Substrate processing apparatus, data processing method, and data processing program

Abstract

The present invention reduces a workload on a learning task. A substrate processing apparatus includes: a reservoir feature amount generation unit that inputs first sensor data of acquired time series and outputs a reservoir feature amount; a learning unit that in a learning period, learns weighting parameters so that prediction result data acquired by calculating the reservoir feature amount under the weighting parameters correlates with second sensor data of the acquired time series; a prediction unit that in a prediction period, receives input of the first sensor data of the time series to calculate the reservoir feature amount output by the reservoir feature amount generation unit under the learned weighting parameters to thereby output the prediction result data; and a determination unit that in the prediction period, compares the prediction result data and the second sensor data of the acquired time series to thereby determine a state of a substrate manufacturing process.

Description

[Technical Field]

[0001] The present disclosure relates to a substrate processing apparatus, a data processing method, and a data processing program. [Background technology]

[0002] In the field of substrate processing equipment, the state of a substrate manufacturing process (e.g., the presence or absence of an abnormality) has been determined using time-series sensor data measured by various sensors. Recently, the use of machine learning models has also been proposed to improve the accuracy of determination. [Prior art documents] [Patent documents]

[0003] [Patent Document 1] Japanese Patent Application Publication No. 2018-77779 [Non-patent literature]

[0004] [Non-Patent Document 1] David Sussillo, LFAbbott, “Generating Coherent Patterns of Activity from Chaotic Neural Networks”, Neuron. Author manuscript; available in PMC 2010 Aug 27. Summary of the Invention [Problem to be solved by the invention]

[0005] However, in the case of a general machine learning model, a large amount of learning data needs to be collected to improve the judgment accuracy. Therefore, in the case of a substrate processing apparatus to which a machine learning model is applied, the workload for the learning work at the time of start-up is high. Furthermore, even after start-up, if the judgment accuracy decreases due to changes over time in the substrate processing apparatus, a large amount of learning data needs to be collected again. Therefore, in the case of a substrate processing apparatus to which a machine learning model is applied, the workload for the learning work at the time of maintenance is also high.

[0006] The present disclosure reduces the workload of learning operations in a substrate processing apparatus that determines the state of a substrate manufacturing process using time-series sensor data. [Means for solving the problem]

[0007] A substrate processing apparatus according to an aspect of the present disclosure has, for example, the following configuration. a reservoir feature generation unit that receives first sensor data in time series acquired in a substrate manufacturing process and outputs reservoir feature data; During the learning period, prediction result data obtained by calculating the reservoir feature output from the reservoir feature generation unit under weight parameters is This is the correct data a learning unit that learns the weight parameters so as to correlate with time-series second sensor data acquired in the substrate manufacturing process; a period determination step of determining, when it is determined that a termination condition is satisfied, to transition between the learning period, which is a period during which time-series second sensor data acquired in the substrate manufacturing process is processed as the correct answer data, and a prediction period, which is a period during which the time-series second sensor data acquired in the substrate manufacturing process is processed as comparison data; The aforementioned a prediction unit that, during a prediction period, receives time-series first sensor data acquired in the substrate manufacturing process, calculates reservoir features output from the reservoir feature generation unit under weight parameters learned by the learning unit, and outputs prediction result data; During the prediction period, the prediction result data; Comparative data The state of the substrate manufacturing process is determined by comparing the second sensor data acquired in the substrate manufacturing process with the time-series second sensor data acquired in the substrate manufacturing process. situation and a determination unit. [Effects of the Invention]

[0008] In a substrate processing apparatus that determines the state of a substrate manufacturing process using time-series sensor data, the workload involved in the learning operation can be reduced. [Brief explanation of the drawings]

[0009] [Figure 1] FIG. 1 is a first diagram showing an example of a system configuration of a substrate processing apparatus. [Figure 2] FIG. 2 is a diagram illustrating an example of a hardware configuration of a data processing device. [Figure 3] FIG. 3 is a first diagram illustrating an example of the functional configuration of the RC unit of the data processing device. [Figure 4] FIG. 4 is a flowchart showing the flow of the reservoir computing process. [Figure 5] FIG. 5 is a diagram illustrating an example of the functional configuration of an acquisition unit and a determination unit of the data processing device. [Figure 6] FIG. 6 is a first flowchart showing the flow of the acquisition and determination process. [Figure 7] FIG. 7 is a second flowchart showing the flow of the acquisition and determination process. [Figure 8] FIG. 8 is a diagram showing a specific example of the acquisition and determination process. [Figure 9] FIG. 9 is a second diagram illustrating an example of the functional configuration of the RC unit of the data processing device. [Figure 10A] FIG. 10A is a third diagram illustrating an example of the functional configuration of the RC unit of the data processing device. [Figure 10B] FIG. 10B is a fourth diagram illustrating an example of the functional configuration of the RC unit of the data processing device. [Figure 11] FIG. 11 is a diagram illustrating an example of the functional configuration of the RC unit and the integration unit of the data processing device. [Figure 12] FIG. 12 is a second diagram showing an example of the system configuration of the substrate processing apparatus. [Figure 13]FIG. 13 is a fifth diagram illustrating an example of the functional configuration of the RC unit of the data processing device. [Figure 14] FIG. 14 is a third diagram showing an example of a system configuration of a substrate processing apparatus. DETAILED DESCRIPTION OF THE INVENTION

[0010] Hereinafter, each embodiment will be described with reference to the accompanying drawings. In this specification and drawings, components having substantially the same functional configurations are designated by the same reference numerals, and redundant description will be omitted.

[0011] [First embodiment] <System configuration of substrate processing equipment> First, the system configuration of a substrate processing apparatus according to a first embodiment will be described. Fig. 1 is a first diagram showing an example of the system configuration of a substrate processing apparatus, and arrows in the diagram indicate data flow. Of these, Fig. 1a shows the data flow within the substrate processing apparatus 100 during a learning period, and Fig. 1b shows the data flow within the substrate processing apparatus 100 during a prediction period.

[0012] As shown in Figures 1a and 1b, the substrate processing apparatus 100 has a chamber 110 for processing a substrate, a sensor a 120 (an example of a first sensor), a data processing apparatus 130, a sensor b 140 (an example of a second sensor), a control apparatus 150, and an actuator 160.

[0013] 1A and 1B, while a substrate is being processed in a chamber 110, time-series sensor data measured by a sensor a 120 is transmitted to a reservoir computing unit (hereinafter, referred to as an RC unit) 131 of a data processing device 130. That is, the time-series sensor data transmitted to the RC unit 131 is an example of time-series first sensor data acquired in a substrate manufacturing process.

[0014] The RC unit 131 realizes reservoir computing with high expressive power and fast learning speed. Specifically, the RC unit 131 predicts time-series sensor data measured by the sensor b 140 based on time-series sensor data (input data) measured by the sensor a 120.

