Transfer anomaly prediction system, substrate processing apparatus, transfer anomaly prediction method, recording medium, and completion learning model

Abstract

The transport anomaly prediction system includes an estimation unit with a completion learning model. The completion learning model performs machine learning on the relationship between a dataset containing sensor data output by multiple sensors provided in the substrate transport unit during past substrate transport and the transport anomaly degree during the substrate transport. The estimation unit takes a dataset containing sensor data output by the multiple sensors during new substrate transport as input, estimates the transport anomaly degree during the new substrate transport, and outputs it.

Description

Technical Field

[0001] This disclosure relates to a transport anomaly prediction system. Background Technology

[0002] In semiconductor manufacturing equipment, substrates must be processed. However, during prolonged operation, processing errors can occur due to component wear and equipment deviations. When processing errors occur, the equipment stops, resulting in reduced productivity, or the substrate itself may be damaged.

[0003] Japanese Patent No. 6325325 discloses a system that compares the output of a sensor in operation with the output of a sensor stored when the wafer processing equipment was aligned before operation. If the difference exceeds a certain threshold, it is determined that the wafer has deviated from the alignment.

[0004] The problem the invention aims to solve

[0005] However, in the system disclosed in Japanese Patent No. 6325325, when the difference between the output of the sensor stored when the output of the chip processing device during operation is aligned with the position before operation exceeds a certain threshold, it is judged as an anomaly. This is a simple method for anomaly detection, and the detection probability is not high. Summary of the Invention

[0006] We hope to provide a transport anomaly prediction system that can improve the detection probability of transport anomalies.

[0007] One aspect of the present disclosure of a transport anomaly prediction system includes an estimation unit having a completion learning model. The completion learning model performs machine learning on the relationship between a dataset containing sensor data output from multiple sensors provided in the substrate transport unit during past substrate transport and the transport anomaly degree during the substrate transport. The estimation unit takes a dataset containing sensor data output from the multiple sensors during new substrate transport as input, estimates the transport anomaly degree during the new substrate transport, and outputs it. Attached Figure Description

[0008] Figure 1 This is a schematic configuration diagram of a substrate processing apparatus according to one embodiment.

[0009] Figure 2 This is a block diagram representing one implementation of a transport anomaly prediction system.

[0010] Figure 3 This is an example diagram showing sensor data during normal substrate transport.

[0011] Figure 4 This diagram illustrates the substrate transport section when a transport malfunction occurs.

[0012] Figure 5 This is an example diagram showing sensor data when a transport anomaly occurs.

[0013] Figure 6 This is a schematic diagram illustrating the structure of the learning model in the first approach.

[0014] Figure 7 This is a schematic diagram illustrating the structure of the learning model in the second approach.

[0015] Figure 8 This is a schematic diagram illustrating the structure of the learning model in the third approach.

[0016] Figure 9 This is a schematic diagram illustrating the structure of the learning model in the fourth approach.

[0017] Figure 10 This is a schematic diagram illustrating the structure of the learning model in the fifth approach.

[0018] Figure 11 This is a schematic diagram illustrating the structure of the learning model in the sixth approach.

[0019] Figure 12 This is a schematic diagram illustrating the structure of the learning model in the seventh approach.

[0020] Figure 13 This is a schematic diagram illustrating the structure of the learning model in the eighth approach.

[0021] Figure 14 This is a flowchart illustrating an example of a method for predicting transport anomalies in one implementation. Detailed Implementation

[0022] The first embodiment of the conveying anomaly prediction system includes an estimation unit.

[0023] The estimation unit has a completion learning model, which performs machine learning on the relationship between a dataset containing sensor data output by multiple sensors provided in the substrate transport unit during past substrate transport and the transport anomaly degree during the substrate transport. The estimation unit takes a dataset containing sensor data output by the multiple sensors during new substrate transport as input, estimates the transport anomaly degree during the new substrate transport, and outputs it.

[0024] In this approach, the estimation unit uses a complete learning model to perform machine learning on the relationship between a dataset containing sensor data from past substrate transport and the degree of transport anomalies during those substrate transports. For a dataset containing sensor data from new substrate transports, the unit can synthesize multiple index data to estimate and output the degree of transport anomalies during the new substrate transport. This improves the detection probability of transport anomalies compared to the previous method where anomalies were judged when the difference between the sensor output during past substrate transport and the stored sensor output before the substrate was aligned with its position exceeded a certain threshold. Furthermore, by using the complete learning model, sensor data such as vibration, sound, and image data from equipment, which were difficult to process using the previous method, can be utilized.

[0025] The second embodiment of the conveying anomaly prediction system is similar to the first embodiment of the conveying anomaly prediction system.

[0026] The plurality of sensors are composed of one or more of the following: vibration sensor, sound sensor, image sensor, video sensor, temperature sensor, equipment moving speed sensor, equipment action torque sensor, and equipment parallelism sensor.

[0027] The third-party transport anomaly prediction system of the implementation method, such as the transport anomaly prediction system of the first or second method,

[0028] It further includes an output signal transmitting unit, which compares the transport anomaly degree output by the estimation unit with a predetermined threshold. If the transport anomaly degree exceeds the threshold, it sends an output signal for outputting maintenance notification and / or alarm to the output device.

[0029] The fourth embodiment of the conveying anomaly prediction system is any of the conveying anomaly prediction systems in the first to third embodiments.

[0030] The estimation unit takes a dataset containing sensor data from the start of the new substrate transport to the present as input, estimates the transport anomaly during the new substrate transport, and outputs the result.

[0031] The fifth embodiment of the conveying anomaly prediction system is any of the conveying anomaly prediction systems in the first to third embodiments.

[0032] The estimation unit takes a dataset containing sensor data from the start to the end of the transfer when the new substrate is being transferred as input, estimates the degree of transfer anomaly during the transfer of the new substrate, and outputs the result.

[0033] The sixth embodiment of the conveying anomaly prediction system is any of the conveying anomaly prediction systems in the first to third embodiments.

[0034] The estimation unit takes a dataset of sensor data, which includes data from the beginning of the transport of the first substrate to the end of the transport of the last substrate, as input, estimates the transport anomaly during the transport of the multiple new substrates, and outputs the result.

[0035] The seventh embodiment of the conveying anomaly prediction system is any of the conveying anomaly prediction systems in the first to sixth embodiments.

[0036] It further includes a relearning unit that uses a dataset containing sensor data output during the transport of the new substrate as teacher data to enable the completed learning model to relearn.

