Substrate processing apparatus, data processing method, and program product

Abstract

To reduce the work load of learning work. A substrate processing apparatus has: a reservoir feature amount generation section that inputs acquired time-series first sensor data and outputs a reservoir feature amount; a learning section that, during a learning period, learns a weight parameter so that predicted result data obtained by operating the reservoir feature amount based on the weight parameter is correlated with acquired time-series second sensor data; a prediction section that, during a prediction period, operates the reservoir feature amount output by the reservoir feature amount generation section by inputting the time-series first sensor data based on the learned weight parameter, and outputs predicted result data; and a determination section that, during the prediction period, determines a state of a substrate manufacturing process by comparing the predicted result data and the acquired time-series second sensor data.

Description

Technical Field

[0001] This disclosure relates to a substrate processing apparatus, a data processing method, and a data processing program. Background Technology

[0002] Previously, in the field of substrate processing equipment, time-series sensor data measured by various sensors were used to determine the status of the substrate manufacturing process (e.g., whether there are any abnormalities). In addition, recently, machine learning models have been proposed to improve the accuracy of the determination.

[0003] [Existing Technical Documents]

[0004] [Patent Documents]

[0005] [Patent Document 1] Japanese Patent Application Publication No. 2018-77779

[0006] [Non-patent literature]

[0007] [Non-patent document 1] David Sussillo, LFAbbott, "Generating CoherentPatterns of Activity from Chaotic Neural Networks", Neuron. Authormanuscript; available in PMC 2010 Aug 27. Summary of the Invention

[0008] [The problem the invention aims to solve]

[0009] However, in the case of a typical machine learning model, a large amount of training data is required to improve the judgment accuracy. Therefore, in the case of a board processing device that applies a machine learning model, the workload involved in the learning operation during startup is high. In addition, after startup, as the judgment accuracy decreases over time with the board processing device, a large amount of training data needs to be collected again. Therefore, in the case of a board processing device that applies a machine learning model, the workload involved in the learning operation during maintenance is also high.

[0010] This disclosure reduces the workload of learning operations in a substrate processing apparatus that uses time-series sensor data to determine the state of a substrate manufacturing process.

[0011] [Methods for solving the problem]

[0012] According to one aspect of the present disclosure, a substrate processing apparatus has, for example, the following structure:

[0013] The reservoir feature quantity generation unit inputs the first sensor data of the time series obtained in the substrate manufacturing process and outputs the reservoir feature quantity;

[0014] The learning unit learns weight parameters during the learning process so that the prediction result data obtained by calculating the reservoir feature quantity output by the reservoir feature quantity generation unit based on the weight parameters is correlated with the second sensor data of the time series obtained in the substrate manufacturing process.

[0015] The prediction unit, during the prediction period, calculates the reservoir feature quantity output by the reservoir feature quantity generation unit based on the weight parameters learned by the learning unit, using the first sensor data obtained from the time series during the substrate manufacturing process as input, and outputs the prediction result data; and

[0016] The determination unit determines the state of the substrate manufacturing process by comparing the prediction result data with the time series data of the second sensor obtained during the substrate manufacturing process during the prediction period.

[0017] [The effects of the invention]

[0018] In a substrate processing apparatus that uses time-series sensor data to determine the state of the substrate manufacturing process, the workload of learning operations can be reduced. Attached Figure Description

[0019] Figure 1 Figure 1 shows an example of the system structure of a substrate processing device.

[0020] Figure 2 This is a diagram illustrating an example of the hardware structure of a data processing device.

[0021] Figure 3 Figure 1 is an example of the functional structure of the RC section of a data processing device.

[0022] Figure 4 This is a flowchart illustrating the calculation and processing flow of the storage pool.

[0023] Figure 5 This diagram illustrates an example of the functional structure of the acquisition unit and the determination unit of a data processing device.

[0024] Figure 6 This is the first flowchart representing the process of obtaining and determining the process.

[0025] Figure 7 This is the second flowchart representing the process of obtaining and determining the process.

[0026] Figure 8 This is a diagram illustrating specific examples of acquisition and judgment processing.

[0027] Figure 9 Figure 2 is an example of the functional structure of the RC section of a data processing device.

[0028] Figure 10A Figure 3 shows an example of the functional structure of the RC section of a data processing device.

[0029] Figure 10B Figure 4 shows an example of the functional structure of the RC section of a data processing device.

[0030] Figure 11 This is a diagram illustrating an example of the functional structure of the RC section and the integration section of a data processing device.

[0031] Figure 12 Figure 2 shows an example of the system structure of a substrate processing device.

[0032] Figure 13 Figure 5 shows an example of the functional structure of the RC section of a data processing device.

[0033] Figure 14 Figure 3 shows an example of the system structure of a substrate processing device. Detailed Implementation

[0034] Hereinafter, each embodiment will be described with reference to the accompanying drawings. Furthermore, in this specification and the accompanying drawings, structural elements having substantially the same functional structure are labeled with the same symbols, thereby omitting redundant descriptions.

[0035] [First Implementation]

[0036] <System Structure of Substrate Processing Device>

[0037] First, the system structure of the substrate processing apparatus according to the first embodiment will be described. Figure 1 Figure 1 shows an example of the system structure of a substrate processing apparatus, with arrows indicating data flow. Figure 1 1a represents the flow of data within the substrate processing apparatus 100 during the learning period. Figure 1 1b represents the flow of data within the substrate processing apparatus 100 during the prediction period.

[0038] like Figure 1 As shown in 1a and 1b, the substrate processing apparatus 100 includes a substrate processing chamber 110, a sensor a120 (an example of a first sensor), a data processing device 130, a sensor b140 (an example of a second sensor), a control device 150, and an actuator 160.

[0039] like Figure 1As shown in 1a and 1b, during the process of processing the substrate in chamber 110, time-series sensor data measured by sensor a120 is sent to the Reservoir Computing (hereinafter referred to as RC unit) 131 of the data processing device 130. That is, the time-series sensor data sent to RC unit 131 is an example of the first time-series sensor data obtained in the substrate manufacturing process.

[0040] The RC unit 131 performs reservoir calculations with high performance and fast learning speed. Specifically, the RC unit 131 predicts sensor data measured by the time series measured by the sensor a120 (input data).

[0041] Therefore, as Figure 1 As shown in 1a, during the learning period, the RC unit 131 acquires the time-series sensor data (forward solution data) measured by the sensor b140 via the acquisition unit 132 of the data processing device 130. Then, the RC unit 131 learns the weighting parameter so that the prediction result data obtained by calculating the time-series sensor data (input data) measured by the sensor a120 based on the weighting parameter is correlated with the forward solution data.

[0042] On the other hand, such as Figure 1 As shown in 1b, during the prediction period, the RC unit 131 calculates the time-series sensor data (input data) measured by the sensor a120 based on the learned weight parameters, and outputs the time-series sensor data (prediction result data). In addition, the time-series sensor data (prediction result data) output by the RC unit 131 is notified to the acquisition unit 132 of the data processing device 130.

[0043] Sensor b140 transmits time-series sensor data measured during substrate processing in chamber 110 to acquisition unit 132 of data processing device 130. That is, the time-series sensor data transmitted to acquisition unit 132 is an example of second-series sensor data acquired during substrate manufacturing process.

[0044] like Figure 1 As shown in 1a, the acquisition unit 132 of the data processing device 130 will notify the RC unit 131 of the time-series sensor data (forward solution data) measured by the sensor b140 during the learning period.

