Information processing method, information processing apparatus and substrate processing system

Abstract

This invention provides an information processing method, an information processing apparatus, and a substrate processing system. The information processing method of this invention includes the following steps: acquiring time-series data from a sensor disposed on the substrate processing apparatus; learning a first learning model, i.e., when time-series data from the sensor is input based on the acquired time-series data, outputting information related to the substrate processing apparatus; inputting the time-series data from the sensor into the learned first learning model, and outputting an estimation result based on the information obtained from the first learning model.

Description

Technical Field

[0001] This invention relates to an information processing method, an information processing apparatus, and a substrate processing system. Prior Technology

[0002] In a substrate processing system, multiple edge devices are used to perform various processes in multiple chambers having multiple edge devices. [Previous Technical Documents] [Patent Literature]

[0003] [Patent Document 1] International Publication No. 2018 / 173121 [Patent Document 2] International Publication No. 2019 / 163823 Summary of the Invention

[0004] [The problem the invention aims to solve]

[0005] In view of the above, the present invention provides an information processing method, an information processing apparatus, and a substrate processing system that can efficiently create and manage models applied by various edge devices set in a substrate processing system. [Technical means to solve the problem]

[0006] One aspect of the information processing method of the present invention includes the following steps: acquiring time-series data from a sensor disposed on a substrate processing device; learning a first learning model, that is, when time-series data from the sensor is input based on the acquired time-series data, outputting information related to the substrate processing device; and inputting time-series data from the sensor into the learned first learning model, and outputting an estimation result based on the information obtained from the first learning model. [Effects of the Invention]

[0007] According to the present invention, inferential models of various edge devices set in a substrate processing system can be created and managed efficiently. Simple Explanation of the Diagram

[0008] Figure 1 is a diagram showing an example of the configuration of the substrate processing system according to the embodiment. Figure 2 is a schematic diagram showing the configuration of the drive system of the substrate processing device. Figure 3 is a cross-sectional view showing an example of the structure of a chamber. Figure 4 is a block diagram illustrating the structure of the control system of the substrate processing apparatus. Figure 5 is a block diagram showing the internal structure of the edge devices in the substrate processing apparatus. Figure 6 is a schematic diagram illustrating an example of the configuration of an observation model possessed by an edge device. Figure 7 is a schematic diagram illustrating an example of the configuration of a control model possessed by an edge device. Figure 8 is a block diagram showing the internal structure of the control equipment in the substrate processing apparatus. Figure 9 is a conceptual diagram illustrating an example of the structure of a database. Figure 10 is a block diagram showing the internal structure of the device cluster server. Figure 11 is a conceptual diagram illustrating an example of the structure of a database. Figure 12 is a flowchart showing the steps of generating the first learning model performed by the edge device. Figure 13 is a flowchart showing the sequence of processing steps performed inside the substrate processing device during the application phase. Figure 14 is a flowchart illustrating the processing steps performed between the substrate processing device and the device group server during the application phase. Figure 15 is a schematic diagram showing an example of the evaluation results. Implementation

[0009] Hereinafter, an embodiment will be described with reference to the drawings. In the description, the same symbols are used to mark the same elements or elements with the same function, and repeated descriptions are omitted.

[0010] (Implementation Method 1) Figure 1 is a diagram illustrating an example configuration of the substrate processing system according to the embodiment. The substrate processing system of this embodiment is a system for managing learning models applied by a plurality of substrate processing devices 100A-100D. The substrate processing system includes a plurality of substrate processing devices 100A-100D and a device group server 200 that collects data from the plurality of substrate processing devices 100A-100D. Each substrate processing device 100A-100D and the device group server 200 can be communicatively connected via a communication network such as a LAN (Local Area Network) or a dedicated line.

[0011] In this embodiment, the substrate processing system is configured to include four substrate processing devices 100A to 100D, but the number of devices is not limited to four. In the following description, when it is not necessary to distinguish between substrate processing devices 100A to 100D, they are also referred to as substrate processing device 100 (see Figure 2). In this embodiment, the device cluster server 200 may be a single computer or a computer system containing multiple computers or peripheral devices. Furthermore, the device cluster server 200 may be a virtual machine in which the physical entity is virtualized, or it may be the cloud.

[0012] Figure 2 is a schematic diagram showing the configuration of the drive system of the substrate processing apparatus 100. The substrate processing apparatus 100 includes a transport unit HU for moving the substrate W in and out, and a processing unit SU for performing substrate processing on the substrate W.

[0013] The transfer unit HU includes a case placement stage 10 and a transfer stage 20. The case placement stage 10 includes a case container 11. The case container 11 can hold up to 25 substrates W in a stacked state, for example.

[0014] The transfer table 20 is equipped with a substrate transfer mechanism 21 for transferring substrates W. The substrate transfer mechanism 21 has two transfer arms 21A and 21B that hold the substrates W approximately horizontally. Using the transfer arms 21A and 21B, the substrate transfer mechanism 21 removes the substrates W one by one from the cassette container 11. The substrate transfer mechanism 21 transfers the substrates W removed from the cassette container 11 to either of the loading lock chambers 25A and 25B. The loading lock chambers 25A and 25B connect the transfer unit HU to the processing unit SU.

[0015] The processing unit SU includes a transfer chamber 30 and four chambers 40A to 40D. The transfer chamber 30 has a sealable structure, for example, that is polygonal when viewed from above (hexagonal in the illustrated example). The transfer chamber 30 is connected to each of the chambers 40A to 40D via airtight gate valves. The transfer chamber 30 includes a substrate transport mechanism 31 for transporting a substrate W. The substrate transport mechanism 31 has two transport arms 31A and 31B that hold the substrate W approximately horizontal. Using the transport arms 31A and 31B, the substrate transport mechanism 31 removes the substrate W from the loading and locking chambers 25A and 25B and transports the removed substrate W to any of the chambers 40A to 40D.

[0016] With this configuration, the processing unit SU transfers the substrate W, which has been conveyed to the loading lock chambers 25A and 25B, into the chambers 40A to 40D via the transfer chamber 30, where substrate processing is performed. After performing the substrate processing, the processing unit SU removes the processed substrate W from the chambers 40A to 40D and transfers it out to the loading lock chambers 25A and 25B via the transfer chamber 30. One example of substrate processing performed in the chambers 40A to 40D is film deposition using CVD (Chemical Vapor Deposition). Substrate processing performed in the chambers 40A to 40D can also be replaced by diffusion processing, etching processing, ashing processing, sputtering processing, etc. Furthermore, in the example of Figure 2, a single-piece substrate processing apparatus 100 is shown that takes out substrates W one by one from the box container 11 for substrate processing. However, the substrate processing apparatus 100 can also be a batch substrate processing apparatus that processes multiple substrates W at the same time, and the conveying unit HU can be configured arbitrarily.

[0017] In the following description, when there is no need to distinguish between chambers 40A to 40D, they will be simply referred to as chamber 40 (see Figure 3).

[0018] Figure 3 is a cross-sectional view showing an example of the configuration of the chamber 40. The chamber 40 shown in Figure 3 is an apparatus for performing film formation processing on a substrate W, and includes a processing chamber 41 for performing substrate processing and an exhaust chamber 42 communicating with the processing chamber 41.

[0019] The processing chamber 41 has a plate-shaped top 411 and bottom 413, and a side wall portion 412 connecting the top 411 and bottom 413. The processing chamber 41 is, for example, generally cylindrical in shape. A loading / unloading outlet is formed between the side wall portion 412 and the transfer chamber 30 for moving the substrate W in and out. By opening a gate valve provided between the processing chamber 41 and the transfer chamber 30, the substrate W can be moved in and out through the loading / unloading outlet. An opening 413a is formed in the center of the bottom 413. The exhaust chamber 42 is connected to the bottom 413 of the processing chamber 41 in communication with the opening 413a.

[0020] The exhaust chamber 42 has an annular flange portion 421, a plate-shaped bottom 423, and a side wall portion 422 connecting the flange portion 421 and the bottom 423. The flange portion 421 is engaged with the bottom 413 of the processing chamber 41. An exhaust hole 424 is formed in the side wall portion 422.