[0037] The eighth embodiment of the conveying anomaly prediction system is any of the conveying anomaly prediction systems in the first to seventh embodiments.

[0038] The dataset further includes at least one time information among the equipment operation time during substrate transport, the elapsed time after maintenance, and the component usage time of the substrate processing unit.

[0039] The ninth embodiment of the conveying anomaly prediction system is any of the conveying anomaly prediction systems in the first to eighth embodiments.

[0040] The completion learning model performs machine learning on teacher data in a dataset containing sensor data from past substrate transports, along with the remaining time or number of transports from the time of the substrate transport until the transport anomaly occurs. The estimation unit takes a dataset containing sensor data from new substrate transports as input, and estimates and outputs the transport anomaly degree based on the remaining time or number of transports predicted by the completion learning model.

[0041] The tenth embodiment of the conveying anomaly prediction system is any of the conveying anomaly prediction systems in the first to eighth embodiments.

[0042] The completed learning model uses a dataset containing sensor data from past normal substrate transport as teacher data and performs machine learning using the k-nearest neighbor algorithm. The estimation unit uses a dataset containing sensor data from new substrate transport as input and estimates the transport anomaly degree based on the distance to the k-nearest neighbor calculated by the completed learning model and outputs it.

[0043] The eleventh embodiment of the conveying anomaly prediction system is any of the conveying anomaly prediction systems in the first to eighth embodiments.

[0044] The completed learning model uses a dataset containing sensor data from past normal substrate transport as teacher data and performs machine learning using LSTM (Long Short-Term Memory). The estimation unit takes a dataset containing actual sensor data before the new substrate transport as input, calculates the deviation between the dataset containing the sensor data predicted by the completed learning model for the new substrate transport and the dataset containing the actual sensor data for the new substrate transport, and infers and outputs the transport anomaly degree based on the deviation.

[0045] The twelfth embodiment of the conveying anomaly prediction system is any of the conveying anomaly prediction systems in the first to eighth embodiments.

[0046] The completed learning model performs machine learning on teacher data that includes a dataset containing sensor data from past substrate transports and labels indicating whether a transport anomaly occurred during the substrate transport. The estimation unit takes a dataset containing sensor data from new substrate transports as input, estimates the transport anomaly degree based on the probability of a transport anomaly predicted by the completed learning model, and outputs the result.

[0047] The thirteenth embodiment of the conveying anomaly prediction system is similar to the twelfth embodiment of the conveying anomaly prediction system.

[0048] The completed learning model adds a label to the dataset containing sensor data from past substrate transports, indicating whether a transport anomaly occurred during the transport. When a transport anomaly occurs, machine learning is performed on teacher data labeled with the cause of the transport anomaly. The estimation unit takes the dataset containing sensor data from new substrate transports as input, and estimates the transport anomaly degree of each cause of transport anomaly based on the probability of a transport anomaly predicted by the completed learning model.

[0049] The fourteenth embodiment of the conveying anomaly prediction system is any of the conveying anomaly prediction systems in the first to thirteenth embodiments.

[0050] The estimation unit has multiple learning models and estimates and outputs the transport anomaly degree based on a combination of predictions from the multiple learning models.

[0051] The substrate processing apparatus of the fifteenth embodiment includes: a substrate transport unit; and a transport anomaly prediction system as in any one of the first to fourteenth embodiments.

[0052] The sixteenth embodiment of the method for predicting transport anomalies.

[0053] This is a computer-executed method for predicting transport anomalies, which includes the following steps:

[0054] The completed learning model performs machine learning on the relationship between a dataset containing sensor data output by multiple sensors located in the substrate transport section during past substrate transport and the transport anomaly degree during the current substrate transport. The model takes a dataset containing sensor data output by the multiple sensors during a new substrate transport as input, estimates the transport anomaly degree during the new substrate transport, and outputs it.

[0055] The seventeenth implementation method includes a transport anomaly prediction procedure.

[0056] Have the computer perform the following steps:

[0057] A fully learned learning model, which performs machine learning on the relationship between a dataset containing sensor data output by multiple sensors provided in the substrate transport section during past substrate transport and the transport anomaly degree during the current substrate transport, takes a dataset containing sensor data output by the multiple sensors during the new substrate transport as input, estimates the transport anomaly degree during the new substrate transport, and outputs it.

[0058] The eighteenth embodiment of the computer-readable recording medium.

[0059] It is a non-transitory record used by the computer to perform the following steps of the transport anomaly prediction procedure.

[0060] A fully learned learning model, which performs machine learning on the relationship between a dataset containing sensor data output by multiple sensors provided in the substrate transport section during past substrate transport and the transport anomaly degree during the current substrate transport, takes a dataset containing sensor data output by the multiple sensors during the new substrate transport as input, estimates the transport anomaly degree during the new substrate transport, and outputs it.

[0061] The nineteenth implementation method completes the learning model as a tuned neural network-like system.

[0062] The model comprises: an input layer; one or more intermediate layers connected to the input layer; and an output layer connected to the intermediate layers. The completed learning model inputs a dataset containing sensor data output from multiple sensors located in the substrate transport section during past substrate transports into the input layer. By repeatedly comparing the dataset containing sensor data from multiple past substrate transports with the transport anomaly degree during the substrate transport, and updating the parameters of each node according to the error, machine learning is performed to study the relationship between the dataset containing sensor data output from past substrate transports and the transport anomaly degree during the substrate transport.

[0063] The computer is configured to perform the following functions: when a dataset containing sensor data output from the multiple sensors during the transfer of a new substrate is input into the input layer, the transfer anomaly during the transfer of the new substrate is estimated and output from the output layer.

[0064] Hereinafter, specific examples of embodiments will be described in detail with reference to the accompanying drawings. Furthermore, in the following description and the drawings used in the following description, the same symbols will be used for parts with the same structure, and repeated descriptions will be omitted.

[0065] Figure 1 This is a schematic configuration diagram of a substrate processing apparatus 1 according to one embodiment.

[0066] like Figure 1 As shown, the substrate processing apparatus 1 includes: a top ring 6, a substrate conveying section 2, nozzles 3a and 3b, a conveying anomaly prediction system 10, and an output device 4.

[0067] The top ring 6 has an air bladder on its lower surface to hold the substrate W downward, and is configured to move the substrate W between a substrate transport position (the position opposite to the substrate transport section 2) and a substrate processing position (e.g., the position opposite to a polishing table not shown).