[0045] In addition, such as Figure 1As shown in 1b, during the prediction period, the acquisition unit 132 of the data processing device 130 notifies the determination unit 133 of the prediction result data notified by the RC unit 131 and the time series sensor data (comparison data) measured by the sensor b140. In addition, the time series sensor data measured by the sensor b140 is called "positive solution data" during the learning period and "comparison data" during the prediction period.

[0046] In addition, such as Figure 1 As shown in Figure 1b, during the prediction period, the determination unit 133 calculates the difference between the prediction result data notified by the acquisition unit 132 and the comparison data, and determines the state of the substrate manufacturing process based on the calculated difference data. The determination result obtained by the determination unit 133 is notified to the control device 150. The control device 150 controls the actuator 160 according to the notified determination result.

[0047] Thus, in the substrate processing apparatus 100 that uses time-series sensor data to determine the state of the substrate manufacturing process, in the first embodiment...

[0048] • Configure the RC unit 131 to perform reservoir calculations with high performance and fast learning speed.

[0049] • During the learning process, two types of time-series sensor data were used as input data and forward solution data for learning.

[0050] • During the prediction period, the status of the substrate manufacturing process is determined by using two time series sensor data as input data and comparison data.

[0051] Therefore, the substrate processing apparatus 100 according to the first embodiment can effectively perform learning during startup and maintenance, thereby reducing the workload of the learning operation.

[0052] <Hardware Structure of Data Processing Device>

[0053] Next, the hardware structure of the data processing device 130 included in the substrate processing apparatus 100 will be described. Figure 2 This is a diagram illustrating an example of the hardware structure of a data processing device. Among them, Figure 2 2a represents the case where the reservoir feature generation function is implemented in the reservoir feature generator as hardware (details will be discussed later). Additionally, Figure 2 2b refers to the case where the function of generating reservoir characteristics is implemented by computer-executed software (e.g., a reservoir characteristic generation program).

[0054] like Figure 2 As shown, the data processing device 130 includes a processor 201, a memory 202, an auxiliary storage device 203, and an external I / F device 206 (in Figure 2 In the case of 2a, I / F device 204 and reservoir feature generator 205 can also be used. In addition, the various hardware components of data processing device 130 are interconnected via bus 207.

[0055] Figure 2 Processors 201 of 2a and 2b have various computing devices such as CPU and GPU. Processor 201 processes various programs (e.g., data processing programs described later). Figure 2 In the case of 2b, the data processing program (including the reserve pool feature quantity generation program, etc.) is read into the memory 202 and executed.

[0056] Figure 2 The memory 202 of 2a and 2b has main storage devices such as ROM and RAM. The processor 201 and the memory 202 form a so-called computer. The processor 201 executes various programs read from the memory 202, thereby enabling the computer to perform various functions (RC unit 131, acquisition unit 132, determination unit 133).

[0057] Figure 2 The auxiliary storage devices 203 of 2a and 2b store various programs and various data used by the processor 201 when the various programs are executed.

[0058] Figure 2 The I / F device 204 in 2a is a connection device connected to the reservoir characteristic quantity generator 205. Figure 2 The reservoir feature generator 205 of 2a is implemented, for example, by an FPGA (Field-Programmable Gate Array) board. The reservoir feature generator 205 has the function of generating reservoir features, which takes as input the time series sensor data measured by sensor a120 and outputs the reservoir features.

[0059] Figure 2 The external I / F device 206 of 2a and 2b is a connection device connected to sensor a120 and sensor b140.

[0060] <Functional Structure of the RC Section of the Data Processing Unit>

[0061] Next, the functional structure of the RC section 131 of the data processing device 130 will be explained. Figure 3 Figure 1 is an example of the functional structure of the RC section of a data processing device.

[0062] like Figure 3As shown, the RC unit 131 includes a reservoir feature quantity generation unit 300, a prediction unit 310, a period determination unit 320, a switching unit 330, and a learning unit 340. As described above, the reservoir feature quantity generation unit 300 can also be implemented using hardware such as an FPGA board, i.e., a reservoir feature quantity generator 205. However, in this embodiment, it is explained that the reservoir feature quantity generation unit 300 is implemented by a computer executing a reservoir feature quantity generation program.

[0063] In addition, the prediction unit 310 to the learning unit 340 are implemented by a computer executing a data processing program (a data processing program other than the reservoir feature quantity generation program).

[0064] like Figure 3 As shown, the reservoir feature quantity generation unit 300 has the function of generating reservoir feature quantities by taking in sensor data a (input data) of time series obtained from sensor a120 and outputting reservoir feature quantities.

[0065] The prediction unit 310 has a weight acquisition unit 311 and a calculation unit 312. The weight acquisition unit 311 acquires the weight parameters learned by the learning unit 340 and stored in the data storage unit 350, and sets them in the calculation unit 312. Alternatively, during the initial learning phase, the default weight parameters can be set as initial values ​​in the calculation unit 312, or values ​​generated by random numbers can be set.

[0066] The calculation unit 312 notifies the period determination unit 320 of the prediction result data obtained by calculating the reserve pool feature quantity output from the reserve pool feature quantity generation unit 300 based on the weight parameters set by the weight acquisition unit 311.

[0067] The period determination unit 320 determines whether the current period is a learning period or a prediction period based on the prediction result data notified by the calculation unit 312 and the sensor data b (forward solution data) notified by the acquisition unit 132, and notifies the switching unit 330 of the period determination result. Alternatively, the period determination unit 320 notifies the switching unit 330 of the prediction result data regardless of whether the current period is a learning period or a prediction period.

[0068] During the learning period, the switching unit 330 notifies the learning unit 340 of the prediction result data notified by the period determination unit 320. Furthermore, during the prediction period, the switching unit 330 notifies the acquisition unit 132 of the prediction result data notified by the period determination unit 320.

[0069] The learning unit 340 includes a weight calculation unit 341. The weight calculation unit 341 calculates the error of the following data:

[0070] • Sensor data b, measured by sensor b140 and notified by acquisition unit 132 as positive solution data;

[0071] • Prediction result data notified by the switching unit 330.

[0072] Furthermore, the weight calculation unit 341 learns weight parameters based on the calculated error. At this time, in the weight calculation unit 341, the weight parameters are learned through matrix operations to make the error close to zero (i.e., to associate the predicted result data with the correct answer data).

[0073] Additionally, the weight calculation unit 341 stores the learned weight parameters in the data storage unit 350. The data storage unit 350 is implemented, for example, in the auxiliary storage device 203 of the data processing device 130.

[0074] <Reservoir Calculation Flow>

[0075] Next, the process of calculating and processing the storage pool based on RC section 131 will be explained. Figure 4 This is an example of a flowchart illustrating the process of calculating and processing the reservoir.

[0076] In step S401, when the RC unit 131 begins learning, the weight acquisition unit 311 of the prediction unit 310 initializes the weight parameters and sets the default weight parameters to the calculation unit 312. Meanwhile, the reservoir feature generation unit 300 begins inputting time-series sensor data measured by the sensor a120 and outputting reservoir feature quantities. Consequently, the calculation unit 312 of the prediction unit 310 calculates the output reservoir feature quantities based on the weight parameters and outputs the prediction result data.

[0077] In step S402, the period determination unit 320 determines whether the period is a learning period or a prediction period. When it is determined in step S402 that it is a learning period, the switching unit 330 switches the notification destination of the prediction result data to the learning unit 340 and proceeds to step S403. Furthermore, in this embodiment, the RC unit 131 starts the period from the learning period.