[0021] The processing chamber 41 and the exhaust chamber 42 are configured to maintain a reduced gas atmosphere (vacuum state) in their internal space. O-rings as sealing members are installed at the joint of the processing chamber 41 and the exhaust chamber 42, as well as at the joint of the components constituting the processing chamber 41 and the exhaust chamber 42, to ensure the airtightness of the joint.

[0022] The chamber 40 includes an exhaust device 51 disposed outside the processing chamber 41 and the exhaust chamber 42, an exhaust pipe 52 connecting the exhaust port 424 to the exhaust device 51, and a valve 53 disposed in the middle of the exhaust pipe 52. The valve 53 can maintain the airtightness of the processing chamber 41 and the exhaust chamber 42 when closed, and can depressurize the processing chamber 41 and the exhaust chamber 42 by means of the exhaust device 51 when open. By activating the exhaust device 51, the processing chamber 41 and the exhaust chamber 42 are depressurized to the vacuum level required for their internal space.

[0023] The chamber 40 includes a base 61 disposed within the processing chamber 41, and a support member 62 supporting the base 61 within the processing chamber 41 and the exhaust chamber 42. The base 61 is a substrate mounting stage for horizontally supporting the substrate W. The base 61 has a substrate mounting surface (upper surface) for mounting the substrate W, and a lower surface on the opposite side. One end of the support member 62 is fixed to the center of the lower surface of the base 61. The other end of the support member 62 is fixed to the bottom 423 of the exhaust chamber 42.

[0024] The base 61 has a plurality of support pins that are recessed into the substrate mounting surface, but are not shown. The plurality of support pins are configured to be able to move up and down by any lifting mechanism, and at the raised position, the substrate W is transferred between the base 61 and the substrate conveying mechanism 31.

[0025] Chamber 40 includes a heater 63, a heater power supply 64, and a temperature sensor TS. The temperature measuring components of the heater 63 and the temperature sensor TS are embedded in the base 61. The heater power supply 64 is located outside the processing chamber 41 and the exhaust chamber 42. The heater 63 is connected to the heater power supply 64, for example, via wiring inside the support member 62. The heater power supply 64 supplies electrical output to the heater 63 to heat the substrate W placed on the base 61 to the desired temperature. The temperature of the base 61 is measured by the temperature sensor TS. The temperature sensor TS is constructed using known components such as thermocouples and thermistors.

[0026] The chamber 40 has a cluster head 71 disposed on the top 411 of the processing chamber 41. The cluster head 71 has a gas diffusion space 71a formed therein and a plurality of gas ejection holes 71b formed to extend from the gas diffusion space 71a to the base 61.

[0027] The chamber 40 includes: a gas inlet pipe 72 disposed in the cluster nozzle 71 on the side opposite to the plurality of gas outlet holes 71b and communicating with the gas diffusion space 71a; a gas supply source 73 disposed outside the processing chamber 41 and the exhaust chamber 42; a gas piping 74 connecting the gas inlet pipe 72 and the gas supply source 73; an MFC (mass flow controller) 75 disposed in the middle of the gas piping 74; and a valve not shown. The gas supply source 73 supplies the cluster nozzle 71 with film-forming raw material gas for film-forming treatment, cleaning gas for cleaning the processing chamber 41 and the exhaust chamber 42, and flushing gas for replacing the gas atmosphere in the processing chamber 41 and the exhaust chamber 42. These gases are supplied to the gas diffusion space 71a via the gas piping 74 and the gas inlet pipe 72, and are ejected into the processing chamber 41 from the plurality of gas outlet holes 71b.

[0028] The chamber 40 includes a high-frequency power supply 76 disposed outside the processing chamber 41 and the exhaust chamber 42, a wiring 77 connecting the cluster head 71 to the high-frequency power supply 76, and a matching device 78 disposed in the middle of the wiring 77. The high-frequency power supply 76 supplies high-frequency power to the cluster head 71 to plasmaize the film-forming raw material gas supplied to the processing chamber 41.

[0029] With the above configuration, a film-forming process can be performed on a substrate W placed on a base 61 within the chamber 40. Specifically, by transporting the substrate W to be processed into the vacuum-controlled processing chamber 41, the substrate W placed on the base 61 is heated using a heater 63, and a raw material gas is supplied to the substrate W from the nozzle 71, thereby forming a thin film on the surface of the substrate W. Furthermore, to promote the film-forming reaction, high-frequency power can be supplied to the nozzle 71 from a high-frequency power source 76. In this case, the raw material gas supplied to the processing chamber 41 via the nozzle 71 can be plasma-entrained to form a film.

[0030] In the example of Figure 3, the chamber 40 for performing film deposition on the substrate W is described, but the chamber 40 can also be a process module for performing diffusion treatment, etching treatment, ashing treatment, sputtering treatment, etc. on the substrate W.

[0031] Figure 4 is a block diagram illustrating the configuration of the control system of the substrate processing apparatus 100. The substrate processing apparatus 100 includes various sensors S1 to S3, edge devices 110 to 130 that receive data from the output of the sensors S1 to S3, actuators A1 to A3 that are controlled by the edge devices 110 to 130, and a control device 150 that controls the operation of the entire apparatus.

[0032] Sensors S1 to S3 are installed on the substrate processing device 100 and are used to measure the physical quantity to be measured in time series. Sensors S1 to S3 output time series data (hereinafter also referred to as sensor data) representing the measurement results to the back-end edge devices 110 to 130.

[0033] One example of sensor S1 is an RF (Radio Frequency) sensor. The RF sensor is located on the output side of the high-frequency power supply 76 and measures the RF power of the high-frequency power supply 76 in a time-series manner. The RF sensor can also be replaced by a sensor that measures voltage, current, capacitance, impedance, phase, load power, etc., in a time-series manner. One example of sensor S2 is a temperature sensor. The temperature sensor includes a temperature sensing element embedded in the base 61 and measures the temperature of the substrate mounting surface (i.e., the temperature of the substrate W to be processed) in a time-series manner. The temperature sensor can also be replaced by a sensor that measures electrode temperature or the internal temperature of the processing chamber 41 in a time-series manner. One example of sensor S3 is a torque sensor. The torque sensor measures the torque received by the actuator (e.g., actuator A3) mounted on the substrate transport mechanisms 21 and 31 in a time-series manner.

[0034] The sensors S1 to S3 mounted on the substrate processing apparatus 100 are not limited to the aforementioned RF sensors, temperature sensors, and torque sensors. Furthermore, the number of sensors mounted on the substrate processing apparatus 100 is not limited to three. For example, the substrate processing apparatus 100 may also include one or more sensors, such as a gas sensor, an OES (Optical Emission Spectroscopy) sensor, and a flow sensor. Here, the gas sensor is installed in the processing chamber 41 and measures the amount of a specific component of the gas filling the processing chamber 41 in a time series. The gas sensor may be a mass spectrometer, an infrared spectrometer, or a gas chromatography analyzer. The OES sensor is installed in the processing chamber 41 and measures the plasma emission intensity inside the processing chamber 41 in a time series. The flow sensor is installed in the processing chamber 41 and measures the flow rate of the gas introduced into the processing chamber 41 in a time series.

[0035] Furthermore, the sampling frequencies of sensors S1 to S3 are arbitrary and can be appropriately set for each sensor. For example, the sampling frequency of the RF sensor is 1 to 10 μsec. The sampling frequency of the temperature sensor is, for example, 100 msec. The sampling frequency of the torque sensor is, for example, 2.5 msec. The sampling frequency of the OES sensor is, for example, 10 to 100 msec. The sampling frequency of the gas sensor is, for example, 1 to 10 msec. The sampling frequency of the flow sensor is, for example, 10 msec.

[0036] Edge devices 110-130 perform the following processes: estimating the state of the substrate processing device 100 based on sensor data input from self-sensors S1-S3, and estimating control values ​​for actuators A1-A3. For example, edge devices 110-130 output the state estimation results to the control device 150 and control the operation of actuators A1-A3 based on the estimated control values. The internal structure of edge devices 110-130 and the content of the processes performed by edge device 110 are described in detail below.