[0068] Nozzles 3a and 3b are positioned on the side of the top ring 6 in the substrate transport position, and are configured to spray air and shower between the air bladder of the top ring 6 and the substrate W, thereby peeling the substrate W from the top ring 6.

[0069] The substrate transport unit 2 is positioned below the top ring 6 in the substrate transport position and is configured to receive the substrate W that is peeled off from the top ring 6 and falls down. The substrate transport unit 2 may be, for example, a lift or a linear transporter (LTP).

[0070] like Figure 1As shown, the substrate transport section 2 is equipped with multiple sensors 51 to 53 (three in the illustrated example). In the illustrated example, the multiple sensors 51 to 53 are vibration sensors (accelerometers) that measure the vibration of the substrate transport section 2, and are installed in the arm part and shaft part of the substrate transport section 2 to directly transmit the vibration when the substrate W is transferred.

[0071] In addition, the multiple sensors 51 to 53 are not limited to vibration sensors, but may also consist of one or more of the following: vibration sensor, sound sensor, image sensor, video sensor, temperature sensor, equipment moving speed sensor, equipment operating torque sensor, and equipment parallelism sensor.

[0072] Figure 3 This is an example diagram showing the sensor data output from sensors 51 to 53 during normal substrate transport. Figure 3 In the diagram, the area enclosed by a solid square represents the sensor data collected from the start to the end of one substrate transport cycle (one cycle). For example... Figure 3 As shown, during normal transport, sensor data with the same waveform is repeatedly output from sensors 51 to 53 in each cycle.

[0073] Figure 4 This diagram illustrates the substrate transport unit 2 in the event of a transport malfunction. Figure 5 This is an example diagram showing sensor data when a transport anomaly occurs.

[0074] like Figure 4 As shown, a typical symptom of a transport abnormality is that the substrate W cannot be symmetrically peeled off from the top ring 6, and the substrate W falls at an angle. At this time, because the impact on the substrate W when it is in the substrate transport section 2 is concentrated in one place, the substrate W is easily damaged. If the symptoms worsen, the substrate W will fall off the substrate transport section 2. Figure 4 The example shown can be considered as the cause (type of abnormality) of the conveying abnormality: (1) the air sprayed from nozzles 3a and 3b and the position touched by the shower; (2) the misalignment of the axis of the top ring 6 and the axis of the substrate conveying part 2.

[0075] like Figure 5 As shown, when a conveying abnormality occurs, sensor data with a waveform different from the normal waveform will be output from sensors 51 to 53. Figure 5 In the middle, the area enclosed by the solid squares represents the sensor data from the start of the transport process to the end of the transport process when a transport anomaly occurs.

[0076] Output device 4 is an interface for outputting various information to a user (e.g., the operator of the substrate processing device 1), such as an image display mechanism (monitor) using a liquid crystal display, a lamp, and a speaker.

[0077] like Figure 1 As shown, the transport anomaly prediction system 10 is communicatively connected to multiple sensors 51-53 and output device 4.

[0078] Secondly, the composition of the transport anomaly prediction system 10 will be explained. Figure 2 This is a block diagram representing the configuration of the transport anomaly prediction system 10. At least a portion of the transport anomaly prediction system 10 is configured using a single computer or quantum computing system, or multiple computers or quantum computing systems interconnected via a network.

[0079] like Figure 2 As shown, the transport anomaly prediction system 10 includes an input unit 11, a control unit 12, a storage unit 13, and an output unit 14. Each unit 11 to 14 is communicatively connected via a bus or an Internet.

[0080] The input unit 11 serves as a communication interface for the multiple sensors 51 to 5N disposed on the substrate transport unit 2. The input unit 11 can be connected via wired or wireless connection to each output terminal of the multiple sensors 51 to 5N.

[0081] Output unit 14 is a communication interface to output device 4. Output unit 14 can be connected to the input terminal of output device 4 by wired or wireless means.

[0082] Storage unit 13 is, for example, a non-volatile data storage medium such as flash memory. Storage unit 13 stores various data processed by control unit 12. For example, storage unit 13 stores dataset 151 referenced by estimation unit 121 (described later) and threshold 152 referenced by output signal transmission unit 122 (described later).

[0083] Data set 151 is output from multiple sensors 51 to 5N provided in the substrate transport section 2 during the transport of a new substrate, and contains sensor data acquired via the input section 11. Data set 151 may also further contain at least one of the following time information: equipment operation time during substrate transport, time elapsed after maintenance, and component usage time of the substrate processing section (the part that gradually wears down in contact with the substrate W).

[0084] like Figure 2 As shown, the control unit 12 includes: an estimation unit 121, an output signal transmission unit 122, and a relearning unit 123. These units can be implemented by executing a specified program through a processor in the anomaly prediction system 10, or they can be installed in hardware.

[0085] The estimation unit 121 has completed the learning model 120 (e.g., a tuned neural network-like system, referring to...). Figures 6 to 13The completed learning model 120 performs machine learning on the relationship between the dataset containing sensor data output by multiple sensors 51 to 5N provided on the substrate transfer section 2 during past substrate transfers and the transfer anomaly degree during the current substrate transfer, and takes the dataset containing sensor data output by multiple sensors 51 to 5N during the new substrate transfer (that is, the dataset 131 stored in the storage section 13) as input, estimates the transfer anomaly degree during the new substrate transfer and outputs it.

[0086] The first example of the timing of the processing in the estimation unit 121 is that the estimation unit 121 can also process in real time. That is, it can also take the dataset 131 containing sensor data from the start of the new substrate to the present as input, estimate the degree of transport anomaly during the transport of the new substrate and output it.

[0087] A second example of the timing of the processing in the estimation unit 121 is that the estimation unit 121 can also process each substrate W. That is, it can also take the dataset 131 containing sensor data from the start of the new substrate to the end of the new substrate as input, estimate the degree of transport anomaly during the transport of the new substrate and output it.

[0088] The third example of the processing timing in the estimation unit 121 is that the estimation unit 121 can also process each batch of substrates W (for example, a batch of 25 pieces). That is, the estimation unit 121 can also take the dataset 131 containing sensor data from the start of the initial substrate transport to the end of the last substrate transport when transporting multiple new substrates (a batch) as input, estimate the transport anomaly degree when transporting the multiple new substrates (a batch) and output it.