[0078] In step S403, the learning unit 340 performs learning processing to learn weight parameters based on the prediction result data and the correct solution data notified by the switching unit 330.

[0079] In step S404, the period determination unit 320 determines whether the learning termination condition is met (whether the state of error between no correct solution data and prediction result data has elapsed for a specified amount of time). If it is determined in step S403 that the learning termination condition is not met, then in step S404 it is determined not to enter the prediction period (determined as "No" in step S404), and proceeds to step S409.

[0080] On the other hand, if it is determined in step S403 that the learning termination condition is met, then it is determined in step S404 that the prediction period will begin (the determination is "yes" in step S404), and the process proceeds to step S405. In this case, the weight acquisition unit 311 of the prediction unit 310 reads the weight parameters at the end of the learning period from the data storage unit 350 and sets them in the calculation unit 312.

[0081] In step S405, the determination unit 320 enters the prediction period and proceeds to step S409. Thereafter, the switching unit 330 switches the notification destination of the prediction result data to the acquisition unit 132.

[0082] On the other hand, when it is determined in step S402 that it is the prediction period, step S406 is entered.

[0083] In step S406, the calculation unit 312 of the prediction unit 310 calculates the reserve pool feature quantity output from the reserve pool feature quantity generation unit 300 according to the weight parameters at the end of learning, and performs prediction processing on the output prediction result data.

[0084] In step S407, the determination unit 320 determines whether relearning is needed. If, in step S407, it is determined that relearning is not needed (if "No" is determined in step S407), the process proceeds to step S409. On the other hand, if, in step S407, it is determined that relearning is needed (if "Yes" is determined in step S407), the process proceeds to step S408.

[0085] In step S408, the determination unit 320 enters the learning period and proceeds to step S409. Thereafter, the switching unit 330 switches the notification destination of the prediction result data to the learning unit 340.

[0086] In step S409, the RC unit 131 determines whether to end the reservoir calculation process.

[0087] In step S409, if it is determined that the reserve pool calculation process should continue (if "No" is determined in step S409), the process returns to step S402. On the other hand, in step S409, if it is determined that the reserve pool calculation process should end (if "Yes" is determined in step S409), the reserve pool calculation process ends.

[0088] <Functional Structure of the Acquisition and Determination Units of the Data Processing Device>

[0089] Next, the functional structure of the acquisition unit 132 and the determination unit 133 of the data processing device 130 will be explained. Figure 5 This diagram illustrates an example of the functional structure of the acquisition unit and the determination unit of a data processing device. For example... Figure 5As shown, the acquisition unit 132 includes a sensor data acquisition unit 501 and a prediction result data acquisition unit 503. Additionally, the determination unit 133 includes a difference calculation unit 511 and a state determination unit 512.

[0090] The sensor data acquisition unit 501 notifies the RC unit 131 of the time series sensor data b measured by the sensor b140 as positive solution data, and notifies the determination unit 133 of the comparison data.

[0091] The prediction result data acquisition unit 503 acquires the prediction result data sent from the RC unit 131 during the prediction period and notifies the determination unit 133.

[0092] The differential calculation unit 511 calculates the difference between the comparison data notified by the sensor data acquisition unit 501 during the prediction period and the prediction result data sent by the RC unit 131 during the prediction period, and generates differential data. In addition, the differential calculation unit 511 notifies the state determination unit 512 of the generated differential data.

[0093] The status determination unit 512 determines the status of the substrate manufacturing process. Specifically, the status determination unit 512 determines whether the differential data notified by the differential calculation unit 511 exceeds a predetermined threshold. If the status determination unit 512 determines that the predetermined threshold has been exceeded, it determines that an anomaly has occurred in the processed substrate. In addition, the status determination unit 512 outputs an anomaly detection result indicating that an anomaly has occurred.

[0094] Furthermore, the status determination unit 512 determines, based on the differential data notified by the differential calculation unit 511, whether the change pattern within a specified time range is similar to a specified change pattern. If the status determination unit 512 determines that the change pattern is similar to a specified change pattern, it determines that an abnormality or malfunction has occurred in the substrate processing apparatus 100. Additionally, the status determination unit 512 outputs a fault diagnosis result indicating that an abnormality or malfunction has occurred in the substrate processing apparatus 100 to the control device 150. Therefore, in the event of an abnormality or malfunction, the control device 150 can issue a stop command to the actuator 160.

[0095] <Process for Acquisition and Judgment>

[0096] Next, the process of acquiring and determining the data processing performed by the acquisition unit 132 and the determination unit 133 of the data processing device 130 will be described as the process of acquiring and determining the data processing when outputting the abnormality detection result. Figure 6 This is the first flowchart representing the process of obtaining and determining the process.

[0097] In step S601, the acquisition unit 132 of the data processing device 130 acquires sensor data b, which is a time series measured by the sensor b140.

[0098] In step S602, the acquisition unit 132 of the data processing device 130 determines whether the current period is a learning period or a prediction period. If it is determined in step S602 that it is a learning period, then the process proceeds to step S603.

[0099] In step S603, the acquisition unit 132 of the data processing device 130 sends the acquired sensor data b as forward solution data to the RC unit 131.

[0100] On the other hand, when it is determined in step S602 that it is a prediction period, step S604 is entered.

[0101] In step S604, the acquisition unit 132 of the data processing device 130 acquires the prediction result data sent from the RC unit 131.

[0102] In step S605, the determination unit 133 of the data processing device 130 compares the sensor data b obtained from the sensor b140 with the prediction result data obtained from the RC unit 131 as comparison data, and calculates the difference data.

[0103] In step S606, the determination unit 133 of the data processing device 130 determines whether the differential data exceeds a predetermined threshold. If it is determined in step S606 that the predetermined threshold has been exceeded (if "yes" is indicated in step S606), the process proceeds to step S607.

[0104] In step S607, the determination unit 133 of the data processing device 130 determines that an abnormality has occurred in the processed substrate, outputs the abnormality detection result indicating that an abnormality has occurred to the control device 150, and then proceeds to step S608.

[0105] On the other hand, in step S606, if it is determined that the specified threshold has not been exceeded (if it is "no" in step S606), the process proceeds directly to step S608.

[0106] In step S608, the acquisition unit 132 of the data processing device 130 determines whether to terminate the acquisition and determination process. If, in step S608, it is determined that the acquisition and determination process should not be terminated (if "No" is determined in step S608), the process returns to step S602. Conversely, if, in step S608, it is determined that the acquisition and determination process should be terminated (if "Yes" is determined in step S608), the acquisition and determination process is terminated.

[0107] Next, the process of acquiring and determining the data in the data processing unit 132 and the determination unit 133 of the data processing device 130 will be described, as well as the process of acquiring and determining the data when outputting the fault diagnosis result. Figure 7 This is the second flowchart representing the process of obtaining and determining the process.

[0108] and Figure 6 The difference in the first flowchart shown is in steps S701 and S702. Therefore, steps S701 and S702 will be explained here.

[0109] In step S701, the determination unit 133 of the data processing device 130 determines whether the change pattern of the differential data over a specified time range is similar to a specified change pattern. If it is determined in step S701 that the change pattern is similar to the specified change pattern (if "yes" is true in step S701), the process proceeds to step S702.

[0110] In step S702, the determination unit 133 of the data processing device 130 determines that an abnormality or malfunction has occurred in the substrate processing device 100, outputs the fault diagnosis result to the control device 150, and then proceeds to step S608.

