Model generation method, abnormality factor estimation device and method, substrate processing device, learning method, device and data creation method

The model generated by machine learning, combined with the feature quantities of the substrate processing device and the main cause information of the anomaly, solves the problem of insufficient accuracy in the existing anomaly diagnosis system and achieves high-precision and rapid anomaly cause estimation.

CN113448307BActive Publication Date: 2024-11-12SCREEN HOLDINGS CO LTD
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
CN202110205886.3
Authority / Receiving Office
CN · China
Patent Type
Patents(China)
Current Assignee / Owner
Priority Date
2020-03-24
Filing Date
2021-02-24
Publication Date
2024-11-12
Estimated Expiration
2041-02-24

AI Technical Summary

Technical Problem

Existing anomaly diagnosis systems suffer from insufficient precision in model data and judgment conditions when estimating the cause of anomalies in process handling devices, resulting in an inability to accurately determine the cause of anomalies. Furthermore, they rely heavily on the skills and experience of inspection technicians, leading to a lengthy process.

Method used

By acquiring learning data and performing machine learning, a learned model is generated to infer the main causes of substrate anomalies after fluid processing. The feature information includes first feature information and main cause information of the anomaly. Combined with the object and component information of the substrate processing device, high-precision anomaly cause inference is achieved.

Benefits of technology

This improves the accuracy of identifying the main causes of substrate abnormalities after processing with fluids and shortens the time required to determine the causes of abnormalities.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN113448307B_ABST
    Figure CN113448307B_ABST
Patent Text Reader

Abstract

The present invention provides a method for generating a model after learning, a device for estimating the main cause of abnormality, a substrate processing device, a method for estimating the main cause of abnormality, a learning method, a learning device, and a method for producing learning data. The method for generating a model after learning includes: a step of acquiring learning data; and a step of generating a model after learning by performing machine learning on the learning data, wherein the model after learning estimates the main cause of abnormality of a processing object substrate after being processed by a processing fluid. The learning data includes a feature quantity and information on the main cause of abnormality. The information on the main cause of abnormality indicates the main cause of abnormality of the learning object substrate after being processed by the processing fluid. The feature quantity includes first feature quantity information representing the time-lapse feature of interval data in time series data, wherein the time series data represents the physical quantity of an object used by a substrate processing device that processes the learning object substrate using a processing fluid. The first feature quantity information is represented by time.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present invention relates to a method for generating a model after learning, a device for estimating the main cause of an abnormality, a substrate processing device, a method for estimating the main cause of an abnormality, a learning method, a learning device, and a method for producing learning data. Background Art

[0002] The abnormality diagnosis system described in Patent Document 1 detects abnormality occurring in a resist coating and developing processing device and estimates the cause of the abnormality.

[0003] The abnormality diagnosis system is connected to sensors respectively installed in a plurality of process processing devices through wiring. The sensors output time series data that changes over time to the abnormality diagnosis system. For example, the sensors are pressure sensors, temperature sensors, flow sensors, liquid level sensors, position sensors, torque sensors, or speed sensors.

[0004] The abnormality diagnosis system includes a data holding unit, a data collecting unit, an abnormality detecting unit, a determination data generating unit, and a diagnosis unit.

[0005] The data storage unit stores time series data, abnormality data, determination data, and model data.

[0006] The data collection unit collects time-varying time series data from sensors installed in a plurality of process processing devices, and stores the time series data in the data storage unit. The time series data may be, for example, pressure, temperature, flow rate, liquid level, position, torque or speed that varies with time.

[0007] The abnormality detection unit compares the time series data collected by the data collection unit with the up / down fluctuation threshold data, thereby detecting abnormal data of the time series data that is abnormal.

[0008] The judgment data production unit divides the abnormal data detected by the abnormality detection unit into multiple monitoring intervals in time series based on the judgment condition data, judges whether it is consistent with multiple judgment conditions in each monitoring interval, and produces judgment data including a combination of judgment results, which is stored in the data retention unit as judgment data.