[0015] 1, during the learning period, the RC unit 131 acquires time-series sensor data (ground truth data) measured by the sensor b 140 via the acquisition unit 132 of the data processing device 130. Then, the RC unit 131 learns the weight parameters so that prediction result data obtained by calculating the time-series sensor data (input data) measured by the sensor a 120 under the weight parameters correlates with the ground truth data.

[0016] 1, during the prediction period, the RC unit 131 calculates time-series sensor data (input data) measured by the sensor a 120 based on the learned weight parameters, and outputs time-series sensor data (prediction result data). 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.

[0017] The sensor b 140 transmits time-series sensor data measured during substrate processing in the chamber 110 to the acquisition unit 132 of the data processing device 130. In other words, the time-series sensor data transmitted to the acquisition unit 132 is an example of time-series second sensor data acquired in the substrate manufacturing process.

[0018] As shown in FIG. 1A, the acquisition unit 132 of the data processing device 130 notifies the RC unit 131 of time-series sensor data (correct data) measured by the sensor b 140 during the learning period.

[0019] 1B, 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 b 140 during the prediction period. Note that the time-series sensor data measured by the sensor b 140 is referred to as "correct answer data" during the learning period and as "comparison data" during the prediction period.

[0020] 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 by the determination unit 133 is notified to the control device 150, and the control device 150 controls the actuator 160 in accordance with the notified determination result.

[0021] In this way, in the substrate processing apparatus 100 that determines the state of the substrate manufacturing process using time-series sensor data, in the first embodiment, The RC unit 131 realizes reservoir computing with high expressive power and fast learning speed. During the learning period, learning is performed by using two types of time-series sensor data as input data and correct answer data. During the forecast period, the status of the substrate manufacturing process is determined by using two types of time-series sensor data as input data and comparison data.

[0022] As a result, the substrate processing apparatus 100 according to the first embodiment can efficiently perform learning during start-up and maintenance, thereby reducing the workload of the learning work.

[0023] <Hardware configuration of data processing device> Next, a hardware configuration of the data processing device 130 included in the substrate processing device 100 will be described. Fig. 2 is a diagram showing an example of the hardware configuration of the data processing device. 2a in Fig. 2 shows a case where the reservoir feature generation function (described in detail later) is realized by a reservoir feature generator, which is hardware. 2b in Fig. 2 shows a case where the reservoir feature generation function is realized by a computer executing software (for example, a reservoir feature generation program).

[0024] 2, 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 the case of 2a in FIG. 2, an I / F device 204 and a reservoir feature generator 205 are further included). The hardware components of the data processing device 130 are connected to each other via a bus 207.

[0025] The processor 201 in 2a and 2b in Fig. 2 has various arithmetic devices such as a CPU, a GPU, etc. The processor 201 reads various programs (for example, a data processing program to be described later, in the case of 2b in Fig. 2, a data processing program including a reservoir feature generation program, etc.) into a memory 202 and executes them.

[0026] 2 includes a main storage device such as a ROM, a RAM, etc. The processor 201 and the memory 202 form a so-called computer, and the processor 201 executes various programs read onto the memory 202, whereby the computer realizes various functions (RC unit 131, acquisition unit 132, determination unit 133).

[0027] The auxiliary storage devices 203 denoted by 2a and 2b in FIG. 2 store various programs and various data used when the processor 201 executes the various programs.

[0028] The I / F device 204 shown in 2a of Fig. 2 is a connection device that connects to the reservoir feature generator 205. The reservoir feature generator 205 shown in 2a of Fig. 2 is realized by, for example, an FPGA (Field-Programmable Gate Array) board. The reservoir feature generator 205 has a reservoir feature generation function that inputs time-series sensor data measured by the sensor a120 and outputs reservoir features.

[0029] The external I / F devices 206 denoted by 2a and 2b in FIG. 2 are connection devices that connect to the sensor a 120 and the sensor b 140.

[0030] <Functional configuration of the RC unit of the data processing device> Next, a description will be given of the functional configuration of RC unit 131 of data processing device 130. Fig. 3 is a first diagram showing an example of the functional configuration of the RC unit of the data processing device.

[0031] 3, the RC unit 131 includes a reservoir feature generation unit 300, a prediction unit 310, a period determination unit 320, a switching unit 330, and a learning unit 340. Of these, the reservoir feature generation unit 300 may be realized by the reservoir feature generator 205, which is hardware such as an FPGA board, as described above. However, in this embodiment, the reservoir feature generation unit 300 will be described as a function realized by a reservoir feature generation program being executed by a computer.

[0032] The prediction unit 310 to the learning unit 340 are functions realized by a computer executing a data processing program (a data processing program other than the reservoir feature generation program).

[0033] As shown in FIG. 3, the reservoir feature generating unit 300 has a reservoir feature generating function of inputting time-series sensor data a (input data) acquired from the sensor a 120 and outputting the reservoir feature.

[0034] Prediction unit 310 includes weight acquisition unit 311 and calculation unit 312. Weight acquisition unit 311 acquires weight parameters learned by learning unit 340 and stored in data storage unit 350, and sets the weight parameters in calculation unit 312. Note that in the first learning of a learning period, default weight parameters may be set as initial values ​​in calculation unit 312, or values ​​generated by random numbers, for example, may be set.

[0035] The calculation unit 312 calculates the reservoir feature output from the reservoir feature generation unit 300 based on the weight parameters set by the weight acquisition unit 311, and notifies the period determination unit 320 of the prediction result data.

[0036] 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 (correct data) notified by the acquisition unit 132, and notifies the switching unit 330 of the period determination result. Furthermore, 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.

[0037] 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 acquiring unit 132 of the prediction result data notified by the period determination unit 320.

[0038] The learning unit 340 includes a weight calculation unit 341. The weight calculation unit 341 calculates Sensor data b measured by the sensor b 140 and notified as correct data by the acquisition unit 132; The prediction result data notified by the switching unit 330, The weight calculation unit 341 calculates an error in the weight parameter based on the calculated error. At this time, the weight calculation unit 341 learns the weight parameter by matrix calculation so that the error approaches zero (that is, so that the prediction result data correlates with the correct answer data).

[0039] Furthermore, the weight calculation unit 341 stores the learned weight parameters in the data storage unit 350. The data storage unit 350 is realized in the auxiliary storage device 203 of the data processing device 130, for example.

[0040] <Reservoir computing processing flow> Next, a description will be given of the flow of reservoir computing processing by the RC unit 131. Fig. 4 is an example of a flowchart showing the flow of reservoir computing processing.