[0037] Actuators A1 to A3 are the controlled objects of peripheral devices 110 to 130. In this embodiment, the actuator is not limited to mechanical components such as motors that convert electrical energy into physical momentum; all types of drive circuits, including circuits, are collectively referred to as actuators.

[0038] For example, when sensor S1 is an RF sensor, actuator A1 can be a high-frequency power supply 76. In this case, edge device 110 obtains time-series data of RF power from the RF sensor, estimates a control value for the high-frequency power supply 76 based on the obtained time-series data, and controls the operation of the high-frequency power supply 76 based on the estimated control value. Similarly, when sensor S2 is a temperature sensor TS, actuator A2 can be a heater power supply 64. In this case, edge device 120 obtains time-series data of temperature from the temperature sensor TS, estimates a control value for the heater power supply 64 based on the obtained time-series data, and controls the operation of the heater power supply 64 based on the estimated control value. Furthermore, when sensor S3 is a torque sensor, actuator A3 can be a motor provided by the substrate transport mechanisms 21 and 31. In this case, the edge device 130 obtains time-series data of the torque received by the motor drive shaft from the torque sensor, estimates control values ​​for the substrate conveying mechanisms 21 and 31 based on the obtained time-series data, and controls the operation of the substrate conveying mechanisms 21 and 31 based on the estimated control values.

[0039] In this embodiment, each edge device 110-130 is configured to connect one sensor and one actuator. However, the number of sensors and actuators connected to each edge device 110-130 is not limited to one. Multiple sensors and multiple actuators can also be connected to each edge device 110-130. Furthermore, in this embodiment, the substrate processing apparatus 100 is configured to have three edge devices 110-130. However, the number of edge devices mounted on the substrate processing apparatus 100 is not limited to three; having one or more edge devices is sufficient.

[0040] The control device 150 controls the overall operation of the substrate processing apparatus 100 based on various information input from edge devices 110-130 and various information input from the outside. The internal structure of the control device 150 and the processing performed by the control device 150 will be described in detail below.

[0041] Figure 5 is a block diagram showing the internal structure of the edge device 110 provided in the substrate processing apparatus 100. The edge device 110 is a dedicated or general-purpose computer provided in the substrate processing apparatus 100, and includes a control unit 111, a memory unit 112, an input unit 113, an output unit 114, a communication unit 115, etc. The edge device 110 monitors the state of the substrate processing apparatus 100 and controls the operation of the actuator A1 based on sensor data from the sensor S1 input via the input unit 113.

[0042] The control unit 111 includes a CPU (Central Processing Unit), ROM (Read Only Memory), and RAM (Random Access Memory). The ROM of the control unit 111 stores control programs that regulate the operation of the various hardware components of the edge device 110. The CPU within the control unit 111 reads and executes the control programs stored in the ROM or various computer programs stored in the memory unit 112 to control the operation of the various hardware components, thereby enabling the entire device to function as the information processing device of this invention. The RAM of the control unit 111 temporarily stores data used during the computation process.

[0043] In this embodiment, the control unit 111 is configured with a CPU, ROM, and RAM, but the configuration of the control unit 111 is not limited to those described above. The control unit 111 may, for example, be one or more control circuits or arithmetic circuits including a GPU (Graphics Processing Unit), FPGA (Field Programmable Gate Array), DSP (Digital Signal Processor), quantum processor, volatile or non-volatile memory, etc. Furthermore, the control unit 111 may have functions such as a clock that outputs date and time information, a timer that measures the elapsed time from the start of measurement to the end of measurement, and a counter that counts the number of measurements.

[0044] The memory unit 112 is equipped with memory devices such as HDD (Hard Disk Drive), SSD (Solid State Drive), and EEPROM (Electrically Erasable Programmable Read Only Memory). Various computer programs executed by the control unit 111 or various data used by the control unit 111 are stored in the memory unit 112.

[0045] The computer programs stored in the memory unit 112 include a learning processing program PG11 for generating the learning models (observation model MD11 and control model MD12) described later, and an estimation processing program PG12 for using the learning models to estimate the state of the board processing device 100 or the control value of the actuator A1. These computer programs may be a single program or may contain multiple programs. Furthermore, these computer programs may partially utilize existing libraries.

[0046] The computer programs, such as the learning processing program PG11 or the estimated processing program PG12, stored in the memory unit 112 are provided by a non-temporary recording medium RM10 that can readablely record the computer programs. The recording medium RM10 is a portable memory such as a CD (Compact Disc)-ROM, USB (Universal Serial Bus) memory, SD (Secure Digital) card, microSD card, or mini flash memory. The control unit 111 uses a reading device (not shown) to read various computer programs from the recording medium RM10 and stores the read computer programs in the memory unit 112. The computer programs stored in the memory unit 112 can be provided via communication. In this case, the control unit 111 only needs to obtain the computer program via the communication unit 115 and store the obtained computer program in the memory unit 112.

[0047] The memory unit 112 has a learning model configured to output information related to the substrate processing device 100 when time-series data output by the self-sensor S1 is input. The memory unit 112 stores, for example, information on the structure of the layers of the learning model, information on the nodes contained in each layer, and parameters for weighting and biasing between nodes, as information defining the learning model.

[0048] The edge device 110 of this embodiment includes an observation model MD11 and a control model MD12 as learning models. The observation model MD11 is a model used to estimate the state of the substrate processing device 1. The control model MD12 is a model used to estimate the control value of the control object of the edge device 110, namely the actuator A1. In the example shown in FIG5, it is assumed that both the observation model MD11 and the control model MD12 are used as learning models, but it is also possible to have only one of them. Furthermore, in the example shown in FIG5, it is assumed that one observation model MD11 and one control model MD12 are used, but the observation model MD11 can also be prepared for each sensor to be observed, and the control model MD12 can also be prepared for each actuator to be controlled.

[0049] The input unit 113 has an interface for connecting the sensor S1. The sensor S1 connected to the input unit 113 is, for example, an RF sensor. The sensor S1 connected to the input unit 113 is not limited to those described above, and any sensor required for observing process status (performance) can be connected. The time-series data (sensor data) input via the input unit 113 is temporarily stored in the memory unit 112.

[0050] The output unit 114 has an interface for connecting the control object, namely the actuator A1, of the edge device 110. The control unit 111 uses the control model MD12 described above to estimate control values ​​and outputs control commands based on the estimated control values ​​from the output unit 114 to the actuator A1, thereby controlling the operation of the actuator A1.

[0051] The communication unit 115 has a communication interface for sending and receiving various data between itself and the control device 150. The communication interface of the communication unit 115 can be a communication interface based on communication standards such as LAN. When the control unit 111 inputs data to be sent, the communication unit 115 sends the data to the control device 150; when it receives data sent from the control device 150, it outputs the received data to the control unit 111.

[0052] Figure 6 is a schematic diagram illustrating an example of the configuration of the observation model MD11 possessed by the edge device 110. The observation model MD11 is configured to output relevant information (hereinafter referred to as state information) about the state of the substrate processing device 100 equipped with the sensor S1 when input time-series data (sensor data) obtained by the sensor S1 is received. As the observation model MD11, any model capable of parsing time-series data can be used. In one example, the observation model MD11 includes a machine learning model using deep learning, constructed from RNN (Recurrent Neural Network), LSTM (Long Short-Term Memory), etc. The observation model MD11 can also be replaced by learning models based on CNN (Convolutional Neural Network), R-CNN (Region-based CNN), YOLO (You Only Look Once), SSD (Single Shot Multibox Detector), GAN (Generative Adversarial Network), SVM (Support Vector Machine), decision trees, etc. Furthermore, the observation model MD11 can also be an autoregressive model, moving average model, autoregressive moving average model, and other learning models other than deep learning.

[0053] The observation model MD11 has an input layer MD11a, intermediate layers MD11b and MD11c, and an output layer MD11d. In the example in Figure 6, the observation model MD11 is assumed to have two intermediate layers MD11b and MD11c, but it can also have three or more intermediate layers.

