioned above, adjust the weights attached to those individual errors.

7. The computer program according to claim 6, wherein, To obtain information related to the weights of each measurement of the sensor data during the aforementioned machine learning process. Based on the weights obtained above, the sensors other than the objects to be detected are determined.

8. The computer program according to claim 7, wherein, Based on a comparison between the sensor data input to the learning model and the sensor data output by the learning model, the degree of anomaly for each sensor is calculated. Information related to one or more sensors that are used for state detection is output in a manner that corresponds to the degree of anomaly of that sensor.

9. The computer program according to claim 8, wherein, Information related to multiple sensors is output in order of its degree of anomaly.

10. The computer program according to claim 7, wherein, Information related to the sensor that is the object of state detection and information related to sensors other than the object of state detection are output in a discriminative manner.

11. The computer program according to claim 6, wherein, Calculate the summary value of the acquired sensor data. Based on the calculated summary values, a prediction model is used to predict the summary values ​​of the sensor data. This prediction model is obtained through machine learning, using the summary values ​​of the sensor data as input and predicting the summaries of the sensor data. The predicted summary values ​​are used to preprocess the acquired sensor data. The sensor data, after the above preprocessing, is input into the learning model.

12. A method for generating a learning model, wherein an information processing device generates a learning model through machine learning, the learning model accepting sensor data including multiple measurement values ​​from multiple sensors involved in substrate processing as input, and outputting corresponding sensor data including multiple measurement values. Obtain the sensor data involved in the above substrate processing; The acquired sensor data is input into the learning model to obtain the sensor data output by the learning model. Calculate the individual error between each measurement of the sensor data input to the learning model and the sensor data output by the learning model. The calculated individual errors are weighted separately to calculate the comprehensive error; Based on the calculated comprehensive error, the internal parameters of the learning model are updated; and Based on the convergence status of the individual errors mentioned above, adjust the weights attached to those individual errors.

13. An information processing method, wherein an information processing device causes a computer to perform the following processing: Acquire sensor data, which includes multiple measurements from multiple sensors, involved in substrate processing; Input the acquired sensor data into the learning model, and obtain the sensor data output by the learning model, wherein, The aforementioned learning model is obtained through machine learning by accepting sensor data as input and outputting corresponding sensor data; and Based on the comparison between the sensor data input to the learning model and the sensor data output by the learning model, the state involved in the substrate processing is detected. The machine learning of the above learning model is performed as follows: The sensor data is input into the learning model to obtain the sensor data output by the learning model. Calculate the individual error between each measurement of the sensor data input to the learning model and the sensor data output by the learning model. The calculated individual errors are weighted separately to calculate the comprehensive error; Based on the calculated comprehensive error, the internal parameters of the learning model are updated; and Based on the convergence status of the individual errors mentioned above, adjust the weights attached to those individual errors.

14. An information processing apparatus comprising a processing unit, wherein the processing unit generates a learning model by machine learning, the learning model receiving sensor data including multiple measurement values ​​measured by multiple sensors involved in substrate processing as input, and outputting corresponding sensor data including multiple measurement values. The above processing department shall execute: Obtain the sensor data involved in the above substrate processing; The acquired sensor data is input into the learning model to obtain the sensor data output by the learning model. Calculate the individual error between each measurement of the sensor data input to the learning model and the sensor data output by the learning model. The calculated individual errors are weighted separately to calculate the comprehensive error; Based on the calculated comprehensive error, the internal parameters of the learning model are updated; and Based on the convergence status of the individual errors mentioned above, adjust the weights attached to those individual errors.

15. An information processing apparatus comprising a processing unit, The above processing unit causes the computer to perform the following processes: Acquire sensor data, which includes multiple measurements from multiple sensors, involved in substrate processing; Input the acquired sensor data into the learning model, and obtain the sensor data output by the learning model, wherein, The aforementioned learning model is obtained through machine learning by accepting sensor data as input and outputting corresponding sensor data; and Based on the comparison between the sensor data input to the learning model and the sensor data output by the learning model, the state involved in the substrate processing is detected. The machine learning of the above learning model is performed as follows: The sensor data is input into the learning model to obtain the sensor data output by the learning model. Calculate the individual error between each measurement of the sensor data input to the learning model and the sensor data output by the learning model. The calculated individual errors are weighted separately to calculate the comprehensive error; Based on the calculated comprehensive error, the internal parameters of the learning model are updated; and Based on the convergence status of the individual errors mentioned above, adjust the weights attached to those individual errors.

Images