[0041] In step S401, when the RC unit 131 starts a learning period, the weight acquisition unit 311 of the prediction unit 310 initializes weight parameters and sets default weight parameters to the calculation unit 312. In addition, the reservoir feature generation unit 300 starts inputting time-series sensor data measured by the sensor a120 and outputting reservoir features. As a result, the calculation unit 312 of the prediction unit 310 calculates the output reservoir features based on the weight parameters and outputs prediction result data.

[0042] In step S402, the period determination unit 320 determines whether the period is a learning period or a prediction period. If it is determined in step S402 that the period is a learning period, the switching unit 330 switches the notification destination of the prediction result data to the learning unit 340, and the process proceeds to step S403. In this embodiment, the RC unit 131 starts the period from the learning period.

[0043] In step S403, the learning unit 340 performs a learning process to learn weight parameters based on the prediction result data notified by the switching unit 330 and the correct answer data.

[0044] In step S404, period determination unit 320 determines whether the learning end condition is met (whether a specified time has passed since there was no error between the correct answer data and the prediction result data). If it is determined in step S403 that the learning end condition is not met, it determines in step S404 not to move to the prediction period (determined as NO in step S404), and proceeds to step S409.

[0045] On the other hand, if it is determined in step S403 that the learning end condition is satisfied, it is determined in step S404 to transition to the prediction period (determined as 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 out the weight parameters at the time of completion of learning from the data storage unit 350 and sets them in the calculation unit 312.

[0046] In step S405, the period determination unit 320 moves to the prediction period, and the process proceeds to step S409. As a result, the switching unit 330 switches the notification destination of the prediction result data to the acquisition unit 132.

[0047] On the other hand, if it is determined in step S402 that it is the prediction period, the process proceeds to step S406.

[0048] In step S406, the calculation unit 312 of the prediction unit 310 performs prediction processing to calculate the reservoir feature output from the reservoir feature generation unit 300 based on the weight parameters at the time of completion of learning, and to output prediction result data.

[0049] In step S407, period determination unit 320 determines whether relearning is necessary. If it is determined in step S407 that relearning is not necessary (NO in step S407), the process proceeds to step S409. On the other hand, if it is determined in step S407 that relearning is necessary (YES in step S407), the process proceeds to step S408.

[0050] In step S408, the period determination unit 320 shifts to the learning period, and the process proceeds to step S409. As a result, the switching unit 330 switches the notification destination of the prediction result data to the learning unit 340.

[0051] In step S409, the RC unit 131 determines whether or not to end the reservoir computing process.

[0052] In step S409, if it is determined that the reservoir computing process is to be continued (NO in step S409), the process returns to step S402. On the other hand, in step S409, if it is determined that the reservoir computing process is to be ended (YES in step S409), the reservoir computing process is ended.

[0053] <Functional configuration of the acquisition unit and determination unit of the data processing device> Next, the functional configurations of the acquisition unit 132 and determination unit 133 of the data processing device 130 will be described. Fig. 5 is a diagram showing an example of the functional configurations of the acquisition unit and determination unit of the data processing device. As shown in Fig. 5, the acquisition unit 132 has a sensor data acquisition unit 501 and a prediction result data acquisition unit 503. Furthermore, the determination unit 133 has a difference calculation unit 511 and a state determination unit 512.

[0054] The sensor data acquisition unit 501 notifies the RC unit 131 of the time-series sensor data b measured by the sensor b 140 as correct answer data, and notifies the determination unit 133 of the time-series sensor data b as comparison data.

[0055] The prediction result data acquisition unit 503 acquires the prediction result data transmitted from the RC unit 131 during the prediction period, and notifies the determination unit 133 of the data.

[0056] The difference 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 transmitted by the RC unit 131 during the prediction period, and generates difference data. In addition, the difference calculation unit 511 notifies the state determination unit 512 of the generated difference data.

[0057] The state determination unit 512 determines the state of the substrate manufacturing process. Specifically, the state determination unit 512 determines whether the difference data notified by the difference calculation unit 511 exceeds a predetermined threshold. If the state determination unit 512 determines that the predetermined threshold has been exceeded, it determines that an abnormality has occurred in the processed substrate. Furthermore, the state determination unit 512 outputs an abnormality detection result indicating that an abnormality has occurred.

[0058] Furthermore, the state determination unit 512 determines whether or not a change pattern in a predetermined time range for the difference data notified by the difference calculation unit 511 resembles a predetermined change pattern. If the state determination unit 512 determines that the change pattern resembles the predetermined change pattern, it determines that an abnormality or a malfunction has occurred in the substrate processing apparatus 100. Furthermore, the state determination unit 512 outputs a malfunction diagnosis result indicating that an abnormality or a malfunction has occurred in the substrate processing apparatus 100 to the control device 150. This allows the control device 150 to notify the actuator 160 of a stop command when an abnormality or a malfunction has occurred.

[0059] <Acquisition and determination process flow> Next, a flow of the acquisition and determination process when an abnormality detection result is output will be described as a flow of the acquisition and determination process by the acquisition unit 132 and the determination unit 133 of the data processing device 130. Fig. 6 is a first flowchart showing the flow of the acquisition and determination process.

[0060] In step S601, the acquisition unit 132 of the data processing device 130 acquires the time-series sensor data b measured by the sensor b140.

[0061] 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 the current period is a learning period, the process proceeds to step S603.

[0062] In step S603, the acquisition unit 132 of the data processing device 130 transmits the acquired sensor data b to the RC unit 131 as correct answer data.

[0063] On the other hand, if it is determined in step S602 that it is the prediction period, the process proceeds to step S604.

[0064] In step S604, the acquisition unit 132 of the data processing device 130 acquires the prediction result data transmitted from the RC unit 131.

[0065] In step S605, the determination unit 133 of the data processing device 130 compares the sensor data b acquired from the sensor b 140 as comparison data with the prediction result data acquired from the RC unit 131, and calculates difference data.

[0066] In step S606, the determination unit 133 of the data processing device 130 determines whether the differential data exceeds a predetermined threshold value. If it is determined in step S606 that the differential data exceeds the predetermined threshold value (YES in step S606), the process proceeds to step S607.

[0067] In step S607, the determination unit 133 of the data processing device 130 determines that an abnormality has occurred in the processed substrate, outputs an abnormality detection result indicating that an abnormality has occurred to the control device 150, and then proceeds to step S608.

[0068] On the other hand, if it is determined in step S606 that the predetermined threshold value is not exceeded (NO in step S606), the process proceeds directly to step S608.

[0069] In step S608, the acquisition unit 132 of the data processing device 130 determines whether or not to end the acquisition and determination process. If it is determined in step S608 that the acquisition and determination process should not be ended (NO in step S608), the process returns to step S602. On the other hand, if it is determined in step S608 that the acquisition and determination process should be ended (YES in step S608), the acquisition and determination process is ended.