[0054] One or more nodes are provided in the input layer MD11a, intermediate layers MD11b, MD11c, and output layer MD11d. Nodes in each layer are unidirectionally linked with nodes in the preceding and following layers using the required weights and biases. The same number of data points as the number of nodes in the input layer MD11a are input to the input layer MD11a of the observation model MD11. In this embodiment, the sensor data input to the nodes in the input layer MD11a is time-series data obtained by sensor S1. The sensor data input to the input layer MD11a can be a plurality of continuous measurements over time, or it can be a graph (image data) obtained by plotting the measurements over time.

[0055] The input sensor data is output from the nodes of the input layer MD11a to the nodes of the first intermediate layer MD11b. The data input to the first intermediate layer MD11b is then output from the nodes of the second intermediate layer MD11c to the nodes of the second intermediate layer MD11c. During this process, feature values ​​of the sensor data are extracted using an activation function that incorporates the weights and biases set between the nodes of each layer.

[0056] The feature values ​​of the sensor data extracted from the intermediate layers MD11b and MD11c are output to the output layer MD11d and extracted to the outside of the observation model MD11. The output layer MD11d uses the feature values ​​input from the second intermediate layer MD11c to perform a preset calculation and outputs the status information of the substrate processing device 100 as the final calculation result.

[0057] The status information output by the output layer MD11d includes evaluation values ​​indicating the status of the substrate processing apparatus 100. Evaluation values ​​may, for example, indicate the degree of degradation of a specific component constituting the substrate processing apparatus 100. Alternatively, evaluation values ​​may indicate whether a specific component is faulty. Specific components constituting the substrate processing apparatus 100 include an exhaust device 51, a heater power supply 64, a gas supply source 73, and a high-frequency power supply 76.

[0058] The observation model MD11 is learned using an arbitrary learning algorithm. Supervised learning can be employed as the learning algorithm. In this case, learning is performed as follows: when sensor data, including sensor data from sensor S1, and the correct solution data to be output by the observation model MD11 are used as training data, and sensor data is input, the status information of the board processing device 100 is output. The training data can be provided by the administrator of the board processing device 100, etc. For example, accumulated sensor data of sensor S1, the date and time of component replacement, the date and time of fault discovery, etc., can be used as historical data. Based on this historical data, sensor data of sensor S1 and correct solution data indicating whether there was degradation or fault at the date and time the sensor data was obtained can be provided as training data.

[0059] The observation model MD11 shown in Figure 6 is configured with an input layer MD11a, intermediate layers MD11b and MD11c, and an output layer MD11d. However, the configuration of the observation model MD11 is not limited to that shown in Figure 6. For example, the observation model MD11 can also be a model that does not have intermediate layers MD11b and MD11c and only specifies the relationship between input and output (i.e., the relationship between sensor data and state information).

[0060] In this embodiment, supervised learning is described as the learning algorithm for the observation model MD11, but any learning algorithm that includes unsupervised learning can be used to generate the observation model MD11.

[0061] When the edge device 110's control unit 111 has performed the calculations by the observation model MD11, it obtains status information from the output layer MD11d and acquires feature values ​​of the sensor data extracted from the intermediate layers MD11b and MD11c. The control unit 111 then sends the acquired status information of the substrate processing device 100 and the feature values ​​of the sensor data to the upper-level device of the edge device 110, namely the control device 150.

[0062] Figure 7 is a schematic diagram illustrating an example of the configuration of the control model MD12 provided by the edge device 110. The control model MD12 is configured to output information related to the control values ​​of the substrate processing device 100 equipped with the sensor S1 (hereinafter referred to as control information) when time-series data (sensor data) obtained from the sensor S1 is input. That is, when the sensor data has been input to the input layer MD12a of the control model MD12, calculations are performed to extract feature values ​​of the sensor data from the intermediate layers MD12b and MD12c. The feature values ​​obtained from the intermediate layers MD12b and MD12c are output to the output layer MD12d and extracted to the outside of the control model MD12. The output layer MD12d uses the feature values ​​input from the second intermediate layer MD12c to perform a predetermined calculation and outputs the control information of the substrate processing device 100 as the final calculation result. The characteristic values ​​of the sensor data output from the intermediate layer MD12c and the control information output from the output layer MD12d are input to the control unit 111. The control information output from the output layer MD12d includes control values ​​for controlling at least one component of the substrate processing device 100.

[0063] Similar to the observation model MD11, the control model MD12 can be learned through supervised learning or unsupervised learning. Furthermore, the control model MD12 can also be learned through reinforcement learning. For example, by providing rewards based on the state of the substrate processing device 100 and learning the value in reinforcement learning to maximize the total future rewards. For instance, in Q-learning, a type of reinforcement learning, the value Q of choosing an action (control value) is learned under a certain state in an environment. At the start of Q-learning, the correct value Q for the combination of the state and action (control value) of the substrate processing device 100 is unknown. Therefore, under certain operating data, various control values ​​are selected, and the total reward is calculated based on the reward given to the action at that time (control based on the control value), selecting a better control value, thereby learning the correct value Q.

[0064] The control model MD12 is not limited to the models described above, and can also be other models capable of analyzing time series data. For example, the control model MD12 can also be an autoregressive model, a moving average model, an autoregressive moving average model, or other learning models other than deep learning. Furthermore, the structure of the control model MD12 is not limited to that shown in Figure 7. For example, the control model MD12 can also be a model that does not have intermediate layers MD12b and MD12c, but only specifies the relationship between input and output (i.e., the relationship between sensor data and control information).

[0065] When the control unit 111 of the edge device 110 has performed the calculations performed by the control model MD12, it obtains control information from the output layer MD12d and acquires the feature values ​​of the sensor data calculated in the intermediate layers MD12b and MD12c. The control unit 111 sends the acquired control information from the substrate processing device 100 and the feature values ​​of the sensor data to the upper-level device, i.e., the control device 150, of the edge device 110. Furthermore, the control unit 111 controls the operation of the actuator A1 based on the control information obtained from the control model MD12.

[0066] In this embodiment, feature values ​​are extracted from intermediate layers MD11c and MD12c. However, the final calculation results obtained from output layers MD11d and MD12d can also be considered as feature values ​​of the sensor data. Furthermore, the control unit 111 can also directly extract feature values ​​from the sensor data. Feature value extraction uses appropriate statistical processing such as peak detection and interval averaging. The control unit 111 can extract feature values ​​by detecting abnormal regions in the sensor data and weighting the data of the detected abnormal regions. Additionally, the control unit 111 can also extract a snapshot of time-series data containing abnormal regions in the sensor data as feature values.

[0067] Figures 5-7 illustrate the internal structure of edge device 110, but the internal structures of edge devices 120 and 130 are similarly described. That is, edge devices 120 and 130 each possess an observation model and a control model, respectively, and based on sensor data input from self-sensors S2 and S3, they estimate the state of the substrate processing device 100 and the control values ​​of actuators A1 and A2. Edge devices 120 and 130 send the state information of the substrate processing device 100 obtained from the observation model and control model, the control information of actuators A1 and A2, and the feature values ​​of the sensor data output from sensors S2 and S3 to the host device, i.e., the control device 150. Furthermore, edge devices 120 and 130 control the actions of actuators A2 and A3 respectively based on the control information obtained from the self-control model.

[0068] Figure 8 is a block diagram showing the internal structure of the control device 150 included in the substrate processing apparatus 100. The control device 150 is a dedicated or general-purpose computer installed inside the substrate processing apparatus 100, including a control unit 151, a memory unit 152, a first communication unit 153, a second communication unit 154, an operation unit 155, and a display unit 156. The control device 150 collects data (feature values ​​of sensor data) transmitted from edge devices 110-130 and accumulates it in a database DB20 within the memory unit 152. Based on the accumulated interim data in the database DB20, the control device 150 generates a second learning model that absorbs individual differences in the sensors.

[0069] The control unit 151 includes a CPU, ROM, RAM, etc. The ROM in the control unit 151 stores control programs that control the operation of various hardware components in the control device 150. The CPU in the control unit 151 reads and executes the control programs stored in the ROM or various computer programs stored in the memory unit 152 to control the operation of various hardware components.