[0111] On the other hand, if it is determined in step S702 that the change pattern is not similar to the specified change pattern (if it is "no" in step S702), the process proceeds directly to step S608.

[0112] <Specific Examples of Acquisition and Judgment Processing>

[0113] Next, a specific example of the acquisition and determination processing performed by the acquisition unit 132 and determination unit 133 of the data processing device 130 will be described. Figure 8 This is a diagram illustrating specific examples of acquisition and judgment processing. In Figure 8 In the diagram, the horizontal axis represents time, and the vertical axis represents data values.

[0114] In addition, Figure 8 In the graph, line graph 801 represents the prediction result data sent from RC unit 131, and line graph 802 represents the time series sensor data b (comparison data) measured by sensor b140. Additionally, bar graph 803 represents the difference data calculated based on the difference between the prediction result data and the comparison data.

[0115] exist Figure 8 In the example shown in symbol 810, the difference between the predicted result data and the comparison data exceeds a specified threshold at the timing indicated by symbol 810, which is thus determined to be an anomaly, and the anomaly detection result is output.

[0116] In addition, Figure 8In the case of the example, it indicates that the change pattern of the differential data in the time range indicated by symbol 820 is not similar to the specified change pattern.

[0117] On the other hand, Figure 8 In the case of example, if the variation pattern of the differential data in the time range shown by symbol 830 is similar to the specified variation pattern, it is determined that a fault has occurred in the substrate processing device 100, and a fault diagnosis result is output.

[0118] <Summary>

[0119] As can be seen from the above description, the substrate processing apparatus 100 according to the first embodiment:

[0120] • It has a reservoir characteristic quantity generation unit, which takes in time-series sensor data a (time-series sensor data a measured by sensor a) obtained in the substrate manufacturing process, and outputs reservoir characteristic quantities.

[0121] • During the learning period, weight parameters are learned so that the predicted data obtained by calculating the reservoir feature quantity output from the reservoir feature quantity generation unit based on the weight parameters is correlated with the time series sensor data b measured by sensor b.

[0122] • During the prediction period, by inputting time-series sensor data 'a' obtained in the substrate manufacturing process, the reservoir feature quantity output from the reservoir feature quantity generation unit is calculated based on the learned weight parameters, and the prediction result data is output.

[0123] • Compare the predicted data with the time-series sensor data b obtained during the substrate manufacturing process to determine the status of the substrate manufacturing process.

[0124] Therefore, the substrate processing apparatus 100 according to the first embodiment can effectively perform learning during startup and maintenance. As a result, according to the first embodiment, in the substrate processing apparatus that uses time-series sensor data to determine the state of the substrate manufacturing process, the workload of the learning operation can be reduced.

[0125] [Second Implementation]

[0126] In the first embodiment described above, the case where the RC unit 131 has only one set of reservoir feature quantity generating units was described. However, the number of reservoir feature quantity generating units in the RC unit 131 is not limited to one set, and for example, it may have multiple sets. Hereinafter, the second embodiment will be described focusing on the differences from the first embodiment described above.

[0127] <Functional Structure of the RC Section of the Data Processing Unit>

[0128] First, the functional structure of the RC unit 900 in the data processing device 130 of the substrate processing apparatus 100 according to the second embodiment will be explained. Figure 9 Figure 2 is an example of the functional structure of the RC section of a data processing device.

[0129] and Figure 3 The differences in the functional structure shown in Figure 1 are as follows: Figure 9 In the following cases:

[0130] • The RC section 900 has multiple reserve pool characteristic quantity generation sections 300_1 to 300_L (L is any integer);

[0131] • The time-series sensor data a measured by sensor a120 is input in parallel to multiple reservoir feature generation units 300_1 to 300_L;

[0132] • The reservoir feature quantities output by the multiple reservoir feature quantity generation units 300_1 to 300_L are input in parallel to the calculation unit 312 of the prediction unit 310.

[0133] Therefore, the calculation unit 312 notifies the period determination unit 320 of the prediction result data obtained by weighting the reserve pool feature quantities output by the multiple reserve pool feature quantity generation units 300_1 to 300_L respectively using weight parameters.

[0134] <Summary>

[0135] As can be seen from the above description, the substrate processing apparatus 100 according to the second embodiment:

[0136] • It has multiple reservoir feature quantity generation units that input time-series sensor data a (time-series sensor data a measured by sensor a) obtained in the substrate manufacturing process in parallel, and output reservoir feature quantities respectively.

[0137] • During the learning process, weight parameters are learned so that the predicted data obtained by weighting the reservoir features output from the multiple reservoir feature generation parts using the weight parameters are correlated with the time series sensor data b.

[0138] • During the prediction period, by inputting time-series sensor data 'a' obtained in the substrate manufacturing process in parallel, the reservoir feature quantities output from multiple reservoir feature quantity generation units are weighted and added using the learned weight parameters to output the prediction result data.

[0139] • Compare the predicted data with the time-series sensor data b obtained during the substrate manufacturing process to determine the status of the substrate manufacturing process.

[0140] Therefore, the substrate processing apparatus 100 according to the second embodiment can enjoy the same effect as the first embodiment, and can output prediction result data with higher accuracy.

[0141] [Third Implementation]

[0142] In the first and second embodiments described above, the data processing apparatus 130 was described as having only one set of RC units 131. However, the number of RC units in the data processing apparatus is not limited to one set; for example, it may have multiple sets. Hereinafter, the third embodiment will be described focusing on the differences from the first and second embodiments described above.

[0143] <Functional Structure of the RC Section of the Data Processing Unit>

[0144] First, the functional structure of the RC section of the data processing device 1000 included in the substrate processing apparatus 100 according to the third embodiment will be explained. Figure 10A and Figure 10B Figures 3 and 4 show an example of the functional structure of the RC section of a data processing device.

[0145] and Figure 3 The difference in the functional structure shown in Figure 1 is that, Figure 10A , Figure 10B In the following cases:

[0146] • Multiple RC sections 1010_1 to 1010_L are connected in series;

[0147] • The time-series sensor data measured by sensor a120 is input to the RC unit 1010_1 at the beginning;

[0148] • During the learning period, the prediction result data output by the prediction unit 310 of each RC unit 1010_1 to 1010_L-1 are respectively input into the reserve pool feature quantity generation unit 300_2 to 300_L of the next RC unit;

[0149] • During the learning period, the switching units 1020 of each RC unit 1010_1 to 1010_L-1 are turned ON, and the prediction result data output from the corresponding prediction unit 310 is input into the corresponding learning unit 340 for learning.

[0150] • During the learning period, the switching unit 1021 of the end RC unit 1010_L is switched to the learning unit 340 side, and the prediction result data output from the prediction unit 310 is input into the learning unit 340, thereby learning is performed in the learning unit 340.

[0151] • During the prediction period, the switching units 1020 of each RC unit 1010_1 to 1010_L-1 are turned off, and the prediction result data output from the corresponding prediction unit 310 is input to the reserve pool feature quantity generation units 300_2 to 300_L of the next RC unit.

[0152] • During the prediction period, the switching unit 1021 of the end RC unit 1010_L is switched to the acquisition unit 132 side, and the final prediction result data output from the corresponding prediction unit 310 is output to the acquisition unit 132.