[0070] Next, a flow of the acquisition and determination process when outputting a fault diagnosis result will be described as a flow of the acquisition and determination process by the acquisition unit 132 and the determination unit 133 of the data processing device 130. Fig. 7 is a second flowchart showing the flow of the acquisition and determination process.

[0071] The differences from the first flowchart shown in Fig. 6 are steps S701 and S702. Therefore, steps S701 and S702 will be explained here.

[0072] In step S701, the determination unit 133 of the data processing device 130 determines whether the change pattern of the difference data over a predetermined time range is similar to a predetermined change pattern. If it is determined in step S701 that the change pattern is similar to the predetermined change pattern (YES in step S701), the process proceeds to step S702.

[0073] In step S702, the determination unit 133 of the data processing device 130 determines that an abnormality or a failure has occurred in the substrate processing device 100, and outputs the failure diagnosis result to the control device 150, and then proceeds to step S608.

[0074] On the other hand, if it is determined in step S702 that the change pattern is not similar to the predetermined change pattern (NO in step S702), the process proceeds directly to step S608.

[0075] <Specific example of acquisition and determination process> Next, a specific example of the acquisition and determination process by the acquisition unit 132 and determination unit 133 of the data processing device 130 will be described. Fig. 8 is a diagram showing a specific example of the acquisition and determination process. In Fig. 8, the horizontal axis represents time, and the vertical axis represents data values.

[0076] 8, a line graph 801 represents the prediction result data transmitted from the RC unit 131, and a line graph 802 represents the time-series sensor data b (comparison data) measured by the sensor b 140. A bar graph 803 represents difference data calculated from the difference between the prediction result data and the comparison data.

[0077] In the example of Figure 8, at the timing indicated by the symbol 810, the difference between the prediction result data and the comparison data exceeded a predetermined threshold, indicating that an abnormality had occurred and an abnormality detection result was output.

[0078] In the example of FIG. 8, the change pattern of the difference data in the time range indicated by the reference numeral 820 is determined to be not similar to the predetermined change pattern.

[0079] On the other hand, in the example of Figure 8, the change pattern of the differential data in the time range indicated by the symbol 830 is similar to a predetermined change pattern, so it is determined that a malfunction has occurred in the substrate processing apparatus 100, and a malfunction diagnosis result is output.

[0080] <Summary> As is clear from the above description, the substrate processing apparatus 100 according to the first embodiment has the following features: The reservoir feature generator inputs time-series sensor data a (time-series sensor data a measured by sensor a) acquired in the substrate manufacturing process and outputs reservoir feature data. During the learning period, the weight parameters are learned so that the prediction result data obtained by calculating the reservoir feature quantity output from the reservoir feature quantity generation unit under the weight parameters correlates with the time-series sensor data b measured by sensor b. During the prediction period, time-series sensor data a acquired in the substrate manufacturing process is input, and the reservoir feature output from the reservoir feature generation unit is calculated based on the learned weight parameters, and prediction result data is output. The predicted result data is compared with time-series sensor data b acquired in the substrate manufacturing process to determine the state of the substrate manufacturing process.

[0081] As a result, the substrate processing apparatus 100 according to the first embodiment can efficiently perform learning during start-up and maintenance. As a result, according to the first embodiment, the workload of the learning operation can be reduced in the substrate processing apparatus that determines the state of the substrate manufacturing process using time-series sensor data.

[0082] [Second embodiment] In the first embodiment, the RC unit 131 has only one set of reservoir feature generators. However, the number of reservoir feature generators included in the RC unit 131 is not limited to one, and the RC unit 131 may have multiple sets, for example. The second embodiment will be described below, focusing on the differences from the first embodiment.

[0083] <Functional configuration of the RC unit of the data processing device> First, a description will be given of the functional configuration of the RC unit 900 in the data processing device 130 of the substrate processing apparatus 100 according to the second embodiment. Fig. 9 is a second diagram showing an example of the functional configuration of the RC unit of the data processing device.

[0084] The difference between the functional configuration of the first diagram shown in Fig. 3 and that of Fig. 9 is that The RC unit 900 has a plurality of reservoir feature generation units 300_1 to 300_L (L is an arbitrary integer), The time-series sensor data a measured by the sensor a120 is input in parallel to the plurality of reservoir feature generators 300_1 to 300_L. The reservoir features output from the plurality of reservoir feature generation units 300_1 to 300_L are input in parallel to the calculation unit 312 of the prediction unit 310. is.

[0085] As a result, the calculation unit 312 notifies the period determination unit 320 of prediction result data obtained by weighting and adding the reservoir features output from the plurality of reservoir feature generation units 300_1 to 300_L using the weight parameters.

[0086] <Summary> As is clear from the above description, the substrate processing apparatus 100 according to the second embodiment has the following features: The reservoir feature generator has a plurality of reservoir feature generators, which input time-series sensor data a (time-series sensor data a measured by sensor a) acquired in the substrate manufacturing process in parallel and output reservoir feature values ​​respectively. During the learning period, the weight parameters are learned so that the prediction result data obtained by weighting and adding the reservoir features output from the multiple reservoir feature generators using the weight parameters correlates with the time-series sensor data b. During the prediction period, time-series sensor data a acquired in the substrate manufacturing process is input in parallel, and the reservoir features output from multiple reservoir feature generators are weighted and added using the learned weight parameters, and prediction result data is output. The predicted result data is compared with time-series sensor data b acquired in the substrate manufacturing process to determine the state of the substrate manufacturing process.

[0087] As a result, the substrate processing apparatus 100 according to the second embodiment can output more accurate prediction result data while enjoying the same effects as those of the first embodiment.

[0088] [Third embodiment] In the above first and second embodiments, the case where the data processing device 130 has only one set of RC unit 131 has been described. However, the number of RC units that the data processing device has is not limited to one set, and for example, the data processing device may have multiple sets. Below, the third embodiment will be described, focusing on the differences from the above first and second embodiments.

[0089] <Functional configuration of the RC unit of the data processing device> First, a functional configuration of an RC unit of a data processing apparatus 1000 included in a substrate processing apparatus 100 according to the third embodiment will be described. Figures 10A and 10B are third and fourth diagrams showing an example of the functional configuration of the RC unit of the data processing apparatus.

[0090] The difference between the functional configuration of the first diagram shown in FIG. 3 and that of FIG. 10A and FIG. 10B is that A plurality of RC units 1010_1 to 1010_L are connected in series. The time-series sensor data measured by the sensor a120 is input to the RC unit 1010_1 at the start. During the learning period, the prediction result data output from the prediction unit 310 of each RC unit 1010_1 to 1010_L-1 is input to the reservoir feature generation units 300_2 to 300_L of the next RC unit, respectively. During the learning period, the switching units 1020 of the RC units 1010_1 to 1010_L-1 are turned on, and the prediction result data output from the corresponding prediction unit 310 is input to the corresponding learning unit 340, and learning is performed in the learning unit 340. During the learning period, the switching unit 1021 of the terminal 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 to the learning unit 340, whereby learning is performed in the learning unit 340. During the prediction period, the switching units 1020 of the RC units 1010_1 to 1010_L-1 are each turned off, and the prediction result data output from the corresponding prediction unit 310 is input to the reservoir feature generation units 300_2 to 300_L of the next RC unit, respectively. During the prediction period, the switching unit 1021 of the terminal RC unit 1010_L is switched to the side of the acquisition unit 132, and the final prediction result data output from the corresponding prediction unit 310 is output to the acquisition unit 132. is.