[0089] The estimation unit 121 can also use the sensor data output from the multiple sensors 51 to 5N as input, or it can use the intensity of a predetermined frequency region extracted (preprocessed) by FFT (Fast Fourier Transform) as input.

[0090] Figure 6 This is a schematic diagram illustrating the structure of the learning model 120 in the first approach. Figure 6 The completed learning model 120 shown in the example is a tuned neural network system and contains a hierarchical neural network or quantum neural network (QNN) having: an input layer; one or more intermediate layers connected to the input layer; and an output layer connected to the intermediate layers. Figure 6 The diagram illustrates a hierarchical neural network, specifically a feedforward neural network. However, various other types of neural networks, such as convolutional neural networks (CNNs) and recurrent neural networks (RNNs), can be used. The completed learning model 120 can also contain multi-layered neural networks with two or more intermediate layers, i.e., deep learning.

[0091] An example of the method for generating the learning model 120 in the first approach is explained below. Teacher data is prepared by adding a tag indicating whether a transport abnormality occurred during the transport of the substrate (e.g., 0 for normal substrate transport and 1 for transport abnormality) to the sensor data output from the multiple sensors 51-5N located in the substrate transport section 2 during past substrate transport. Figure 6 As shown, sensor data from a substrate transport process contained in the teacher data is input into the input layer. The output from the output layer is repeatedly compared with the label indicating whether a transport anomaly occurred during the substrate transport process, as contained in the teacher data. For each data point from multiple substrate transport processes contained in the teacher data, the parameters (weighting and threshold, etc.) of each node are updated according to their error. Therefore, a completed learning model 120 (tuned neural network system) is generated based on sensor data output from multiple sensors 51-5N located in the substrate transport unit 2 during past substrate transport processes, predicting the probability of a transport anomaly occurring during the substrate transport process (transport anomaly probability).

[0092] In the first approach, the estimation unit 121 takes as input the sensor data output by the multiple sensors 51 to 5N during the transfer of the new substrate, and estimates and outputs the degree of transfer anomaly during the transfer of the new substrate based on the probability of a transfer anomaly occurring (transfer anomaly probability) predicted by the learning model 120. Alternatively, the degree of transfer anomaly can be the probability of a transfer anomaly occurring (transfer anomaly probability) predicted by the learning model 120, or it can be a value that uniquely transforms the transfer anomaly probability using a specified function.

[0093] Figure 7 This is a schematic diagram illustrating the structure of the learning model 120 in the second approach. Figure 7 The completed learning model 120 shown in the example is a tuned neural network system and contains a hierarchical neural network or quantum neural network (QNN) having: an input layer; one or more intermediate layers connected to the input layer; and an output layer connected to the intermediate layers. Figure 7 The diagram illustrates a hierarchical neural network, specifically a feedforward neural network. However, various other types of neural networks, such as convolutional neural networks (CNNs) and recurrent neural networks (RNNs), can be used. The completed learning model 120 can also include multi-layered neural networks with two or more intermediate layers, i.e., deep learning.

[0094] An example of the method for generating the learning model 120 in the second approach is explained below. Teacher data is prepared by adding tags indicating whether a transport abnormality occurred during the transport of the substrate (e.g., 0 for normal substrate transport and 1 for transport abnormality) to a dataset containing sensor data output from multiple sensors 51-5N located in the substrate transport section 2 during past substrate transports, and time information during the substrate transport (i.e., at least one of equipment operation time, post-maintenance elapsed time, and component usage time in the substrate processing section). Figure 7 As shown, a dataset containing one substrate transport time from the teacher data is input into the input layer. The output from the output layer is repeatedly compared with the label indicating whether a transport anomaly occurred during that substrate transport time, as contained in the teacher data. For each data point containing multiple substrate transport times from the teacher data, the parameters (weighting and threshold, etc.) of each node are updated according to their error. Therefore, a completed learning model 120 (tuned neural network system) is generated based on a dataset containing sensor data output from multiple sensors 51-5N installed in the substrate transport unit 2 during past substrate transport times and the time information of that substrate transport time, predicting the probability of a transport anomaly occurring during that substrate transport time (transport anomaly probability).

[0095] In the second method, the estimation unit 121 takes as input a dataset containing sensor data output from multiple sensors 51 to 5N during the transport of a new substrate and time information during the transport of the new substrate, and estimates and outputs the transport anomaly degree during the transport of the new substrate based on the probability of transport anomaly occurring (transport anomaly probability) predicted by the learning model 120. Alternatively, the transport anomaly degree can be the probability of transport anomaly occurring (transport anomaly probability) predicted by the learning model 120, or it can be a value that uniquely transforms the transport anomaly probability using a specified function.

[0096] Figure 8 This is a schematic diagram illustrating the structure of the learning model 120 in the third approach. Figure 8 The completed learning model 120 shown in the example is a tuned neural network system and contains a hierarchical neural network or quantum neural network (QNN) having: an input layer; one or more intermediate layers connected to the input layer; and an output layer connected to the intermediate layers. Figure 8 The diagram illustrates a hierarchical neural network, specifically a feedforward neural network. However, various other types of neural networks, such as convolutional neural networks (CNNs) and recurrent neural networks (RNNs), can be used. The completed learning model 120 can also include multi-layered neural networks with two or more intermediate layers, i.e., deep learning.

[0097] An example of the method for generating the learning model 120 in the third approach is explained below. Teacher data is prepared to add tags to the sensor data output from multiple sensors 51-5N located in the substrate transport section 2 during past substrate transport, indicating whether a transport abnormality occurred during the transport (e.g., 0 for normal substrate transport, 1 for a transport abnormality), and also adding tags indicating the cause (type of abnormality) of the transport abnormality when it occurred. Figure 8 As shown, sensor data from a substrate transport process contained in the teacher data is input into the input layer. The output from the output layer is repeatedly compared with the labels indicating whether a transport anomaly occurred during substrate transport, based on the causes (anomaly types) of each transport anomaly contained in the teacher data. For each data point from multiple substrate transport processes contained in the teacher data, the parameters (weighting and thresholds, etc.) of each node are updated according to their errors. Therefore, a completed learning model 120 (tuned neural network system) is generated based on sensor data output from multiple sensors 51-5N located in the substrate transport unit 2 during past substrate transport processes, predicting the probability (transport anomaly probability) of a transport anomaly occurring during substrate transport based on the causes (anomaly types) of each transport anomaly.