[0153] <Summary>

[0154] As can be seen from the above description, the substrate processing apparatus 100 according to the third embodiment is as follows:

[0155] • It has multiple RC units connected in series. During prediction, time-series sensor data a (time-series sensor data a measured by sensor a) obtained in the substrate manufacturing process is input to the starting RC unit, and the final prediction result data is output from the ending RC unit.

[0156] • During the learning and prediction periods, the prediction result data output by the preceding RC unit is input into the reserve pool characteristic quantity generation unit of each RC unit after the starting point.

[0157] • During the learning period, each RC unit learns its own weight parameters so that the prediction results obtained by calculation based on the weight parameters are correlated with the time series sensor data b measured by sensor b.

[0158] • During the prediction period, each RC unit calculates the reserve pool features output from the reserve pool feature generation unit based on the learned weight parameters, and outputs the prediction result data.

[0159] • The final prediction results are compared with the time-series sensor data b obtained during the substrate manufacturing process to determine the status of the substrate manufacturing process.

[0160] Therefore, the substrate processing apparatus 100 according to the third embodiment can output more accurate prediction result data while enjoying the same effect as the first embodiment.

[0161] [Fourth Implementation]

[0162] In the third embodiment described above, the case where multiple sets of RC units of the data processing device 1000 are connected in series was explained. In contrast, in the fourth embodiment, the case where multiple sets of RC units are connected in parallel will be described. Hereinafter, the fourth embodiment will be described focusing on the differences from the third embodiment described above.

[0163] <Functional Structure of the RC Section and Integration Section of the Data Processing Unit>

[0164] First, the functional structure of the RC section and the integration section of the data processing device 1100 included in the substrate processing device 100 according to the fourth embodiment will be explained. Figure 11 This is a diagram illustrating an example of the functional structure of the RC section and the integration section of a data processing device.

[0165] For Figure 10A , Figure 10B The difference in the functional structures shown is that, in Figure 11 In the following cases:

[0166] Multiple RC sections 1110_1 to 1010_L are connected in parallel;

[0167] • For multiple RC sections 1110_1 to 1010_L, the integration section 1130 is connected in series.

[0168] In addition, Figure 11 In the following cases:

[0169] The integration unit 1130 includes a prediction unit 1140 (an example of an integrated prediction unit), a period determination unit 1150, a switching unit 1160, and a learning unit 1170 (an example of an integrated learning unit).

[0170] The prediction unit 1140 also includes a weight acquisition unit 1141 and a calculation unit 1142. The weight acquisition unit 1141 acquires the weight parameters learned by the learning unit 1170 and stored in the data storage unit 1180, and sets them in the calculation unit 1142.

[0171] • The calculation unit 1142 uses the weight parameters set by the weight acquisition unit 1141 to perform weighted addition on the prediction result data output by the prediction units 1120 of each RC unit 1110_1 to 1110_L, and notifies the period determination unit 1150 of the prediction result data.

[0172] • During the learning period, the switching unit 1160 notifies the learning unit 1170 and the prediction units 1120 of each RC unit 1110_1 to 1110_L of the prediction result data notified by the period determination unit 1150.

[0173] • During the prediction period, the switching unit 1160 notifies the acquisition unit 132 and the prediction units 1120 of each RC unit 1110_1 to 1110_L of the prediction result data notified by the period determination unit 1150 as the final prediction result data.

[0174] • The learning unit 1170 has a weight calculation unit 1171;

[0175] • During the learning process, the weight calculation unit 1171 calculates the error between the sensor data b measured by the sensor b140 and notified as positive solution data by the acquisition unit 132 and the prediction result data notified by the switching unit 1160.

[0176] • During the learning process, the weight calculation unit 1171 learns the weight parameters based on the calculated error and stores the learned weight parameters in the data storage unit 1180.

[0177] In addition, Figure 11 In the following cases:

[0178] • The time-series sensor data measured by sensor a120 is input in parallel to the reservoir characteristic quantity generation units 300_1 to 300_L of each RC unit 1110_1 to 1110_L, and the reservoir characteristic quantities are output respectively.

[0179] • During the learning period, the prediction units 1120 of each RC unit 1110_1 to 1110_L calculate the reserve pool characteristic quantity according to the weight parameters, and correct the prediction result data obtained by calculation according to the prediction result data notified by the integration unit 1130 and output it.

[0180] • During the prediction period, the prediction unit 1120 of each RC unit 1110_1 to 1110_L calculates the reserve pool characteristic quantity according to the weight parameters, corrects the prediction result data obtained by calculation according to the prediction result data notified by the integration unit 1130, and outputs it.

[0181] • During the learning and prediction periods, the prediction result data output by the corresponding prediction units 1120 of the RC units 1110_1 to 1110_L are input into the integration unit 1130.

[0182] • During the learning period, the corresponding switching units 1020 of each RC unit 1110_1 to 1110_L are turned ON, and the prediction result data output from the corresponding prediction unit 1120 is input into the corresponding learning unit 340 for learning.

[0183] <Summary>

[0184] As can be seen from the above description, the substrate processing apparatus 100 according to the fourth embodiment is as follows:

[0185] • It has multiple RC units connected in parallel. Each RC unit receives time-series sensor data a (time-series sensor data a measured by sensor a) obtained in the substrate manufacturing process in parallel, and outputs its respective prediction result data to the integration unit.

[0186] • During the prediction period, the prediction results data output by multiple RC units are integrated in the integration unit according to the weight parameters and output as the final prediction result data.

[0187] • During the learning period, each RC unit learns its own weight parameters so that the prediction results obtained by calculation based on the weight parameters are correlated with the time series sensor data b measured by sensor b.

[0188] • During the learning and prediction periods, the prediction results data integrated by the integration department are input into the prediction departments of each RC department to correct the prediction results data obtained by calculating the reserve pool features based on weight parameters.

[0189] • During the learning process, the integration unit learns the weight parameters so that the integrated prediction data is correlated with the time series sensor data b measured by sensor b.

[0190] • During the prediction period, the final prediction result data is compared with the time series sensor data b obtained in the substrate manufacturing process to determine the status of the substrate manufacturing process.

[0191] Therefore, the substrate processing apparatus 100 according to the third embodiment can output more accurate prediction result data while enjoying the same effect as the first embodiment.

[0192] [Fifth Implementation]

[0193] In the first to fourth embodiments described above, the prediction of sensor data for a time series measured by a single sensor b140 was explained based on sensor data for a time series measured by a single sensor a120. In contrast, in the fifth embodiment, sensor data for a time series measured by a single sensor b140 is predicted based on sensor data for a time series measured by multiple sensors. Hereinafter, the fifth embodiment will be described focusing on its differences from the first embodiment described above.

[0194] <System Structure of Substrate Processing Device>

[0195] First, the system structure of the substrate processing apparatus according to the fifth embodiment will be described. Figure 12 Figure 2 is an example of the system structure of a substrate processing apparatus. (And...) Figure 1 The difference in the system architecture shown is that, Figure 12 In the case where sensors a1120_1 to a3120_3 (another example of the first sensor) are configured instead of sensor a120, the function of RC unit 1210 is different from that of RC unit 131.

[0196] In addition, with Figure 1 same, Figure 12 12a indicates the flow of data within the substrate processing apparatus 1200 during the learning period. Figure 12 12b indicates the flow of data within the substrate processing apparatus 1200 during the prediction period.

[0197] like Figure 12 As shown in 12a and 12b, during the process of processing the substrate in the chamber 110, sensor data of a time series measured by sensors a1120_1 to a3120_3 (another example of the first sensor data) is sent to the RC unit 1210.