[0091] <Summary> As is clear from the above description, the substrate processing apparatus 100 according to the third embodiment has the following features: It has multiple RC units connected in series. During the prediction period, time-series sensor data a acquired in the board manufacturing process (time-series sensor data a measured by sensor a) is input to the RC unit at the beginning, and the final prediction result data is output from the RC unit at the end. During the learning period and the prediction period, the reservoir feature generation unit of each RC unit after the starting point receives the prediction result data output from the previous RC unit. During the learning period, each RC unit learns its weight parameters so that the prediction result data obtained by calculation under the weight parameters correlates with the time-series sensor data b measured by sensor b. During the prediction period, each RC unit calculates the reservoir features output from the reservoir feature generation unit based on the learned weight parameters and outputs the prediction result data. The final prediction result data is compared with the time-series sensor data b acquired during the substrate manufacturing process to determine the state of the substrate manufacturing process.

[0092] As a result, the substrate processing apparatus 100 according to the third embodiment can output more accurate prediction result data while enjoying the same effects as those of the first embodiment.

[0093] [Fourth embodiment] In the third embodiment, a case where multiple sets of RC units included in the data processing device 1000 are connected in series has been described. In contrast, in the fourth embodiment, a case where multiple sets of RC units are connected in parallel will be described. The fourth embodiment will be described below, focusing on the differences from the third embodiment.

[0094] <Functional configuration of the RC unit and integration unit of the data processing device> First, a description will be given of the functional configuration of the RC unit and the integration unit of the data processing apparatus 1100 included in the substrate processing apparatus 100 according to the fourth embodiment. Fig. 11 is a diagram showing an example of the functional configuration of the RC unit and the integration unit of the data processing apparatus.

[0095] The difference between the functional configuration shown in FIG. 10A and FIG. 10B is that in the case of FIG. 11, A plurality of RC units 1110_1 to 1010_L are connected in parallel, The integrating unit 1130 is connected in series to the multiple RC units 1110_1 to 1110_L. In addition, in the case of Figure 11, 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), The prediction unit 1140 further includes a weight acquisition unit 1141 and a calculation unit 1142, and the weight acquisition unit 1141 acquires weight parameters learned by the learning unit 1170 and stored in the data storage unit 1180, and sets the weight parameters in the calculation unit 1142. The calculation unit 1142 performs weighted addition on the prediction result data output from the prediction units 1120 of the RC units 1110_1 to 1110_L using the weight parameters set by the weight acquisition unit 1141, and notifies the period determination unit 1150 of the prediction result data. During the learning period, the switching unit 1160 notifies the learning unit 1170 and the prediction unit 1120 of each of the RC units 1110_1 to 1110_L of the prediction result data notified by the period determination unit 1150. During the prediction period, the switching unit 1160 notifies the acquisition unit 132 and the prediction units 1120 of the RC units 1110_1 to 1110_L of the prediction result data notified by the period determination unit 1150 as the final prediction result data. The learning unit 1170 includes a weight calculation unit 1171. During the learning period, the weight calculation unit 1171 calculates the error between the sensor data b measured by the sensor b 140 and notified as correct data by the acquisition unit 132 and the prediction result data notified by the switching unit 1160. During the learning period, the weight calculation unit 1171 learns weight parameters based on the calculated errors, and the weight parameters obtained by learning are stored in the data storage unit 1180. In addition, in the case of Figure 11, The time-series sensor data measured by the sensor a120 is input in parallel to the reservoir feature generation units 300_1 to 300_L of the RC units 1110_1 to 1110_L, and the reservoir features are output respectively. During the learning period, the prediction unit 1120 corresponding to each of the RC units 1110_1 to 1110_L calculates reservoir features based on weight parameters, corrects the prediction result data obtained by the calculation based on the prediction result data notified by the integration unit 1130, and outputs the corrected data. During the prediction period, the corresponding prediction unit 1120 of each of the RC units 1110_1 to 1110_L calculates reservoir features based on weight parameters, corrects the prediction result data obtained by the calculation based on the prediction result data notified by the integration unit 1130, and outputs the corrected data. During the learning period and the prediction period, the prediction result data output from the corresponding prediction unit 1120 of each of the RC units 1110_1 to 1110_L is input to the integration unit 1130. During the learning period, the corresponding switching units 1020 of the RC units 1110_1 to 1110_L are turned on, and the prediction result data output from the corresponding prediction units 1120 is input to the corresponding learning units 340, and learning is performed in the learning units 340. is.

[0096] <Summary> As is clear from the above description, the substrate processing apparatus 100 according to the fourth embodiment has the following features: The system has multiple RC units connected in parallel. Each RC unit receives time-series sensor data a acquired in the substrate manufacturing process (time-series sensor data a measured by sensor a) in parallel and outputs the respective prediction result data to the integrating unit. During the prediction period, the prediction result data output from each of the multiple RC units is integrated in the integration unit using weight parameters and output as the final prediction result data. During the learning period, each RC unit learns its weight parameters so that each prediction result data obtained by calculation under the weight parameters correlates with the time-series sensor data b measured by sensor b. During the learning and prediction periods, the prediction result data integrated by the integration unit is input to the prediction unit of each RC unit, and the prediction result data obtained by calculating the reservoir feature values ​​based on the weight parameters is corrected. During the learning period, the integration unit learns weight parameters so that the integrated prediction result data correlates with the time-series sensor data b measured by sensor b. During the prediction period, the final prediction result data is compared with the time-series sensor data b acquired in the substrate manufacturing process to determine the state of the substrate manufacturing process.

[0097] As a result, the substrate processing apparatus 100 according to the third embodiment can output more accurate prediction result data while enjoying the same effects as those of the first embodiment.

[0098] [Fifth embodiment] In the above first to fourth embodiments, cases have been described in which time-series sensor data measured by one sensor b 140 is predicted based on time-series sensor data measured by one sensor a 120. In contrast, in the fifth embodiment, time-series sensor data measured by one sensor b 140 is predicted based on time-series sensor data measured by multiple sensors. The fifth embodiment will be described below, focusing on the differences from the above first embodiment.