[0098] In the third method, the estimation unit 121 takes the sensor data output by the multiple sensors 51 to 5N during the transfer of the new substrate as input, completes the learning model 120, and estimates the transfer anomaly degree during the transfer of the new substrate based on the probability of the transfer anomaly occurring (transfer anomaly probability) predicted for each cause (anomaly type) of the transfer anomaly. Alternatively, the transfer anomaly degree can be the probability of the transfer anomaly occurring (transfer anomaly probability) predicted by the learning model 120, or it can be a value that uniquely transforms the transfer anomaly probability using a specified function.

[0099] Figure 9 This is a schematic diagram illustrating the structure of the learning model 120 in the fourth approach. Figure 9 The completed learning model 120 shown in the example is a tuned neural network system and contains a hierarchical neural network or quantum neural network (QNN) having: an input layer; one or more intermediate layers connected to the input layer; and an output layer connected to the intermediate layers. Figure 9 The diagram illustrates a hierarchical neural network, specifically a feedforward neural network. However, various other types of neural networks, such as convolutional neural networks (CNNs) and recurrent neural networks (RNNs), can be used. The completed learning model 120 can also include multi-layered neural networks with two or more intermediate layers, i.e., deep learning.

[0100] An example of the method for generating the learning model 120 in the fourth approach is explained below. Teacher data is prepared that includes sensor data output from multiple sensors 51-5N located in the substrate transport section 2 during past substrate transports, and time information during substrate transport (i.e., at least one of the equipment operation time, post-maintenance elapsed time, and component usage time in the substrate processing section during substrate transport). This data includes labels indicating whether a transport abnormality occurred during substrate transport (e.g., 0 for normal substrate transport, 1 for transport abnormality), and labels indicating the cause (type) of the transport abnormality when one occurs. Figure 9 As shown, a dataset containing one substrate transport time from the teacher data is input into the input layer. The output from the output layer is repeatedly compared with the labels indicating whether a transport anomaly occurred during substrate transport, based on the causes (anomaly types) of each transport anomaly contained in the teacher data. For each of the multiple substrate transport times contained in the teacher data, the parameters (weighting and thresholds, etc.) of each node are updated according to their errors. Therefore, a complete learning model 120 (tuned neural network system) is generated, based on a dataset containing sensor data output from multiple sensors 51-5N installed in the substrate transport unit 2 during past substrate transport times and the time information of that substrate transport time, predicting the probability of a transport anomaly occurring during substrate transport for each cause (anomaly type) of transport anomaly.

[0101] In the fourth method, the estimation unit 121 takes as input a dataset containing sensor data output from multiple sensors 51 to 5N during the transport of a new substrate and time information during the transport of the new substrate. It then uses the learning model 120 to estimate the transport anomaly degree during the transport of the new substrate based on the probability of a transport anomaly occurring (transport anomaly probability) predicted by the learning model 120 for each cause (anomaly type). Alternatively, the transport anomaly degree can be the probability of a transport anomaly occurring predicted by the learning model 120, or it can be a value that uniquely transforms the transport anomaly probability using a specified function.

[0102] Figure 10 This is a schematic diagram illustrating the structure of the learning model 120 in the fifth approach. Figure 10 The completed learning model 120 shown in the example is a tuned neural network system and contains a hierarchical neural network or quantum neural network (QNN) having: an input layer; one or more intermediate layers connected to the input layer; and an output layer connected to the intermediate layers. Figure 10The diagram illustrates a hierarchical neural network, specifically a feedforward neural network. However, various other types of neural networks, such as convolutional neural networks (CNNs) and recurrent neural networks (RNNs), can be used. The completed learning model 120 can also include multi-layered neural networks with two or more intermediate layers, i.e., deep learning.

[0103] An example of the method for generating the learning model 120 in the fifth approach is explained below. Teacher data, such as the remaining time or remaining number of transports from the time the substrate was transported until the transport abnormality occurred, is included in the sensor data output from the multiple sensors 51-5N located in the substrate transport section 2 during past substrate transport. Figure 10 As shown, sensor data from a substrate transport process included in the teacher data is input to the input layer. The output from the output layer is repeatedly compared with the remaining time or number of transports from the substrate transport time to the occurrence of a transport anomaly, as contained in the teacher data. For each data point from multiple substrate transport times included in the teacher data, the parameters (weighting and thresholds, etc.) of each node are updated according to their error. Therefore, a complete learning model 120 (tuned neural network system) is generated based on sensor data output from multiple sensors 51-5N located in the substrate transport unit 2 during past substrate transport processes, predicting the remaining time or number of transports from the substrate transport time to the occurrence of a transport anomaly.

[0104] In the fifth method, the estimation unit 121 takes as input the sensor data output by the multiple sensors 51 to 5N during the transfer of the new substrate, and estimates and outputs the transfer anomaly degree during the transfer of the new substrate based on the remaining time or remaining transfer count predicted by the completion learning model 120. Alternatively, the transfer anomaly degree can be the remaining time or remaining transfer count predicted by the completion learning model 120, or it can be a value that uniquely transforms the remaining time or remaining transfer count using a specified function. For example, the transfer anomaly degree can also be the value of dividing the remaining time predicted by the completion learning model 120 by the average time from the start of maintenance to the occurrence of the transfer anomaly, or it can be the value of dividing the remaining transfer count predicted by the completion learning model 120 by the average number of transfers from the start of maintenance to the occurrence of the transfer anomaly.

[0105] Figure 11 This is a schematic diagram illustrating the structure of the learning model 120 in the sixth approach. Figure 11 The completed learning model 120 shown in the example is a tuned neural network system and contains a hierarchical neural network or quantum neural network (QNN) having: an input layer; one or more intermediate layers connected to the input layer; and an output layer connected to the intermediate layers. Figure 11The diagram illustrates a hierarchical neural network, specifically a feedforward neural network. However, various other types of neural networks, such as convolutional neural networks (CNNs) and recurrent neural networks (RNNs), can be used. The completed learning model 120 can also include multi-layered neural networks with two or more intermediate layers, i.e., deep learning.