[0198] The RC unit 1210 performs reservoir calculations with high performance and fast learning speed. Specifically, the RC unit 1210 predicts the time series sensor data (input data) measured by the sensor a1120_1 to the sensor a3120_3 based on the sensor data (input data) measured by the sensor a1120_1 to the sensor a3120_3.

[0199] Therefore, as Figure 12 As shown in 12a, during the learning period, the RC unit 1210 acquires the time-series sensor data (forward solution data) measured by the sensor b140 via the acquisition unit 132 of the data processing device 130. Then, the RC unit 1210 learns weight parameters so that the prediction result data obtained by calculating the time-series sensor data (input data) measured by the sensor a1120_1 to the sensor a3120_3 based on the weight parameters is correlated with the forward solution data.

[0200] On the other hand, such as Figure 12 As shown in 12b, during the prediction period, the RC unit 1210 calculates the time-series sensor data (input data) measured by sensors a1120_1 to a3120_3 based on the learned weight parameters. Then, the RC unit 1210 outputs the time-series sensor data (prediction result data). Furthermore, the time-series sensor data (prediction result data) output by the RC unit 1210 is notified to the acquisition unit 132 of the data processing device 130.

[0201] <Functional Structure of the RC Section of the Data Processing Unit>

[0202] Next, the functional structure of the RC unit 1210 in the data processing device 130 of the substrate processing apparatus 1200 according to the fifth embodiment will be described. Figure 13 This is an example of the functional structure of the RC section of a data processing device. Figure 5 .

[0203] and Figure 3 The difference in the functional structure shown in Figure 1 is that, Figure 13 In the following circumstances:

[0204] • The time series sensor data a1 to a3 measured by sensors a1120_1 to a3120_3 are input into the storage pool feature quantity generation unit 300.

[0205] Therefore, in the computing unit 312, it is possible to input the reservoir feature quantity based on the time series sensor data a1 to sensor data a3 measured by multiple sensors a1120_1 to a3120_3 respectively, and output the prediction result data.

[0206] <Summary>

[0207] As can be seen from the above description, the substrate processing apparatus 100 according to the fifth embodiment:

[0208] • It has a reservoir feature quantity generation unit, which takes in multiple time series sensor data a1 to a3 obtained in the substrate manufacturing process (multiple time series sensor data a1 to a3 measured by multiple sensors a1 to a3) and outputs reservoir feature quantities.

[0209] • During the learning period, weight parameters are learned so that the predicted data obtained by calculating the reservoir feature quantity output from the reservoir feature quantity generation unit based on the weight parameters is correlated with the time series sensor data b measured by sensor b.

[0210] • During the prediction period, multiple time-series sensor data a1 to a3 obtained in the substrate manufacturing process are input, and the reservoir feature quantity output by the reservoir feature quantity generation unit is calculated according to the learned weight parameters, and the prediction result data is output.

[0211] • Compare the predicted data with the time-series sensor data b obtained during the substrate manufacturing process to determine the status of the substrate manufacturing process.

[0212] Therefore, according to the fifth embodiment, the substrate processing apparatus that uses multiple time-series sensor data to determine the state of the substrate manufacturing process can enjoy the same effect as the first embodiment described above.

[0213] [Sixth Implementation]

[0214] In the above embodiments, it is described how to predict sensor data b, which is a time series measured by sensor b140, based on sensor data of a time series measured by sensor a120 (or sensor a1120_1 to sensor a3120_3).

[0215] In contrast, in the sixth embodiment, sensor data of the time series measured by sensor a120 is predicted based on sensor data of the time series measured by sensor a120. Hereinafter, the sixth embodiment will be described focusing on the differences from the embodiments described above.

[0216] <System Structure of Substrate Processing Device>

[0217] First, the system structure of the substrate processing apparatus according to the sixth embodiment will be described. Figure 14 Figure 3 shows an example of the system structure of a substrate processing apparatus. This is similar to the system structure used in the first embodiment described above. Figure 1 The difference in the described system architecture lies in that, Figure 14 In this case, sensor b140 is not included.

[0218] In addition, with Figure 1 same, Figure 14 14a indicates the flow of data within the substrate processing apparatus 1400 during the learning period. Figure 14 14b indicates the flow of data within the substrate processing apparatus 1400 during the prediction period.

[0219] like Figure 14 As shown in 14a and 14b, during the process of processing the substrate in the chamber 110, the time-series sensor data measured by sensor a120 is sent to the RC unit 131 and the acquisition unit 132. That is, the time-series sensor data measured by sensor a120 is an example of the first sensor data of the time series acquired in the substrate manufacturing process, and another example of the second sensor data.

[0220] The RC unit 131 performs reservoir calculations with high performance and fast learning speed. Specifically, the RC unit 131 predicts the time series sensor data measured by the sensor a120 based on the sensor data (input data) measured by the sensor a120.

[0221] Therefore, as Figure 14 As shown in 14a, during the learning period, the RC unit 131 acquires the time-series sensor data (forward solution data) measured by the sensor a120 via the acquisition unit 132 of the data processing device 130. Then, the RC unit 131 learns the weighting parameter so that the prediction result data obtained by performing calculations on the time-series sensor data (input data) measured by the sensor a120 based on the weighting parameter is correlated with the forward solution data.

[0222] On the other hand, such as Figure 14As shown in 14b, during the prediction period, the RC unit 131 calculates the time-series sensor data (input data) measured by the sensor a120 based on the learned weight parameters. Then, the RC unit 131 outputs the time-series sensor data (prediction result data). Furthermore, the time-series sensor data (prediction result data) output by the RC unit 131 is notified to the acquisition unit 132 of the data processing device 130.

[0223] In addition, such as Figure 14 As shown in 14b, during the prediction period, the acquisition unit 132 notifies the determination unit 133:

[0224] • Time-series sensor data (prediction result data) output by RC unit 131; and

[0225] • Sensor data a (comparison data) of the time series measured by sensor a120.

[0226] Furthermore, the time-series sensor data a measured by sensor a120 and sent to the acquisition unit 132 of the data processing device 130 is called "forward solution data" during the learning period and "comparison data" during the prediction period.

[0227] <Summary>

[0228] As can be seen from the above description, the substrate processing apparatus 100 according to the sixth embodiment is as follows:

[0229] • It has a reservoir characteristic quantity generation unit, which takes in time-series sensor data a (time-series sensor data a measured by sensor a) obtained in the substrate manufacturing process, and outputs reservoir characteristic quantities.

[0230] • During the learning period, weight parameters are learned so that the predicted data obtained by calculating the reservoir feature quantity output by the reservoir feature quantity generation unit based on the weight parameters is correlated with the time series sensor data a measured by sensor a.

[0231] • During the prediction period, by inputting time-series sensor data 'a' obtained in the substrate manufacturing process, the reservoir feature quantity output by the reservoir feature quantity generation unit is calculated based on the learned weight parameters, and the prediction result data is output.

[0232] • Compare the predicted data with the time-series sensor data 'a' obtained during the substrate manufacturing process to determine the status of the substrate manufacturing process.

[0233] Therefore, the substrate processing apparatus 100 according to the sixth embodiment can enjoy the same effects as the first embodiment described above.

[0234] [Other Implementation Methods]

[0235] In the above embodiments, it is explained that the reservoir feature quantity generation unit realizes the function of inputting time-series sensor data obtained in the substrate manufacturing process and outputting reservoir feature quantities.