[0099] <System configuration of substrate processing equipment> First, a system configuration of a substrate processing apparatus according to a fifth embodiment will be described. Fig. 12 is a second diagram showing an example of the system configuration of the substrate processing apparatus. The system configuration in Fig. 1 differs from that shown in Fig. 1 in that sensors a1120_1 to a3120_3 (another example of the first sensor) are provided instead of the sensor a120, and that the function of the RC unit 1210 is different from the function of the RC unit 131.

[0100] 1, 12a in FIG. 12 shows the data flow in the substrate processing apparatus 1200 during the learning period, and 12b in FIG. 12 shows the data flow in the substrate processing apparatus 1200 during the prediction period.

[0101] As shown in 12a and 12b in FIG. 12, while the substrate is being processed in the chamber 110, the time-series sensor data measured by the sensors a1120_1 to a3120_3 (another example of the first sensor data) is transmitted to the RC unit 1210.

[0102] The RC unit 1210 realizes reservoir computing with high expressive power and fast learning speed. Specifically, the RC unit 1210 predicts time-series sensor data measured by the sensor b 140 based on time-series sensor data (input data) measured by the sensors a 1120_1 to a 3120_3.

[0103] 12, during the learning period, the RC unit 1210 acquires time-series sensor data (ground truth data) measured by the sensor b 140 via the acquisition unit 132 of the data processing device 130. Then, the RC unit 1210 learns the weight parameters so that prediction result data obtained by calculating the time-series sensor data (input data) measured by the sensors a1120_1 to a3120_3 under the weight parameters correlates with the ground truth data.

[0104] 12, during the prediction period, the RC unit 1210 calculates time-series sensor data (input data) measured by the sensors a1120_1 to a3120_3 under the learned weight parameters. As a result, the RC unit 1210 outputs time-series sensor data (prediction result data). 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.

[0105] <Functional configuration of the RC unit of the data processing device> Next, a description will be given of the functional configuration of the RC unit 1210 in the data processing device 130 of the substrate processing device 1200 according to the fifth embodiment. Fig. 13 is a fifth diagram showing an example of the functional configuration of the RC unit of the data processing device.

[0106] The difference between the functional configuration of the first diagram shown in Fig. 3 and that of Fig. 13 is that The time-series sensor data a1 to a3 measured by the sensors a1 to a3 are input to the reservoir feature generator 300. is.

[0107] As a result, the calculation unit 312 can input reservoir feature amounts based on the time-series sensor data a1 to a3 measured by the plurality of sensors a1120_1 to a3120_3, respectively, and output prediction result data.

[0108] <Summary> As is clear from the above description, the substrate processing apparatus 100 according to the fifth embodiment has the following features: The reservoir feature generator inputs a plurality of time-series sensor data a1 to a3 (a plurality of time-series sensor data a1 to a3 measured by a plurality of sensors a1 to a3) acquired in the substrate manufacturing process, and outputs reservoir features. During the learning period, the weight parameters are learned so that the prediction result data obtained by calculating the reservoir feature quantity output from the reservoir feature quantity generation unit under the weight parameters correlates with the time-series sensor data b measured by sensor b. During the prediction period, multiple time-series sensor data a1 to a3 acquired in the substrate manufacturing process are input, and the reservoir feature output from the reservoir feature generation unit is calculated based on the learned weight parameters, and prediction result data is output. The predicted result data is compared with time-series sensor data b acquired in the substrate manufacturing process to determine the state of the substrate manufacturing process.

[0109] As a result, according to the fifth embodiment, in a substrate processing apparatus that determines the state of a substrate manufacturing process using a plurality of time-series sensor data, it is possible to obtain the same effects as those of the first embodiment.

[0110] [Sixth embodiment] In each of the above embodiments, the case where the time-series sensor data b measured by the sensor b 140 is predicted based on the time-series sensor data measured by the sensor a 120 (or the sensors a 1120_1 to a 3120_3) has been described.

[0111] In contrast to this, in the sixth embodiment, time-series sensor data measured by the sensor a 120 is predicted based on the time-series sensor data measured by the sensor a 120. The sixth embodiment will be described below, focusing on the differences from the above-mentioned embodiments.

[0112] <System configuration of substrate processing equipment> First, a system configuration of a substrate processing apparatus according to a sixth embodiment will be described. Fig. 14 is a third diagram showing an example of the system configuration of a substrate processing apparatus. The difference from the system configuration described in Fig. 1 in the first embodiment is that the sensor b140 is not included in Fig. 14.

[0113] 1, 14a in FIG. 14 shows the data flow in the substrate processing apparatus 1400 during the learning period, and 14b in FIG. 14 shows the data flow in the substrate processing apparatus 1400 during the prediction period.

[0114] 14, during substrate processing in the chamber 110, the time-series sensor data measured by the sensor a 120 is transmitted to the RC unit 131 and the acquisition unit 132. In other words, the time-series sensor data measured by the sensor a 120 is an example of first time-series sensor data acquired in the substrate manufacturing process, and is another example of second sensor data.

[0115] The RC unit 131 realizes reservoir computing with high expressive power and fast learning speed. Specifically, the RC unit 131 predicts time-series sensor data measured by the sensor a120 based on the time-series sensor data (input data) measured by the sensor a120.

[0116] 14, during the learning period, the RC unit 131 acquires time-series sensor data (ground truth data) measured by the sensor a 120 via the acquisition unit 132 of the data processing device 130. Then, the RC unit 131 learns the weight parameters so that prediction result data obtained by calculating the time-series sensor data (input data) measured by the sensor a 120 under the weight parameters correlates with the ground truth data.

[0117] 14, during the prediction period, the RC unit 131 calculates time-series sensor data (input data) measured by the sensor a 120 based on the learned weight parameters. As a result, the RC unit 131 outputs time-series sensor data (prediction result data). 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.

[0118] Furthermore, as shown in 14b of FIG. 14, during the prediction period, the acquisition unit 132 Time-series sensor data (prediction result data) output by the RC unit 131; Time-series sensor data a (comparison data) measured by the sensor a 120; to the determination unit 133. The time-series sensor data a measured by the sensor a 120 and transmitted to the acquisition unit 132 of the data processing device 130 is referred to as "correct data" during the learning period, and as "comparison data" during the prediction period.

[0119] <Summary> As is clear from the above description, the substrate processing apparatus 100 according to the sixth embodiment: The reservoir feature generator inputs time-series sensor data a (time-series sensor data a measured by sensor a) acquired in the substrate manufacturing process and outputs reservoir feature data. During the learning period, the weight parameters are learned so that the prediction result data obtained by calculating the reservoir feature quantity output from the reservoir feature quantity generation unit under the weight parameters correlates with the time-series sensor data a measured by the sensor a. During the prediction period, time-series sensor data a acquired in the substrate manufacturing process is input, and the reservoir feature output from the reservoir feature generation unit is calculated based on the learned weight parameters, and prediction result data is output. The predicted result data is compared with time-series sensor data a acquired in the substrate manufacturing process to determine the state of the substrate manufacturing process.