[0106] An example of the method for generating the learning model 120 in the sixth approach is explained below. A dataset containing sensor data output from multiple sensors 51-5N located in the substrate transport section 2 during past substrate transports, and time information during substrate transport (i.e., at least one of equipment operation time, post-maintenance elapsed time, and component usage time in the substrate processing section), is prepared, along with teacher data indicating the remaining time or remaining transport attempts from the substrate transport time up to the time of the transport abnormality. Figure 10 As shown, a dataset containing one substrate transport time from the teacher data is input into the input layer. The output from the output layer is repeatedly compared with the remaining time or remaining transport count from the substrate transport time to the occurrence of a transport anomaly, as contained in the teacher data. For each data point containing multiple substrate transport times from the teacher data, the parameters (weighting and thresholds, etc.) of each node are updated according to their error. Therefore, a completed learning model 120 (tuned neural network system) is generated based on a dataset containing sensor data output from multiple sensors 51-5N installed in the substrate transport unit 2 during past substrate transport times and the time information of that substrate transport time, predicting the remaining time or remaining transport count from the substrate transport time to the occurrence of a transport anomaly.

[0107] In the sixth method, the estimation unit 121 takes as input a dataset containing sensor data output from multiple sensors 51 to 5N during the transport of a new substrate, and time information during the transport of the new substrate. Based on the remaining time or remaining transport count predicted by the completion learning model 120, it estimates and outputs the transport anomaly degree during the transport of the new substrate. Alternatively, the transport anomaly degree can be the remaining time or remaining transport count predicted by the completion learning model 120, or it can be a value that uniquely transforms the remaining time or remaining transport count using a specified function. For example, the transport anomaly degree can be the value of dividing the remaining time predicted by the completion learning model 120 by the average time from the start of maintenance to the occurrence of the transport anomaly, or it can be the value of dividing the remaining transport count predicted by the completion learning model 120 by the average number of transport counts from the start of maintenance to the occurrence of the transport anomaly.

[0108] Figure 12 This is a schematic diagram illustrating the structure of the learning model 120 in the seventh approach. Figure 12The example shown uses the completed learning model 120 as teacher data, which contains sensor data from past normal substrate transport, and performs machine learning using the k-nearest neighbor algorithm. It also calculates the distance from the dataset to the k-nearest neighbor based on the dataset containing sensor data from new substrate transport. Figure 12 In the example shown, multiple hollow circles (“○”) represent the positions of sensor data in the feature space during normal substrate transport in the past, hollow triangles (“△”) represent the positions of sensor data in the feature space during new substrate transport, and dashed lines represent the distance to the nearest neighbor k when k=3.

[0109] In the seventh method, the estimation unit 121 takes a dataset containing sensor data during the transfer of a new substrate as input, estimates the transfer anomaly degree based on the distance to the k nearest neighbor calculated by the learning model 120, and outputs the result. Alternatively, the transfer anomaly degree can be the distance to the k nearest neighbor calculated by the learning model 120, or it can be a value that is uniquely transformed to the k nearest neighbor using a specified function.

[0110] Figure 13 This is a schematic diagram illustrating the structure of the learning model 120 in the eighth approach. Figure 13 The completed learning model 120 shown is a tuned neural network-like system and contains: an input layer; one or more intermediate layers connected to the input layer; and an output layer connected to the intermediate layers, forming a hierarchical neural network or quantum neural network (QNN). The completed learning model 120 may also contain a multi-layered neural network with two or more intermediate layers, i.e., deep learning.

[0111] The eighth method completes the learning model 120 by using a dataset containing sensor data from past normal substrate handling as teacher data and performing machine learning with LSTM (Long Short-Term Memory). Figure 13 In the example shown, the learning model 120, for example, predicts a dataset containing sensor data (predicted values ​​of sensor data in the nth period) during the transport of the new substrate, based on a dataset containing actual sensor data (sensor data from the 1st to the n-1th period) from the time of maintenance until the new substrate is transported.

[0112] In the eighth method, the estimation unit 121 takes as input a dataset containing actual sensor data (sensor data for cycles 1 to n) from the time of maintenance to the time of new substrate transport. From a dataset containing actual sensor data before the new substrate transport (sensor data for cycles 1 to n-1), it calculates the deviation between a dataset containing the predicted sensor data (predicted values ​​of sensor data for cycle n) during the new substrate transport, as predicted by the learning model 120, and a dataset containing the actual sensor data (sensor data for cycle n) during the new substrate transport. Based on this deviation, it estimates the transport anomaly degree and outputs it. Alternatively, the transport anomaly degree can be the deviation itself, or it can be a value that uniquely transforms the deviation using a specified function.

[0113] The estimation unit 121 has multiple complete learning models 120 (e.g., two or more of the complete learning models 120 in the first to eighth modes), and can also estimate the transport anomaly degree and output it based on the combination of predictions from multiple complete learning models 120 (i.e., overall learning).

[0114] The output signal transmitting unit 122 compares the transport anomaly degree output by the estimation unit 121 with the predetermined threshold 152. If the transport anomaly degree exceeds the threshold 152, the output signal used to output maintenance notification and / or alarm is sent to the output device 4 via the output unit 14.

[0115] The relearning unit 123 uses the dataset 131 containing sensor data output during the transfer of a new substrate as teacher data to enable the completed learning model 120 to relearn.

[0116] Next, an example of a transport anomaly prediction method implemented by the transport anomaly prediction system 10 configured in this way will be explained. Figure 14 This is a flowchart illustrating an example of a transport anomaly prediction method.

[0117] like Figure 14 As shown, the estimation unit 121 acquires a dataset 131 containing sensor data output from multiple sensors 51 to 5N provided on the substrate transfer unit 2 during the transfer of a new substrate (step S11). The dataset 131 acquired by the estimation unit 121 may also include at least one of the following time information: equipment operation time during substrate transfer, time elapsed after maintenance, and component usage time of the substrate processing unit. The dataset 131 acquired by the estimation unit 121 is stored in the storage unit 13.

[0118] Then, the estimation unit 121 uses a completed learning model 120, which performs machine learning on a dataset containing sensor data output from multiple sensors 51 to 5N during past substrate transport and the relationship between the dataset and the transport anomaly degree during the substrate transport, and takes a dataset 131 containing sensor data output from multiple sensors 51 to 5N during new substrate transport as input, to estimate the transport anomaly degree during the new substrate transport and output it (step S12).

[0119] In step S12, the estimation unit 121 may also use a completed learning model 120 (see reference) which is the result of machine learning on a dataset containing sensor data output from multiple sensors 51 to 5N during past substrate transport, and the relationship between the cause (type) of each transport anomaly and the degree of transport anomaly during the substrate transport. Figure 8 , 9 The dataset 131, which contains sensor data output from multiple sensors 51 to 5N during the transport of a new substrate, is used as input. The dataset estimates the degree of transport abnormality during the transport of the new substrate based on the cause (type) of each transport abnormality and outputs the result.