[0070] The control unit 151 is not limited to the above-described configuration, and may also be one or more control circuits or arithmetic circuits equipped with GPU, FPGA, DSP, quantum processor, volatile or non-volatile memory, etc. Furthermore, the control unit 151 may also have functions such as a clock that outputs date and time information, a timer that measures the elapsed time from the time the measurement start instruction is given to the time the measurement end instruction is given, and a counter that counts the number of times.

[0071] The memory unit 152 includes memory devices such as HDD, SSD, and EEPROM. The memory unit 152 also includes the aforementioned database DB20. Figure 9 is a conceptual diagram showing an example of the configuration of the database DB20. The database DB20 is associated with the identification information (device ID) of the edge devices 110-130 to store date and time information and characteristic values ​​of sensor data. Furthermore, the database DB20 can also store status information and control information of the substrate processing device 100.

[0072] In addition to storing the database DB20, the memory unit 152 also stores various computer programs executed by the control unit 151 and various data used by the control unit 151.

[0073] The computer program stored in memory unit 152 includes a learning processing program PG21 for generating the second learning model and an estimation processing program PG22 for using the second learning model to estimate the state or control values ​​of the board processing device 100. The computer program stored in memory unit 152 is provided by a non-temporary recording medium RM20 on which the computer program is readablely stored. Furthermore, the computer program stored in memory unit 152 can also be provided via communication.

[0074] The memory unit 152 has a second learning model, which is configured to output information related to the substrate processing device 100 when feature values ​​of sensor data have been input. The memory unit 152 stores information such as the structure of the layers of the second learning model, the information of the nodes contained in each layer, and the weighting and biasing information between the nodes, as information defining the second learning model.

[0075] The control device 150 includes an observation model MD21 and a control model MD22 as a second learning model. The observation model MD21 is used to estimate the state of the substrate processing device 100. The control model MD22 is used to estimate the control values ​​used in the substrate processing device 100. Since the configuration of the observation model MD21 and the control model MD22 is the same as that of the observation model MD11 and the control model MD12 of the edge devices 110-130, detailed descriptions are omitted.

[0076] In this embodiment, the second learning model is configured to have both an observation model MD21 and a control model MD22, but it may also have only one of them. Furthermore, in this embodiment, the second learning model is configured to have one observation model MD21 and one control model MD22, but the observation model MD21 may be prepared for each observation object, and the control model MD22 may be prepared for each control object.

[0077] The first communication unit 153 has a communication interface for sending and receiving various data between itself and the edge devices 110-130. The communication interface of the first communication unit 153 can be a communication interface based on communication standards such as LAN. When the control unit 151 inputs data to be sent, the first communication unit 153 sends the data to the destination edge devices 110-130; when it receives data sent from the edge devices 110-130, it outputs the received data to the control unit 151.

[0078] The second communication unit 154 has a communication interface for sending and receiving various types of data. The communication interface of the second communication unit 154 is, for example, a communication interface based on the LAN communication standard used in WiFi (registered trademark) or Ethernet (registered trademark). When the control unit 151 inputs data to be sent, the second communication unit 154 sends the data to the designated destination. Furthermore, when the second communication unit 154 receives data sent from an external device, it outputs the received data to the control unit 151.

[0079] The operation unit 155 is equipped with operating devices such as a touch panel, keyboard, and switches, and is subject to various operations and settings performed by administrators. The control unit 151 performs appropriate control based on various operation information provided by the operation unit 155, and stores setting information in the memory unit 152 as needed.

[0080] Display unit 156 is equipped with a display device such as a liquid crystal display or an organic EL (Electro-Luminescence) display device, and displays information that should be reported to the administrator or the like according to the instructions from control unit 151.

[0081] Next, the configuration of the device cluster server 200 will be explained. Figure 10 is a block diagram showing the internal structure of the device cluster server 200. The device cluster server 200 includes a control unit 201, a memory unit 202, a communication unit 203, an operation unit 204, and a display unit 205. The device cluster server 200 collects data sent from multiple board processing devices 100 and accumulates it in a database DB30 located in the memory unit 202. Based on the long-term data accumulated in the database DB30, the device cluster server 200 generates a third learning model that absorbs individual differences among devices.

[0082] The control unit 201 includes a CPU, ROM, RAM, etc. The ROM in the control unit 201 stores control programs for the operation of various hardware components of the control device cluster server 200. The CPU in the control unit 201 reads and executes the control programs stored in the ROM or various computer programs stored in the memory unit 202 to control the operation of various hardware components.

[0083] The control unit 201 is not limited to the above-described configuration, and may also be one or more control circuits or arithmetic circuits equipped with GPU, FPGA, DSP, quantum processor, volatile or non-volatile memory, etc. Furthermore, the control unit 201 may also have functions such as a clock that outputs date and time information, a timer that measures the elapsed time from the time the measurement start instruction is given to the time the measurement end instruction is given, and a counter that counts the number of times.

[0084] The memory unit 202 includes memory devices such as HDD, SSD, and EEPROM. The memory unit 202 also includes the aforementioned database DB30. Figure 11 is a conceptual diagram showing an example of the configuration of the database DB30. The database DB30 is associated with the identification code (device ID) of the substrate processing device 100 to store date and time information and characteristic values ​​of sensor data. It can also store status information indicating the state of the substrate processing device 100 and control information used to control the substrate processing device 100.

[0085] In addition to storing the database DB30, the memory unit 202 also stores various computer programs executed by the control unit 201 and various data used by the control unit 201.

[0086] The computer program stored in the memory unit 202 includes a learning processing program PG31 for generating the third learning model and an estimation processing program PG32 for estimating the state or control values ​​of the board processing device 100. The computer program stored in the memory unit 202 is provided by a non-temporary recording medium RM30 on which the computer program is readablely stored. Furthermore, the computer program stored in the memory unit 202 can also be provided via communication.

[0087] The memory unit 202 has a third learning model, which is configured to output information related to the substrate processing device 100 when feature values ​​of sensor data have been input. The memory unit 202 stores information such as the layer composition information of the third learning model, the node information contained in each layer, and the weighting and bias information between nodes, as information defining the third learning model.

[0088] The device cluster server 200 includes an observation model MD31 and a control model MD32 as a third learning model. The observation model MD31 is used to estimate the state of the substrate processing device 100. The control model MD32 is used to estimate the control values ​​used in the substrate processing device 100. Since the configuration of the observation model MD31 and the control model MD32 is the same as that of the observation model MD21 and the control model MD22 of the control device 150, their detailed description is omitted.

[0089] In this embodiment, the third learning model is configured to have both an observation model MD31 and a control model MD32, but it may also have only one of them. Furthermore, in this embodiment, the third learning model is configured to have one observation model MD31 and one control model MD32, but the observation model MD31 may be prepared for each observation object, and the control model MD32 may be prepared for each control object.

[0090] The communication unit 203 has a communication interface for sending and receiving various types of data. The communication interface of the communication unit 203 may be, for example, a communication interface based on the LAN communication standard used in WiFi (registered trademark) or Ethernet (registered trademark). When the control unit 201 inputs data to be sent, the communication unit 203 sends the data to the designated destination. Furthermore, when the communication unit 203 receives data sent from an external device, it outputs the received data to the control unit 201.

[0091] The operation unit 204 is equipped with operating devices such as a touch panel, keyboard, and switches, and is subject to various operations and settings performed by administrators. The control unit 201 performs appropriate control based on various operation information provided by the operation unit 204, and stores setting information in the memory unit 202 as needed.

[0092] The display unit 205 is equipped with a display device such as a liquid crystal display or an organic EL display, and displays information that should be reported to the administrator according to the instructions from the control unit 201.

[0093] In the example of Figure 10, the device cluster server 200 is configured with an operation unit 204 and a display unit 205. However, in the device cluster server 200, the operation unit 204 and the display unit 205 are not essential components. When the operation unit 204 is not present, the device cluster server 200 can simply receive operations from an external computer communicatively connected via the communication unit 203. Furthermore, when the display unit 205 is not present, the device cluster server 200 can simply send information that should be reported to administrators, etc., to the external computer via the communication unit 203 and display it on the external computer.