[0236] However, the function of inputting time-series sensor data obtained in the substrate manufacturing process and outputting reservoir feature quantities can be implemented, for example, by the reservoir feature quantity generator 205.

[0237] Therefore, in this specification, the reservoir feature quantity generation unit includes both functions implemented by the processor 201 executing a data processing program and functions implemented by the reservoir feature quantity generator 205.

[0238] Furthermore, in the above embodiments, the case of implementing the reservoir feature generator 205 by means of an FPGA board or the like has been described. However, when implemented by means of an FPGA board or the like, it can also be configured to include a specific physical reservoir.

[0239] In the above embodiments, the method of predicting sensor data of a time series measured by other sensors (or, sensor data of a time series measured by the present sensor) based on sensor data of a time series measured by one or more sensors has been described. Furthermore, in the above embodiments, the method of determining the state of the substrate manufacturing process based on the difference between the predicted data and the comparison data has been described. However, the method for determining the state of the substrate manufacturing process is not limited to this; for example, it may be configured to directly predict the state of the substrate manufacturing process based on sensor data of a time series measured by one or more sensors. In this case, data representing the state of the substrate manufacturing process is input to the learning unit as positive solution data.

[0240] Furthermore, while the learning method for the learning unit to learn the weight parameters is not mentioned in the above embodiments, the learning method can be arbitrary. For example, the learning unit can also learn using methods such as successive least squares or RLS (Recursive Least Squares). Alternatively, the learning unit can also learn using methods such as Bayesian linear regression, Kalman filter, information filter, or particle filter.

[0241] Furthermore, in the above-described second to fourth embodiments, the case of using multiple reservoir feature quantity generation units 300_1 to 300_L was explained, but the multiple reservoir feature quantity generation units may have the same structure or different structures.

[0242] Furthermore, in the fifth embodiment described above, the case of predicting sensor data b, a time series measured by a single sensor b140, based on sensor data a1 to a3, a time series measured by three sensors a1120_1 to a3120_3. However, the number of sensor data used for prediction is not limited to three; for example, it could be n sensor data measured by n sensors (n is any integer greater than 2) belonging to the first sensor. Additionally, the number of sensor data for the predicted time series is not limited to one; for example, it could be m sensor data measured by m sensors (m is any integer greater than 1) belonging to the second sensor.

[0243] Furthermore, the present invention is not limited to the structures listed in the above embodiments, or combinations thereof with other elements, as shown herein. Modifications can be made in these aspects without departing from the spirit of the invention, and appropriate provisions can be made according to its application.

[0244] This application asserts priority based on Japanese Patent Application No. 2021-101004, filed on June 17, 2021, and incorporates the entire contents of that Japanese Patent Application.

[0245] [Symbol Explanation]

[0246] 100: Substrate processing apparatus

[0247] 110: Chamber

[0248] 120: Sensor a

[0249] 130: Data processing device

[0250] 131: RC Department

[0251] 132: Obtaining the Department

[0252] 133: Judgment Department

[0253] 140: Sensor b

[0254] 150: Control device

[0255] 160: Actuator

[0256] 205: Reservoir Feature Generator

[0257] 300: Reserve Pool Characteristic Quantity Generation Department

[0258] 310: Forecasting Department

[0259] 311: Weight Acquisition Department

[0260] 312: Arithmetic Unit

[0261] 320: Period Judgment Department

[0262] 330: Switching Unit

[0263] 340: Study Department

[0264] 341: Weight Calculation Department

[0265] 501: Sensor Data Acquisition Department

[0266] 503: Prediction Results Data Acquisition Department

[0267] 511: Difference Calculation Department

[0268] 512: Status Determination Unit

[0269] 900: RC Department

[0270] 300_1~300_L: Reserve pool characteristic quantity generation section

[0271] 1000: Data processing device

[0272] 1010_1~1010_L: RC section

[0273] 1020: Switching Unit

[0274] 1021: Switching Unit

[0275] 1100: Data processing device

[0276] 1110_1~1110_L: RC section

[0277] 1120: Forecasting Department

[0278] 1130: Integration Department

[0279] 1140: Forecasting Department

[0280] 1141: Weight Acquisition Section

[0281] 1142: Arithmetic Unit

[0282] 1150: Period Judgment Department

[0283] 1160: Switching Unit

[0284] 1170: Study Department

[0285] 1171: Weight Calculation Department

[0286] 1200: Substrate processing device

[0287] 1210: RC Department

[0288] 1400: Substrate processing apparatus.

Claims

1. A substrate processing apparatus comprising: The reservoir feature quantity generation unit inputs the first sensor data of the time series obtained in the substrate manufacturing process and outputs the reservoir feature quantity; The learning unit learns weight parameters during the learning process so that the prediction result data obtained by calculating the reservoir feature quantity output by the reservoir feature quantity generation unit based on the weight parameters is correlated with the second sensor data, which is the positive solution data, obtained in the time series during the substrate manufacturing process. The period determination unit determines the entry that occurs between the learning period and the prediction period if the termination condition is met. The learning period is the period during which the second sensor data of the time series obtained in the substrate manufacturing process is processed as the forward solution data, and the prediction period is the period during which the second sensor data of the time series obtained in the substrate manufacturing process is processed as comparison data. During the prediction period, the prediction unit calculates the reservoir feature quantity output by the reservoir feature quantity generation unit based on the weight parameters learned by the learning unit, using the first sensor data of the time series obtained in the substrate manufacturing process as input, and outputs the prediction result data. as well as The state determination unit determines the state of the substrate manufacturing process by comparing the prediction result data with second sensor data obtained in the substrate manufacturing process as comparison data during the prediction period.

2. The substrate processing apparatus according to claim 1, wherein, It has multiple reservoir feature quantity generation units, which input first sensor data of time series obtained in the substrate manufacturing process in parallel, and output reservoir feature quantities respectively. The learning unit learns weight parameters so that the prediction result data obtained by weighting the reservoir feature quantities output by the plurality of reservoir feature quantity generation units using the weight parameters is correlated with the second sensor data, which is the forward solution data, obtained in the time series during the substrate manufacturing process. The prediction unit uses weight parameters learned by the learning unit to perform weighted addition on the reservoir feature quantities output by the plurality of reservoir feature quantity generation units respectively, which are obtained by inputting time series first sensor data obtained in parallel during the substrate manufacturing process, and outputs prediction result data.

3. The substrate processing apparatus according to claim 1, wherein, The system comprises multiple reservoir feature quantity generation units connected in series, wherein: an initial reservoir feature quantity generation unit is input with first sensor data of a time series obtained during the substrate manufacturing process; and subsequent reservoir feature quantity generation units are input with prediction result data obtained by calculating the reservoir feature quantities output by the preceding reservoir feature quantity generation unit based on weight parameters. During the learning period, each learning unit corresponding to the plurality of reservoir feature quantity generation units learns corresponding weight parameters, so that each prediction result data obtained by calculating the reservoir feature quantities output by the plurality of reservoir feature quantity generation units based on the weight parameters is correlated with the second sensor data, which serves as the forward solution data, obtained in the time series during the substrate manufacturing process. During the prediction period, each prediction unit corresponding to the plurality of reserve pool feature generation units calculates the reserve pool feature quantities output by the plurality of reserve pool feature generation units based on the weight parameters learned by the corresponding learning unit, and outputs its respective prediction result data. During the prediction period, the state determination unit determines the state of the substrate manufacturing process by comparing the final prediction result data output by the prediction unit corresponding to the end reservoir feature quantity generation unit with the second sensor data of the time series obtained in the substrate manufacturing process as comparison data.