[0120] As a result, the substrate processing apparatus 100 according to the sixth embodiment can provide the same effects as those of the first embodiment.

[0121] [Other embodiments] In the above embodiments, the reservoir feature generator has been described as realizing the function of inputting time-series sensor data acquired in a substrate manufacturing process and outputting reservoir feature amounts.

[0122] However, the function of inputting time-series sensor data acquired in the substrate manufacturing process and outputting reservoir feature quantities may be realized by the reservoir feature quantity generator 205, for example.

[0123] Therefore, in this specification, the reservoir feature generation unit is considered to include both functions realized by the processor 201 executing the data processing program and functions realized by the reservoir feature generator 205 being executed.

[0124] In addition, in each of the above embodiments, the reservoir feature generator 205 has been described as being realized by an FPGA board or the like, but when realized by an FPGA board or the like, it may be configured to include a specific physical reservoir.

[0125] In each of the above embodiments, a case has been described in which time-series sensor data measured by one or more sensors is used to predict time-series sensor data measured by another sensor (or time-series sensor data measured by the sensor itself). Also, in each of the above embodiments, a case has been described in which the state of the substrate manufacturing process is determined based on differential data between prediction result data and comparison data. However, the method for determining the state of the substrate manufacturing process is not limited to this, and, for example, the state of the substrate manufacturing process may be directly predicted based on time-series sensor data measured by one or more sensors. In this case, data indicating the state of the substrate manufacturing process is input to the learning unit as correct answer data.

[0126] Furthermore, although the above embodiments do not mention the learning method used by the learning unit to learn the weight parameters, any learning method may be used by the learning unit to learn the weight parameters. For example, the learning unit may perform learning using a learning method such as the recursive least squares method (RLS). Alternatively, the learning unit may perform learning using a learning method using Bayesian linear regression, a Kalman filter, an information filter, a particle filter, or the like.

[0127] In addition, in the above second to fourth embodiments, cases have been described in which multiple reservoir feature generation units 300_1 to 300_L are used, but the multiple reservoir feature generation units may have the same configuration as each other or different configurations.

[0128] In the above fifth embodiment, a case has been described in which time-series sensor data b measured by one sensor b140 is predicted based on time-series sensor data a1 to a3 measured by three sensors a1 120_1 to a3 120_3. However, the number of sensor data used for prediction is not limited to three, and may be, for example, n pieces of sensor data measured by n sensors (n is any integer equal to or greater than 2) belonging to a first sensor. Furthermore, the number of time-series sensor data to be predicted is not limited to one, and may be, for example, m pieces of sensor data measured by m sensors (m is any integer equal to or greater than 1) belonging to a second sensor.

[0129] The present invention is not limited to the configurations described in the above embodiments, but may be combined with other elements, etc. These aspects can be changed without departing from the spirit of the present invention, and can be appropriately determined depending on the application form.

[0130] This application claims priority based on Japanese Patent Application No. 2021-101004, filed on June 17, 2021, the entire contents of which are incorporated herein by reference. [Explanation of symbols]

[0131] 100: Substrate processing apparatus 110: Chamber 120: Sensor a 130: Data processing device 131:RC section 132: Acquisition Department 133: Judgment section 140: Sensor b 150: Control device 160: Actuator 205: Reservoir feature generator 300: Reservoir feature generation unit 310: Prediction Department 311: Weight acquisition unit 312: Arithmetic section 320: Period determination section 330: Switching section 340: Learning Department 341: Weight calculation unit 501: Sensor data acquisition unit 503: Prediction result data acquisition unit 511: Difference calculation section 512: Status determination unit 900:RC section 300_1~300_L: Reservoir feature generation section 1000: Data processing device 1010_1~1010_L :RC section 1020: Switching section 1021: Switching section 1100: Data processing device 1110_1~1110_L :RC section 1120: Prediction Department 1130: Integration Department 1140: Prediction Department 1141: Weight acquisition unit 1142: Arithmetic unit 1150: Period determination section 1160: Switching section 1170: Learning Department 1171: Weight calculation unit 1200: Substrate processing equipment 1210 :RC section 1400: Substrate processing equipment

Claims

1. a reservoir feature generation unit that receives first sensor data in time series acquired in a substrate manufacturing process and outputs reservoir feature data; a learning unit that learns weight parameters during a learning period so that prediction result data obtained by calculating the reservoir feature output from the reservoir feature generation unit under the weight parameters correlates with second sensor data in time series acquired in the substrate manufacturing process, which is correct answer data; and a period determination unit that, when it is determined that a termination condition is satisfied, determines to perform a transition between the learning period, which is a period during which time-series second sensor data acquired in the substrate manufacturing process is processed as the correct answer data, and a prediction period, which is a period during which the time-series second sensor data acquired in the substrate manufacturing process is processed as comparison data; a prediction unit that, during the prediction period, receives time-series first sensor data acquired in the substrate manufacturing process, calculates reservoir features output from the reservoir feature generation unit under weight parameters learned by the learning unit, and outputs prediction result data; a state determination unit that determines a state of the substrate manufacturing process by comparing the prediction result data with second sensor data in time series acquired in the substrate manufacturing process, which is comparison data, during the prediction period; A substrate processing apparatus having:

2. a plurality of reservoir feature generators each receiving a first sensor data in a time series acquired in the substrate manufacturing process in parallel and outputting a reservoir feature, the learning unit learns weight parameters so that prediction result data obtained by weighting and adding the reservoir feature amounts output from the plurality of reservoir feature generation units using weight parameters correlates with time-series second sensor data acquired in the substrate manufacturing process, which is correct answer data; 2. The substrate processing apparatus according to claim 1, wherein the prediction unit receives first sensor data in a time series acquired in the substrate manufacturing process in parallel, weights and adds reservoir features output from each of the plurality of reservoir feature generation units using weight parameters learned by the learning unit, and outputs prediction result data.

3. the plurality of reservoir feature amount generation units are connected in series, the plurality of reservoir feature amount generation units including a reservoir feature amount generation unit at a beginning to which first sensor data in a time series acquired in the substrate manufacturing process is input, and a reservoir feature amount generation unit subsequent to the beginning to which prediction result data obtained by calculating the reservoir feature amount output by the immediately preceding reservoir feature amount generation unit under a weight parameter is input, During the learning period, each learning unit corresponding to the plurality of reservoir feature generation units learns a corresponding weight parameter so that each prediction result data obtained by calculating the reservoir feature output from each of the plurality of reservoir feature generation units under a weight parameter correlates with time-series second sensor data acquired in the substrate manufacturing process, which is correct answer data; During the prediction period, each prediction unit corresponding to the plurality of reservoir feature generation units calculates the reservoir feature output from each of the plurality of reservoir feature generation units based on a weight parameter learned by a corresponding learning unit, and outputs each prediction result data; 2. The substrate processing apparatus according to claim 1, wherein during the prediction period, the state determination unit determines the state of the substrate manufacturing process by comparing final prediction result data output from a prediction unit corresponding to a terminal reservoir feature generation unit with second sensor data in time series acquired during the substrate manufacturing process, which is comparison data.