[0094] The operation of the substrate processing system will be explained below. In the substrate processing system of this embodiment, during the learning phase before application, edge devices 110-130 generate a first learning model (observation model MD11 and control model MD12).

[0095] Figure 12 is a flowchart illustrating the steps of generating the first learning model performed by the edge device 110. The control unit 111 of the edge device 110 collects sensor data output from the sensor S1 in a time sequence via the input unit 113 (step S101). The sensor data collection period is, for example, one month. When acquiring sensor data, the control unit 111 receives status information from the substrate processing device 100 externally and obtains control values ​​output to the substrate processing device 100 by the output unit 114. The control values ​​of the substrate processing device 100 used during the stage before the learning of the control model MD12 is completed are determined, for example, by referring to a pre-set process recipe. These status information or control values, along with the sensor data, are stored in the memory unit 112 as training data during the learning of the observation model MD11 or the control model MD12.

[0096] When the required training data for learning is obtained, the control unit 111 selects a set of training data from the training data stored in the memory unit 112 (step S102). The control unit 111 inputs the sensor data contained in the selected set of training data into the observation model MD11 and the control model MD12 respectively, and performs the calculations of the observation model MD11 and the control model MD12 (step S103). In the stage before the start of learning, the model parameters of the observation model MD11 and the control model MD12 are set with initial values.

[0097] The control unit 111 evaluates the computation results of the observation model MD11 and the control model MD12 (step S104) and determines whether the learning of the observation model MD11 and the control model MD12 has been completed (step S105). The control unit 111 can use the error function (also called the objective function, loss function, or cost function) based on the computation results obtained from the model and the state or control values ​​included as positive solution data to evaluate the computation results. For example, in a course where the error function is optimized (minimized or maximized) using gradient descent methods such as the steepest descent method, the control unit 111 determines that the learning of the observation model MD11 and the control model MD12 has been completed when the error function becomes below (or above) the threshold.

[0098] In the case where learning is incomplete (S105: No), that is, when the learning of either the observation model MD11 or the control model MD12 is incomplete, the control unit 111 updates the parameters (weights and biases between nodes, etc.) of the model whose learning is incomplete (step S106), and the processing returns to step S102. The control unit 111 can use the error backpropagation method, which updates the weights and biases between nodes sequentially from the output layers MD11d and MD12d to the input layers MD11a and MD12a, to update the parameters in the model.

[0099] When it is determined that the learning has been completed (S105: Yes), the pre-learned observation model MD11 and control model MD12 are obtained. Therefore, the control unit 111 stores them as the first learning model in the memory unit 112 (step S107).

[0100] Figure 12 illustrates the generation sequence of the first learning model performed by edge device 110. However, edge devices 120 and 130 can also generate the first learning model applied to each of them using the same generation sequence.

[0101] After generating the first learning model in each of the edge devices 110-130, the substrate processing system of this embodiment moves to the application stage. The substrate processing system performs the following processing in the application stage.

[0102] Figure 13 is a flowchart illustrating the processing steps performed inside the substrate processing apparatus 100 during the application phase. When the control unit 111 of the edge device 110 (120, 130) provided in the substrate processing apparatus 100 obtains sensor data output in a time sequence from the self-sensor S1 via the input unit 113 (step S121), it inputs the obtained sensor data to the observation model MD11 or the control model MD12, thereby executing the model (step S122).

[0103] During the model execution process, the control unit 111 extracts feature values ​​from the intermediate layers MD11c and MD12c of each model MD11 and MD12 (step S123). For example, the control unit 111 can extract feature values ​​from the intermediate layers MD11c and MD12c. Alternatively, the control unit 111 can use the final calculation results obtained from the output layers MD11d and MD12d as the feature values ​​of the sensor data, or it can directly extract feature values ​​from the sensor data.

[0104] The control unit 111 sends the extracted feature values ​​along with the estimated results of the state information and control information obtained from each model MD11 and MD12 to the control device 150 (step S124). Furthermore, the control unit 111 executes control of the actuator A1 based on the estimated state information and control information from each model MD11 and MD12 (step S125). Moreover, the control unit 111 only needs to execute steps S122 to S125 each time sensor data is obtained in step S122.

[0105] The control unit 151 of the control device 150 receives the feature value sent from the edge device 110 (120, 130) from the first communication unit 153 (step S126) and accumulates it in the database DB20 (step S127).

[0106] Control unit 151 determines whether the feature value collection period has ended (step S128). The collection period is, for example, 6 months after the start of feature value collection. Alternatively, the determination of whether the collection period has ended can be based on the number of feature values ​​accumulated in database DB20. If the collection period has not ended (S128: No), control unit 151 returns the process to step S126 and repeats the process of receiving feature values ​​and accumulating them in database DB20.

[0107] If the data collection period has ended (S128: Yes), the learning of observation model MD21 and control model MD22 is performed to create the model (step S129). For example, control unit 151 can create observation model MD21 by using a set of feature values ​​and state information stored in database DB20 as training data. Similarly, control unit 151 can create control model MD22 by using a set of feature values ​​and control values ​​stored in database DB20 as training data. The model creation sequence is the same as that of observation model MD11 and control model MD12.

[0108] When the control unit 151 receives new feature values ​​from the edge devices 110 (120, 130) after the model is created, it inputs the received feature values ​​into the observation model MD21 or the control model MD22 and executes the model (step S130).

[0109] Based on the execution result of step S130, control unit 151 determines whether the model used in edge devices 110 (120, 130) needs to be updated (step S131). In control device 150, since the model is created based on accumulated interim data (e.g., data in 6-month units), control unit 151 can execute the model based on newly acquired feature values ​​and determine the deviation from the trend exhibited by the model. If the deviation from the trend exhibited by the model is above a threshold, control unit 151 determines that the model possessed by edge devices 110 (120, 130) is abnormal. If control unit 151 determines that the model is not abnormal, it determines that the model does not need to be updated (S131: No), and control unit 151 returns the process to step S130.

[0110] When the control unit 151 determines that the model is abnormal, it determines that the model needs to be updated (S131: Yes) and sends a model relearning instruction to the edge devices 110 (120, 130) (step S132).

[0111] The control unit 111 of the edge device 110 (120, 130) determines whether it has received a relearning instruction sent by the control device 150 (step S133). If it is determined that no relearning instruction has been received (S133: no), the control unit 111 returns the process to step S121 and repeats the process of steps S121 to S125.

[0112] Upon receiving a relearning instruction (S133: Yes), the control unit 111 performs relearning (step S134). For example, the control unit 111 may relearn the observation model MD11 by using sensor data obtained from the sensor S1 and state information of the substrate processing device 100 as training data for additional learning. Similarly, the control unit 111 may relearn the control model MD12 by using sensor data obtained from the sensor S1 and control values ​​used in the substrate processing device 100 as training data for additional learning. Alternatively, the configuration of the observation model MD11 and the control model MD12 may be relearned from scratch using the aforementioned training data, instead of relearning the configuration of the observation model MD11 and the control model MD12 through additional learning.

[0113] Furthermore, in this flowchart, when it is determined that the model in the control device 150 needs to be updated, a relearning instruction is sent to the edge devices 110 (120, 130). However, correction values ​​used to correct the calculation results of each model MD11 and MD12 can also be sent to the edge devices 110 (120, 130). For example, the correction value can be calculated based on the error between the prediction results obtained from the second learning model (observation model MD21 and control model MD22) and the measured results.

[0114] Figure 14 is a flowchart illustrating the processing steps performed between the substrate processing apparatus 100 and the device group server 200 during the application phase. As described above, during the application phase, the control device 150 of the substrate processing apparatus 100 creates an observation model MD21 and a control model MD22 based on the feature values ​​obtained from the edge devices 110-130. When the control unit 151 of the control device 150 receives new feature values ​​from the edge devices 110 (120, 130) after model creation (step S141), it inputs the received feature values ​​into the observation model MD21 or the control model MD22 and executes the model (step S142).