[0120] Next, the output signal transmitting unit 122 compares the transport anomaly degree output by the estimation unit 121 with the predetermined threshold 152 (step S13).

[0121] If the transport anomaly degree output by the estimation unit 121 exceeds the threshold 152 (step S13: YES), the output signal transmission unit 122 will send the output signal for outputting maintenance notification and / or alarm to the output device 4 via the output unit 14 (step S14).

[0122] In addition, if the abnormality of the transport output by the estimation unit 121 does not exceed the threshold 152 (step S13: No), the relearning unit 123 uses the dataset 131 containing the sensor data output when the new substrate is transported as teacher data with the label attached when it is normal, so that the completed learning model 120 can be relearned (step S15).

[0123] In this embodiment described above, the estimation unit 121 uses a completed learning model 120, which performs machine learning on a dataset containing sensor data from past substrate transports and the relationship between this dataset and the transport anomaly degree during those past substrate transports, to synthesize multiple index data from the dataset 131 containing sensor data from the new substrate transport and output the transport anomaly degree during the new substrate transport. This improves the detection probability of transport anomalies compared to the past method of judging anomalies as abnormal when the difference between the sensor output during substrate transport and the sensor output stored when the substrate was aligned before transport is greater than a certain threshold. Furthermore, by using the completed learning model 120, vibration, sound, and image data from devices that were difficult to process using the past method can also be used as sensor data.

[0124] Furthermore, when this embodiment is adopted, when the transport abnormality estimated by the estimation unit 121 exceeds the threshold 152, the output signal transmission unit 122 sends an output signal for outputting maintenance notification and / or alarm to the output device 4. Therefore, the output from the output device 4 is used as a trigger, and maintenance is performed by the user (e.g., the operator of the board processing device 1), which can prevent processing errors in advance.

[0125] Furthermore, when this embodiment is adopted, since the relearning unit 123 uses the dataset 131 containing the sensor data output when the new substrate is transported as teacher data to relearn the completed learning model 120, a system that follows changes in the operating status of the device can be obtained.

[0126] In addition, adopt Figure 8 , 9 In the illustrated embodiment, since the estimation unit 121 takes the dataset containing sensor data from past substrate transport as input, and estimates the transport abnormality degree of the new substrate transport based on the relationship between the dataset containing sensor data from past substrate transport and the causes (abnormality types) of each transport abnormality, the estimation unit 121 estimates the transport abnormality degree of the new substrate transport based on the causes (abnormality types) of each transport abnormality and outputs it out. By investigating the causes (abnormality types) of transport abnormalities that are estimated to have a high transport abnormality degree, the maintenance time of the device can be shortened.

[0127] Furthermore, the transport anomaly prediction system 10 of this embodiment can be constructed by a single computer or quantum computing system, or by multiple computers or quantum computing systems interconnected via a network. However, in one or more computers or quantum computing systems, the program used to implement the transport anomaly prediction system 10 and the computer-readable recording medium that records the program in a non-transitory manner are also protected by this invention.

[0128] The above embodiments and modifications have been illustrated by way of example. However, the scope of this technology is not limited to these examples, and changes and modifications can be made according to the purpose within the scope of the claims. Furthermore, various embodiments and modifications can be appropriately combined without causing contradictions in the processing content.

Claims

1. A conveying anomaly prediction system, characterized in that, The system includes an estimation unit with a completion learning model. This model performs machine learning on the relationship between a dataset containing sensor data output from multiple sensors located in the substrate transport section during past substrate transports and the transport anomaly degree during those past transports. The estimation unit takes a dataset containing sensor data output from the multiple sensors during new substrate transports as input, estimates the transport anomaly degree during the new substrate transport, and outputs the result. The completed learning model uses a dataset containing sensor data from past normal substrate transport as teacher data and performs machine learning using LSTM (Long Short-Term Memory). The estimation unit takes a dataset containing actual sensor data up to the new substrate transport as input, calculates the deviation between the dataset containing the sensor data predicted by the completed learning model for the new substrate transport and the dataset containing the actual sensor data for the new substrate transport, and the dataset containing the actual sensor data for the new substrate transport, based on the deviation, and estimates the transport anomaly degree and outputs it.

2. A conveying anomaly prediction system, characterized in that, The system includes an estimation unit with a completion learning model. This model performs machine learning on the relationship between a dataset containing sensor data output from multiple sensors located in the substrate transport section during past substrate transports and the transport anomaly degree during those past transports. The estimation unit takes a dataset containing sensor data output from the multiple sensors during new substrate transports as input, estimates the transport anomaly degree during the new substrate transport, and outputs the result. The completed learning model performs machine learning on teacher data that includes a dataset containing sensor data from past substrate transports and labels indicating whether a transport anomaly occurred during that substrate transport. The estimation unit takes a dataset containing sensor data from new substrate transports as input, estimates the transport anomaly degree based on the probability of a transport anomaly predicted by the completed learning model, and outputs the result.

3. The conveying anomaly prediction system as described in claim 2, characterized in that, The completed learning model adds a label to the dataset containing sensor data from past substrate transports, indicating whether a transport anomaly occurred during the transport. When a transport anomaly occurs, machine learning is performed on teacher data labeled with the cause of the transport anomaly. The estimation unit takes the dataset containing sensor data from new substrate transports as input, and estimates the transport anomaly degree of each cause of transport anomaly based on the probability of a transport anomaly predicted by the completed learning model for each cause of transport anomaly, and outputs the result.

4. The conveying anomaly prediction system as described in any one of claims 1-3, characterized in that, The plurality of sensors consist of one or more of the following: vibration sensor, sound sensor, image sensor, video sensor, temperature sensor, equipment movement speed sensor, equipment action torque sensor, and equipment parallelism sensor.

5. The conveying anomaly prediction system as described in any one of claims 1-3, characterized in that, The device further includes an output signal transmitting unit that compares the transport anomaly degree output by the estimation unit with a predetermined threshold. If the transport anomaly degree exceeds the threshold, the output signal for outputting maintenance notification and / or alarm is sent to the output device.

6. The conveying anomaly prediction system as described in any one of claims 1-3, characterized in that, The estimation unit takes a dataset containing sensor data from the start of the new substrate transport to the present as input, estimates the transport anomaly degree of the new substrate transport, and outputs it.