4. The substrate processing apparatus according to claim 1, wherein, It has: multiple reservoir feature quantity generation units that input first sensor data of time series obtained in the substrate manufacturing process in parallel, and output reservoir feature quantities respectively; During the learning period, each of the multiple learning units learns a corresponding weight parameter so that the prediction result data obtained by calculating the reservoir feature quantities output by the multiple reservoir feature quantity generation units based on the corresponding weight parameter is correlated with the second sensor data of the time series obtained in the substrate manufacturing process, which is the positive solution data. The multiple prediction units calculate the reserve pool feature quantities output by the multiple reserve pool feature quantity generation units based on the weight parameters learned by the corresponding learning units, and output the prediction result data respectively. The integrated learning unit learns the weighting parameters used in the weighted addition during the learning process, such that the final prediction result data obtained by performing the weighted addition on the prediction result data output by the multiple prediction units using the weighting parameters is correlated with the time-series second sensor data obtained during the substrate manufacturing process, which serves as the positive solution data. The integrated prediction unit, during the prediction period, uses weight parameters learned by the integrated learning unit to perform a weighted summation of the prediction result data output by multiple prediction units, and outputs the final prediction result data. During the learning period and the prediction period, the prediction result data output by the multiple prediction units are prediction result data obtained by correcting the prediction result data obtained by calculating the reserve pool feature quantities output by the multiple reserve pool feature quantity generation units based on corresponding weight parameters using the final prediction result data.

5. The substrate processing apparatus according to claim 1, wherein, It has a reservoir feature quantity generation unit that receives multiple time-series first sensor data obtained during the substrate manufacturing process and outputs reservoir feature quantities. The learning unit learns weight parameters so that the prediction result data obtained by calculating the reservoir feature quantity output by the reservoir feature quantity generation unit based on the weight parameters is correlated with the second sensor data, which serves as the forward solution data, obtained in the time series during the substrate manufacturing process. The prediction unit calculates the reservoir feature quantity output by the reservoir feature quantity generation unit based on the weight parameters learned by the learning unit by inputting the first sensor data of the multiple time series, and outputs the prediction result data.

6. The substrate processing apparatus according to claim 1, wherein, The first sensor data, which is a time series obtained during the substrate manufacturing process and input into the feature quantity generation unit of the reservoir, is sensor data of a time series measured by the first sensor. The second sensor data, which is a time series obtained during the substrate manufacturing process and used for learning by the learning unit, is sensor data of a time series measured by a second sensor that is different from the first sensor.

7. The substrate processing apparatus according to claim 6, wherein, The first sensor data, obtained in the substrate manufacturing process, consists of n time-series sensor data measured by n sensors belonging to the first sensor, where n is an integer greater than or equal to 2. The second sensor data, which is obtained in the substrate manufacturing process and used for learning by the learning unit, is a time series of sensor data measured by m sensors belonging to the second sensor, which is different from the first sensor, where m is an integer greater than or equal to 1.

8. The substrate processing apparatus according to claim 1, wherein, The first sensor data, obtained in the substrate manufacturing process, is a time-series sensor data measured by the first sensor. The second sensor data, which is obtained during the substrate manufacturing process and used for learning by the learning unit, is the time-series sensor data measured by the first sensor.

9. The substrate processing apparatus according to claim 6, wherein, It also includes: a difference calculation unit that calculates the difference between the time series data of the second sensor measured by the second sensor and the prediction result data output by the prediction unit; and The state determination unit determines the state of the substrate manufacturing process based on the differential data calculated by the differential calculation unit.

10. The substrate processing apparatus according to claim 8, wherein, It also includes: a difference calculation unit that calculates the difference between the time series data of the first sensor measured by the first sensor and the prediction result data output by the prediction unit; and The state determination unit determines the state of the substrate manufacturing process based on the differential data calculated by the differential calculation unit.

11. The substrate processing apparatus according to claim 9, wherein, The state determination unit determines the state of the substrate manufacturing process by determining whether the differential data exceeds a predetermined threshold, or by determining whether the differential data within a predetermined time range is similar to a predetermined pattern.

12. The substrate processing apparatus according to claim 11, wherein, The actuator is controlled based on the determination result of the state of the substrate manufacturing process.

13. The substrate processing apparatus according to claim 1, wherein, If a specified amount of time has elapsed in the state where there is no error between the predicted result data and the correct solution data, the period determination unit determines that the termination condition is met and determines to proceed from the learning period to the prediction period.

14. The substrate processing apparatus according to claim 1, wherein, It also includes: a switching unit that switches the notification destination of the prediction result data, so that the prediction result data is processed in the learning unit during the learning period, and the prediction result data is processed in the state determination unit during the prediction period.

15. The substrate processing apparatus according to claim 1, wherein, The period determination unit determines whether relearning is needed during the prediction period. If it determines that relearning is needed, it determines that the termination condition is met and determines that the transition from the prediction period to the learning period is initiated.

16. A data processing method in a substrate processing apparatus, comprising: The reservoir feature quantity generation process inputs the first sensor data of the time series obtained in the substrate manufacturing process and outputs the reservoir feature quantity; In the learning process, during the learning period, weight parameters are learned so that the predicted result data obtained by calculating the reservoir feature quantity output in the reservoir feature quantity generation process based on the weight parameters is correlated with the second sensor data of the time series obtained in the substrate manufacturing process, which is the positive solution data. The process of determining the period involves determining whether an entry occurs between the learning period and the prediction period if the termination condition is met. The learning period is the period during which the second sensor data of the time series obtained in the substrate manufacturing process is processed as the positive solution data, and the prediction period is the period during which the second sensor data of the time series obtained in the substrate manufacturing process is processed as comparison data. In the prediction process, during the prediction period, based on the weight parameters learned in the learning process, the reservoir feature quantity output in the reservoir feature quantity generation process is calculated by inputting the first sensor data of the time series obtained in the substrate manufacturing process, and the prediction result data is output. as well as In the state determination process, during the prediction period, the state of the substrate manufacturing process is determined by comparing the prediction result data with the time series data of the second sensor obtained in the substrate manufacturing process as comparison data.

17. A program product comprising a data processing program that causes a computer of a substrate processing apparatus to perform the following steps: The reservoir feature quantity generation process inputs the first sensor data of the time series obtained in the substrate manufacturing process and outputs the reservoir feature quantity; In the learning process, during the learning period, weight parameters are learned so that the predicted result data obtained by calculating the reservoir feature quantity output in the reservoir feature quantity generation process based on the weight parameters is correlated with the second sensor data of the time series obtained in the substrate manufacturing process, which is the positive solution data. The process of determining the period involves determining whether an entry occurs between the learning period and the prediction period if the termination condition is met. The learning period is the period during which the second sensor data of the time series obtained in the substrate manufacturing process is processed as the positive solution data, and the prediction period is the period during which the second sensor data of the time series obtained in the substrate manufacturing process is processed as comparison data. In the prediction process, during the prediction period, based on the weight parameters learned in the learning process, the reservoir feature quantity output in the reservoir feature quantity generation process is calculated by inputting the first sensor data of the time series obtained in the substrate manufacturing process, and the prediction result data is output. as well as In the state determination process, during the prediction period, the state of the substrate manufacturing process is determined by comparing the prediction result data with the time series data of the second sensor obtained in the substrate manufacturing process as comparison data.

Images