[0115] During the execution of the model, the control unit 111 extracts feature values ​​of sensor data from the intermediate layer of each model MD21 and MD22 (step S143), and sends the extracted feature values, along with the estimated results of state information and control information obtained from each model MD21 and MD22, to the device cluster server 200 (step S144). In this embodiment, the configuration is set to extract feature values ​​of sensor data from the intermediate layer of each model MD21 and MD22 and send them to the device cluster server 200. However, it can also be configured to treat the final calculation results obtained from the output layer of each model MD21 and MD22 as feature values ​​of sensor data and send them to the device cluster server 200.

[0116] The control unit 201 of the device group server 200 receives the feature value sent from the board processing device 100 from the communication unit 203 (step S145) and accumulates it in the database DB30 (step S146).

[0117] The control unit 201 determines whether the feature value collection period has ended (step S147). The collection period is, for example, 2 to 3 years after the start of feature value collection. Alternatively, the end of the collection period can be determined based on the number of feature values ​​accumulated in the database DB30. If the collection period has not ended (S147: No), the control unit 201 returns the process to step S145 and repeats the process of receiving feature values ​​and accumulating them in the database DB30.

[0118] If the data collection period has ended (S147: Yes), the learning of observation model MD31 and control model MD32 is performed to create the models (step S148). For example, control unit 201 can create observation model MD31 by using a set of feature values ​​and state information stored in database DB30 as training data. Similarly, control unit 201 can create control model MD32 by using a set of feature values ​​and control values ​​stored in database DB30 as training data. The model creation sequence is the same as that of observation model MD11 and control model MD12.

[0119] When the control unit 201 receives new feature values ​​from the substrate processing device 100 after the model is created, it inputs the received feature values ​​into the observation model MD31 or the control model MD32 and executes the model (step S149).

[0120] Based on the execution result of step S149, the control unit 201 determines whether the model used in the edge devices 110 (120, 130) of the substrate processing apparatus 100 needs to be updated (step S150). In the device cluster server 200, since the model is created based on accumulated long-term data (e.g., data in units of 2-3 years), the control unit 201 can execute the model based on newly acquired feature values ​​and determine the deviation from the trend exhibited by the model. If the deviation from the trend exhibited by the model is above a threshold, the control unit 201 determines that the model possessed by the edge devices 110 (120, 130) is abnormal. If the control unit 201 determines that the model is not abnormal, it determines that the model does not need to be updated (S150: No), and the control unit 201 returns the processing to step S149.

[0121] When the control unit 201 determines that the model is abnormal, it determines that the model needs to be updated (S150: Yes) and sends a model relearning instruction to the substrate processing device 100 (step S151).

[0122] The control unit 151 of the control device 150 of the substrate processing apparatus 100 determines whether a relearning instruction sent from the device group server 200 has been received (step S152). If it is determined that no relearning instruction has been received (S152: no), the control unit 151 returns the processing to step S141 and repeats the processing of steps S141 to S144.

[0123] Upon receiving a relearning instruction (S152: Yes), the control unit 151 instructs the edge devices 110 (120, 130) to perform relearning (step S153). The control unit 111 of the edge devices 110 (120, 130) can, for example, perform additional learning using sensor data obtained from the sensor S1 and state information of the substrate processing device 100 as training data to relearn the observation model MD11. Alternatively, the control unit 111 can, for example, perform additional learning using sensor data obtained from the sensor S1 and control values ​​used in the substrate processing device 100 as training data to relearn the control model MD12. Alternatively, the configuration of the observation model MD11 and control model MD12 can be relearned from scratch using the aforementioned training data, instead of relearning the configuration of the observation model MD11 and control model MD12 through additional learning.

[0124] In this flowchart, when the device cluster server 200 issues a relearning instruction, the configurations of the observation model MD11 and control model MD12 are relearned in the edge devices 110 (120, 130). However, it is also possible to relearn the configurations of the observation model MD21 and control model MD22 in the control device 150. The control unit 151 of the control device 150 can use the accumulated feature values ​​and state information in the database DB20 as training data to relearn the observation model MD21. Furthermore, the control unit 151 can use the accumulated feature values ​​and control values ​​in the database DB20 as training data to relearn the control model MD22. The control unit 151 can relearn the observation model MD21 and control model MD22 through supplementary learning, or it can relearn the observation model MD21 and control model MD22 from scratch using the aforementioned training data.

[0125] Furthermore, in this flowchart, it is configured to send a relearning instruction to the board processing device 100 when the model in the determination device group server 200 needs to be updated. However, it is also possible to send a correction value of the calculation result used to correct the model to the board processing device 100. For example, the correction value can be calculated based on the error between the prediction result obtained from the third learning model (observation model MD31 and control model MD32) and the measured result.

[0126] As described above, in this embodiment, high-resolution models (observation model MD11 and control model MD12) based on sensor data can be created in each edge device 110-130. Furthermore, the edge devices 110-130 can also create models that estimate the degradation of components included in the substrate processing apparatus 100 based on the acquired sensor data.

[0127] Each edge device 110-130 sends feature values ​​extracted by each model to the control device 150 instead of sending sensor data, thus reducing the network load between the edge devices 110-130 and the control device 150. The control device 150 can create a model of medium-term trend (observation model MD21 and control model MD22) based on the feature values ​​of the sensor data.

[0128] Each substrate processing device 100 sends the feature values ​​extracted from each model to the device cluster server 200, thereby reducing the network load between the substrate processing device 100 and the device cluster server 200. The device cluster server 200 can create a model (observation model MD31 and control model MD32) that absorbs the long-term trend of individual differences between devices based on the feature values ​​sent from the substrate processing device 100.

[0129] Furthermore, when a new substrate processing device (not shown) is introduced into the substrate processing system, the pre-learned first learning model (observation model MD11 and control model MD12) can be deployed to each edge device 110-130. Additionally, the pre-learned second learning model (observation model MD21 and control model MD22) can also be deployed to the control device 150 of each substrate processing device 100.

[0130] (Implementation Method 2) In Implementation Method 2, the structure for evaluating the completeness or soundness of the first learning model and outputting the evaluation results is explained. Furthermore, the system configuration, the internal configuration of the substrate processing device 100 and the device group server 200 are the same as in Embodiment 1, so their description is omitted.

[0131] At an appropriate time during the learning or application phase, the substrate processing device 100 evaluates the completeness or soundness of the first learning model (observation model MD11 and control model MD12) possessed by the edge devices 110-130 and outputs the evaluation results.

[0132] An evaluation dataset is prepared in the substrate processing apparatus 100 to evaluate the first learning model (observation model MD11 and control model MD12). For example, to evaluate the observation model MD11 of the edge device 110, an evaluation dataset can be used, consisting of sensor data including sensor S1 and the set of forward retrieval data that the observation model MD11 should output when the sensor data of sensor S1 has been input. Similarly, to evaluate the control model MD12 of the edge device 110, an evaluation dataset can be used, consisting of sensor data including sensor S1 and the set of forward retrieval data that the control model MD12 should output when the sensor data of sensor S1 has been input. The same applies to the evaluation datasets used to evaluate the edge devices 120 and 130.

[0133] The substrate processing apparatus 100 can evaluate the completeness or soundness of the first learning model based on the difference between the estimated value obtained when the sensor data contained in the evaluation dataset is input into the observation model MD11 and the control model MD12 and the correct solution data contained in the dataset.

[0134] When the substrate processing apparatus 100 evaluates the completeness or soundness of the first learning model, it displays the evaluation results on the display unit 156. Figure 15 is a schematic diagram showing an example of the display of evaluation results. In the example of Figure 15, the evaluation results of the completeness and soundness of the observation model MD11 and control model MD12 possessed by each of the edge devices 110 to 130 are shown. Furthermore, in each graph, indices A, B, and C represent edge devices 110, 120, and 130, respectively. The upper graph shows that the completeness of the observation model MD11 and control model MD12 increases with the increase of the number of learning iterations. The lower graph shows the soundness of the observation model MD11 and control model MD12 possessed by each of the edge devices 110, 120, and 130 at the evaluation point.

[0135] Thus, in Implementation 2, the performance of each learning model can be displayed at a glance. Therefore, when the completion or soundness of the observation model MD11 and the control model MD12 is insufficient, the administrator can give a relearning instruction through the operation unit 155, thereby improving the completion or soundness of the observation model MD11 and the control model MD12.