7. The conveying anomaly prediction system as described in any one of claims 1-3, characterized in that, The estimation unit takes a dataset containing sensor data from the start to the end of the transport of a new substrate as input, estimates the transport anomaly degree during the transport of the new substrate, and outputs it.

8. The conveying anomaly prediction system as described in any one of claims 1-3, characterized in that, The estimation unit takes as input a dataset of sensor data from the start of the transport of the first substrate to the end of the transport of the last substrate when transporting multiple new substrates, estimates the transport anomaly during the transport of the multiple new substrates, and outputs the result.

9. The conveying anomaly prediction system as described in any one of claims 1-3, characterized in that, Furthermore, a relearning unit is provided, which uses the dataset containing sensor data output during the transport of the new substrate as teacher data to enable the completed learning model to relearn.

10. The conveying anomaly prediction system as described in any one of claims 1-3, characterized in that, The dataset further includes at least one of the following time information: equipment operation time during substrate transport, time elapsed after maintenance, and component usage time in the substrate processing unit.

11. The conveying anomaly prediction system as described in any one of claims 1-3, characterized in that, The estimation unit has multiple completed learning models, and estimates and outputs the transport anomaly degree based on a combination of predictions from the multiple completed learning models.

12. A substrate processing apparatus, characterized in that, have: Substrate transport section; and The transport anomaly prediction system according to any one of claims 1-11.

13. A method for predicting transport anomalies, which is a computer-executed method for predicting transport anomalies, characterized in that: Includes the following steps: A completion learning model was used to perform machine learning on the relationship between a dataset containing sensor data output by multiple sensors located in the substrate conveying section during past substrate conveying and the degree of conveying anomalies during the substrate conveying process. Using a dataset containing sensor data output from the multiple sensors during the transfer of the new substrate as input, the transfer anomaly degree during the transfer of the new substrate is estimated and output. The completed learning model uses a dataset containing sensor data from past normal substrate transport as teacher data and performs machine learning using LSTM (Long Short-Term Memory). It takes a dataset containing actual sensor data up to the new substrate transport as input, calculates the deviation between the dataset containing the sensor data predicted by the completed learning model for the new substrate transport and the dataset containing the actual sensor data for the new substrate transport, and the dataset containing the actual sensor data for the new substrate transport, and infers and outputs the transport anomaly degree based on the deviation.

14. A method for predicting transport anomalies, which is a computer-executed method for predicting transport anomalies, characterized in that: Includes the following steps: A completion learning model was used to perform machine learning on the relationship between a dataset containing sensor data output by multiple sensors located in the substrate conveying section during past substrate conveying and the degree of conveying anomalies during the substrate conveying process. Using a dataset containing sensor data output from the multiple sensors during the transfer of the new substrate as input, the transfer anomaly degree during the transfer of the new substrate is estimated and output. The completed learning model performs machine learning on teacher data that includes sensor data from past substrate transports and labels indicating whether transport anomalies occurred during the transport. It takes a dataset containing sensor data from new substrate transports as input, estimates the transport anomaly degree based on the probability of transport anomalies predicted by the completed learning model, and outputs the result.

15. A computer-readable recording medium containing a transport anomaly prediction program for causing a computer to perform the following steps: A fully learned machine learning model is used to study the relationship between a dataset containing sensor data output by multiple sensors located in the substrate transport section during past substrate transports and the transport anomaly degree during the current substrate transport. This model takes a dataset containing sensor data output by the same multiple sensors during a new substrate transport as input, estimates the transport anomaly degree during the new substrate transport, and outputs the result. The completed learning model uses a dataset containing sensor data from past normal substrate transport as teacher data and performs machine learning using LSTM (Long Short-Term Memory). It takes a dataset containing actual sensor data up to the new substrate transport as input, calculates the deviation between the dataset containing the sensor data predicted by the completed learning model for the new substrate transport and the dataset containing the actual sensor data for the new substrate transport, and the dataset containing the actual sensor data for the new substrate transport, and infers and outputs the transport anomaly degree based on the deviation.

16. A computer-readable recording medium containing a transport anomaly prediction program for causing a computer to perform the following steps: A fully learned machine learning model is used to study the relationship between a dataset containing sensor data output by multiple sensors located in the substrate transport section during past substrate transports and the transport anomaly degree during the current substrate transport. This model takes a dataset containing sensor data output by the same multiple sensors during a new substrate transport as input, estimates the transport anomaly degree during the new substrate transport, and outputs the result. The completed learning model performs machine learning on teacher data that includes sensor data from past substrate transports and labels indicating whether transport anomalies occurred during the transport. It takes a dataset containing sensor data from new substrate transports as input, estimates the transport anomaly degree based on the probability of transport anomalies predicted by the completed learning model, and outputs the result.

17. A completion learning model, characterized in that, The estimation unit uses a completion learning model. This model takes as input a dataset containing actual sensor data output by multiple sensors located in the substrate transport unit before a new substrate is transported. Based on this dataset containing the actual sensor data before the new substrate transport, the estimation unit calculates the deviation between a dataset containing the sensor data predicted by the completion learning model for the new substrate transport and a dataset containing the actual sensor data for the new substrate transport. Based on this deviation, the estimation unit estimates and outputs the transport anomaly degree for the new substrate transport. The completed learning model has: an input layer; one or more intermediate layers connected to the input layer; and an output layer connected to the intermediate layers. The completed learning model uses a dataset containing sensor data from past normal substrate handling as teacher data and performs machine learning using LSTM (Long Short-Term Memory). The completed learning model enables the computer to perform the following functions: when a dataset containing actual sensor data before the new substrate is transported is input into the input layer, predict the sensor data during the transport of the new substrate, and output it from the output layer.

18. A completion learning model, characterized in that, The estimation unit uses a completion learning model, which takes as input a dataset containing sensor data output by multiple sensors installed in the substrate transport unit during the transport of a new substrate, and estimates and outputs the transport anomaly degree during the transport of the new substrate based on the probability of transport anomaly predicted by the completion learning model. The completed learning model has: an input layer; one or more intermediate layers connected to the input layer; and an output layer connected to the intermediate layers. The completed learning model performs machine learning on teacher data, which includes sensor data from past substrate handling and labels indicating whether any handling anomalies occurred during the substrate handling process. When a dataset containing sensor data during the transport of a new substrate is input into the input layer, the probability of a transport anomaly occurring during the transport of the new substrate is predicted and output from the output layer.

Images