4. a plurality of reservoir feature generators each receiving a first sensor data in a time series acquired in the substrate manufacturing process in parallel and outputting a reservoir feature; a plurality of learning units that learn the corresponding weight parameters during the learning period so that each prediction result data obtained by calculating the reservoir feature amounts output from the plurality of reservoir feature generating units under the corresponding weight parameters correlates with time-series second sensor data acquired in the substrate manufacturing process, which is correct answer data; a plurality of prediction units that calculate reservoir features output from the plurality of reservoir feature generation units, respectively, using weight parameters learned by corresponding learning units, and output prediction result data; an integrated learning unit that learns weight parameters used in the weighted addition during the learning period so that final prediction result data obtained by weighting and adding the prediction result data output from each of the plurality of prediction units using weight parameters correlates with second sensor data in time series acquired in the substrate manufacturing process, which is correct answer data; and an integrated prediction unit that, during the prediction period, performs weighted addition of prediction result data output from each of the plurality of prediction units using the weight parameters learned by the integrated learning unit, and outputs final prediction result data; 2. The substrate processing apparatus according to claim 1, wherein during the learning period and the prediction period, the prediction result data output from each of the plurality of prediction units is prediction result data obtained by calculating reservoir features output from each of the plurality of reservoir feature generation units using corresponding weight parameters, and correcting the prediction result data with the final prediction result data.

5. a reservoir feature generation unit that receives a plurality of first sensor data in time series acquired in the substrate manufacturing process and outputs a reservoir feature; the learning unit learns weight parameters so that prediction result data obtained by calculating the reservoir feature output from the reservoir feature generation unit under weight parameters correlates with second sensor data in time series acquired in the substrate manufacturing process, which is correct answer data; 2. The substrate processing apparatus according to claim 1, wherein the prediction unit calculates reservoir features output from the reservoir feature generation unit in response to input of the plurality of time-series first sensor data, based on weight parameters learned by the learning unit, and outputs prediction result data.

6. the first time-series sensor data acquired in the substrate manufacturing process and input to the reservoir feature generation unit is time-series sensor data measured by a first sensor; A substrate processing apparatus according to any one of claims 1 to 5, wherein the second time-series sensor data acquired during the substrate manufacturing process and used for learning by the learning unit is time-series sensor data measured by a second sensor different from the first sensor.

7. the plurality of time-series first sensor data acquired in the substrate manufacturing process are n time-series sensor data measured by n sensors (n is an integer of 2 or more) belonging to a first sensor; The substrate processing apparatus of claim 6, wherein the second time-series sensor data acquired during the substrate manufacturing process and used for learning by the learning unit is m time-series sensor data measured by m sensors (m is an integer greater than or equal to 1) belonging to a second sensor different from the first sensor.

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

9. a difference calculation unit that calculates a difference between time-series second sensor data measured by the second sensor and prediction result data output by the prediction unit; a state determination unit that determines a state of the substrate manufacturing process based on the difference data calculated by the difference calculation unit; The substrate processing apparatus according to claim 6 or 7, further comprising:

10. a difference calculation unit that calculates a difference between time-series first sensor data measured by the first sensor and prediction result data output by the prediction unit; a state determination unit that determines a state of the substrate manufacturing process based on the difference data calculated by the difference calculation unit; The substrate processing apparatus of claim 8 , further comprising:

11. 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 difference data exceeds a predetermined threshold value or by determining whether the difference data for a predetermined time range resembles a predetermined pattern.

12. The substrate processing apparatus according to claim 11 , wherein an actuator is controlled in accordance with a result of determining the state of the substrate manufacturing process.

13. The period determination unit determines that the termination condition is satisfied when a state in which there is no error between the prediction result data and the correct answer data has elapsed for a specified time period, and determines to transition from the learning period to the prediction period. The substrate processing apparatus according to claim 1 .

14. The system further comprises a switching unit that switches a 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. The substrate processing apparatus according to claim 1 .

15. The period determination unit determines whether or not relearning is necessary during the prediction period, and if it determines that relearning is necessary, determines that the termination condition is satisfied and determines to transition from the prediction period to the learning period. The substrate processing apparatus according to claim 1 .

16. A data processing method in a substrate processing apparatus, comprising: a reservoir feature generating step of inputting first sensor data in time series acquired in a substrate manufacturing process and outputting reservoir feature data; a learning step of learning weight parameters during a learning period so that prediction result data obtained by calculating the reservoir feature output in the reservoir feature generating step under the weight parameters correlates with time-series second sensor data acquired in the substrate manufacturing process, which is correct data; a period determination step of determining, when it is determined that a termination condition is satisfied, to transition between the learning period, which is a period during which time-series second sensor data acquired in the substrate manufacturing process is processed as the correct answer data, and a prediction period, which is a period during which the time-series second sensor data acquired in the substrate manufacturing process is processed as comparison data; a prediction step of calculating, during the prediction period, reservoir features output in the reservoir feature generation step based on the weight parameters learned in the learning step by inputting time-series first sensor data acquired in the substrate manufacturing process, and outputting prediction result data; a state determination step of determining a state of the substrate manufacturing process by comparing the prediction result data with second sensor data in time series acquired in the substrate manufacturing process, which is comparison data, during the prediction period; A data processing method comprising:

17. The computer of the substrate processing apparatus a reservoir feature generating step of inputting first sensor data in time series acquired in a substrate manufacturing process and outputting reservoir feature data; a learning step of learning weight parameters during a learning period so that prediction result data obtained by calculating the reservoir feature output in the reservoir feature generating step under the weight parameters correlates with time-series second sensor data acquired in the substrate manufacturing process, which is correct data; a period determination step of determining, when it is determined that a termination condition is satisfied, to transition between the learning period, which is a period during which time-series second sensor data acquired in the substrate manufacturing process is processed as the correct answer data, and a prediction period, which is a period during which the time-series second sensor data acquired in the substrate manufacturing process is processed as comparison data; a prediction step of calculating, during the prediction period, reservoir features output in the reservoir feature generation step based on the weight parameters learned in the learning step by inputting time-series first sensor data acquired in the substrate manufacturing process, and outputting prediction result data; a state determination step of determining a state of the substrate manufacturing process by comparing the prediction result data with second sensor data in time series acquired in the substrate manufacturing process, which is comparison data, during the prediction period; A data processing program for executing the above.

Images