[0136] The embodiments disclosed herein should be considered illustrative in all respects, not restrictive. The scope of this invention is defined by the claims, not by the foregoing meaning, and is intended to include all modifications within the scope and meaning of the claims.

[0137] 10: Box / casing platform 11: Boxes and containers 20:Transportation platform 21: Substrate conveying mechanism 21A:Conveying arm 21B:Conveying arm 25A: Loaded with a locking chamber 25B: Loading a lockout chamber 30: Transfer Chamber 31: Substrate conveying mechanism 31A: Conveying Arm 31B: Conveying Arm 40: Chamber 40A: Chamber 40B: Chamber 40C: Chamber 40D: Chamber 41: Processing Room 51: Exhaust device 52: Exhaust pipe 53: Valve 61: Base 62: Supporting components 63: Heater 64: Heater power supply 71: Cluster Head 71a: Gas diffusion space 71b: Gas ejection port 72: Gas inlet tube 73: Gas supply source 74: Gas piping 75: MFC (Mass Flow Controller) 76: High-frequency power supply 77: Wiring 78: Matcher 100: Substrate processing apparatus 100A: Substrate Processing Apparatus 100B: Substrate Processing Apparatus 100C: Substrate Processing Apparatus 100D: Substrate Processing Apparatus 110: Edge devices 111: Control Department 112: Memory Department 113: Input Section 114: Output Department 115: Ministry of Communications 120: Edge devices 130: Edge devices 150: Control equipment 151: Control Department 152: Memory Department 153: 1st Communications Department 154: 2nd Communications Department 155: Operations Department 156: Display Section 200: Device Cluster Server 201: Control Department 202: Memory Department 203: Ministry of Communications 204: Operations Department 205: Display Section 411: Top 412: Side wall portion 413: Bottom 413a: Opening 421: Flange portion 422: Side wall portion 423: Bottom 424: Exhaust port A1: Actuator A2: Actuator A3: Actuator DB20: Database DB30: Database HU: Transport Unit MD11: Observation Model MD11a: Input Layer MD11b: Intermediate layer MD11c: Intermediate layer MD11d: Output layer MD12: Control Model MD12a: Input Layer MD12b: Intermediate layer MD12c: Intermediate layer MD12d: Output Layer MD21: Observation Model MD22: Control Model MD31: Observation Model MD32: Control Model NW: Communication Network PG11: Learning Processor PG12: Estimated Processing Method PG21: Learning Processing Program PG22: Presumed Processor PG31: Learning Processor PG32: Estimated Processor RM10: Recording Media RM20: Recording Media RM30: Recording Media S1: Sensor S2: Sensor S3: Sensor SU: Processing Unit TS: Temperature sensor W: substrate

Claims

1. An information processing method comprising the following steps: acquiring time-series data from a plurality of sensors with different sampling periods disposed on a substrate processing device; learning a first learning model for each sensor, i.e., when inputting time-series data with different sampling periods from each of the sensors based on the acquired time-series data, outputting information related to the substrate processing device; and inputting the time-series data from the sensors into the corresponding learned first learning model, and outputting an inference result based on the information obtained from the first learning model.

2. The information processing method of claim 1, wherein the first learning model of each sensor is imported for each edge device corresponding to each sensor, the information processing method includes the following procedures: performing the learning processing of the first learning model in each edge device, and the estimation processing performed by the first learning model.

3. The information processing method of claim 1 or 2, wherein the first learning model includes at least one of the following models: an observation model for estimating the state of the substrate processing device based on time-series data from the sensor; and a control model for estimating control values ​​of the substrate processing device based on time-series data from the sensor.

4. The information processing method of claim 1 or 2 further includes the following procedure: extracting a first feature value from the aforementioned time series data; and outputting the extracted first feature value to a first device within the aforementioned substrate processing apparatus.

5. The information processing method of claim 4, comprising performing the following process in the first device: memorizing the first feature value extracted from the time series data; learning a second learning model, i.e., when a first feature value is input based on the memorized first feature value, outputting information related to the substrate processing device; and inputting a newly acquired first feature value into the learned second learning model, and outputting an inference result based on the information obtained from the second learning model.

6. The information processing method of claim 5 further includes the following procedure: performing the following processing in the first device, that is, outputting a relearning instruction for the first learning model based on the estimation result using the second learning model.

7. The information processing method of request item 5 further includes the following procedure: Based on the calculation result obtained by the second learning model above, output a correction value to correct the calculation result obtained by the first learning model above.

8. The information processing method of claim 5 further includes the following procedure: extracting a second feature value of the time series data for each of the aforementioned substrate processing devices; and outputting the extracted second feature value to a second device outside the aforementioned substrate processing devices.

9. The information processing method of claim 8, comprising performing the following process in the second device: memorizing the second feature value extracted for each of the substrate processing devices; learning a third learning model, i.e., when a second feature value is input based on the memorized second feature value, outputting information related to the substrate processing device; and inputting a newly acquired second feature value into the learned third learning model, and outputting an inference result based on the information obtained from the third learning model.

10. The information processing method of claim 9 further includes the following procedure: performing the following process in the second device, namely, outputting a relearning instruction of the first learning model or the second learning model based on the estimation result using the third learning model.

11. The information processing method of claim 9 further includes the following procedure: Based on the calculation result obtained by the third learning model above, outputting a correction value to correct the calculation result obtained by the first learning model above or the second learning model above.

12. The information processing method of claim 1 or 2 further includes the following procedure: when setting up a new substrate processing device, importing the pre-learned first learning model into the new substrate processing device.

13. The information processing method as described in request item 1 or 2 may further include the following procedure: displaying the performance of each learning model at a glance.

14. An information processing apparatus comprising: an acquisition unit that acquires time-series data from a plurality of sensors having different sampling periods disposed on a substrate processing apparatus; a learning unit that learns a first learning model for each sensor, i.e., when inputting time-series data with different sampling periods from each of the sensors based on the acquired time-series data, outputting information related to the substrate processing apparatus; and an estimation unit that inputs the time-series data from the sensors into the corresponding learned first learning model and outputs an estimation result based on the information obtained from the first learning model.

15. A substrate processing system comprising: a plurality of substrate processing apparatuses, each having an edge device connected to a sensor and a host device connected to the edge device, and performing substrate processing inside a chamber; and a device group server communicatively connected to the plurality of substrate processing apparatuses; wherein the edge device comprises: an acquisition unit that acquires time-series data from the sensor; a first learning unit that learns a first learning model, i.e., when time-series data from the sensor is input based on the acquired time-series data, it outputs information related to the substrate processing apparatus on which the sensor is provided; a first estimation unit that inputs the time-series data from the sensor into the learned first learning model and outputs an estimation result based on the information obtained from the first learning model; and an output unit that outputs a first feature value extracted from the time-series data to the host device; wherein the host device comprises: a first feature value memory unit that stores the first feature value input from the edge device; The second learning unit learns a second learning model, that is, when a first feature value is input based on the memorized first feature value, it outputs information related to the aforementioned substrate processing device; the second estimation unit inputs a newly acquired first feature value into the learned second learning model and outputs an estimation result based on the information obtained from the aforementioned second learning model; and the sending unit sends the second feature value of the time series data extracted for each of the aforementioned substrate processing devices to the aforementioned device group server; the aforementioned device group server includes: a second feature value memory unit that memorizes the second feature value received from the aforementioned host device; a third learning unit learns a third learning model, that is, when a second feature value is input based on the memorized second feature value, it outputs information related to the aforementioned substrate processing device; and a second estimation unit that inputs a newly acquired second feature value into the learned third learning model and outputs an estimation result based on the information obtained from the aforementioned third learning model.

16. The substrate processing system of claim 15, wherein the host device and the device group server comprise: a determination unit that determines whether the first learning model needs to be updated based on the estimation results obtained from the learning models possessed by the host device and the device group server respectively; and an instruction unit that, when it is determined that an update is required, instructs the edge device to relearn the first learning model.

Images