[0009] The diagnosis unit collates the determination data created by the determination data creation unit with the model data corresponding to the determination data, thereby estimating the cause of the abnormality of the process processing device.

[0010] The model data is the result of judging the time series data that has a change tendency unique to each abnormal cause using the judgment conditions when creating the judgment data, and is stored in advance. Moreover, the threshold value and range of each judgment condition are determined for each monitoring interval.

[0011] [Prior art literature]

[0012] [Patent Document]

[0013] [Patent Document 1] Japanese Patent Application Publication No. 2012-150721 Summary of the invention

[0014] [Problems to be solved by the invention]

[0015] However, the abnormality diagnosis system described in Patent Document 1 may not be able to accurately estimate the cause of the abnormality of the process processing device, depending on the accuracy of the model data and the determination conditions. In this case, the inspection technician needs to analyze the time series data to determine the cause of the abnormality of the process processing device.

[0016] However, the timing data is very complicated and it is difficult for the inspection technician to find the regularity. In addition, the timing data has many parameters and the analysis object is huge. Therefore, depending on the skills and experience of the inspection technician, it sometimes takes a long time to determine the main cause of the abnormality of the process processing device. As a result, it sometimes takes a long time to determine the main cause of the abnormality of the substrate processed by the process processing device.

[0017] The present invention is made in view of the above-mentioned problems, and its purpose is to provide a model generation method that has been learned, an abnormality main cause inference device, a substrate processing device, an abnormality main cause inference method, a learning method, a learning device and a learning data production method, which can determine the main cause of the abnormality of the processing object substrate after being processed by the processing fluid with high precision, and can shorten the time for determining the main cause of the abnormality.

[0018] [Technical means to solve the problem]

[0019] According to one aspect of the present invention, a method for generating a learned model includes the following steps: acquiring learning data; and generating a learned model by performing machine learning on the learning data, wherein the learned model infers the main cause of abnormality of a processing target substrate after being processed by a processing fluid. The learning data includes a feature quantity and information on the main cause of abnormality. The information on the main cause of abnormality indicates the main cause of abnormality of the learning target substrate after being processed by the processing fluid. The feature quantity includes first feature quantity information, which is a feature indicating the time passage of interval data in time series data, and the time series data indicates the physical quantity of an object used by a substrate processing device that processes the learning target substrate using the processing fluid. The first feature quantity information is represented by time.

[0020] In the learning-completed model generation method of the present invention, the abnormal main cause information preferably includes at least one of first abnormal main cause information and second abnormal main cause information. The first abnormal main cause information is preferably information related to the processing fluid when the substrate processing device processes the learning object substrate. The second abnormal main cause information is preferably information related to the parts of the substrate processing device.

[0021] In the learning-completed model generation method of the present invention, the first feature quantity information preferably includes at least one of the first information, the second information, the third information, the fourth information, the fifth information and the sixth information. The first information is preferably information indicating the state of the physical quantity when the physical quantity increases toward the target value. The second information is preferably information indicating the overshoot of the physical quantity. The third information is preferably information indicating the change of the physical quantity. The fourth information is preferably information indicating the state of the physical quantity when the physical quantity decreases from the target value. The fifth information is preferably information indicating the overlap of the physical quantity of one of the two objects with the physical quantity of the other object when the substrate processing device uses at least two different objects. The sixth information is preferably information indicating the time interval between one of the objects and the other object on the time axis.

[0022] In the learned model generation method of the present invention, the object is preferably the treatment fluid, and the physical quantity preferably represents the physical quantity of the treatment fluid in the discharge pipe from which the treatment fluid is discharged.

[0023] In the learned model generation method of the present invention, the object is preferably the treatment fluid, and the physical quantity preferably represents the flow rate of the treatment fluid, the temperature of the treatment fluid, or the concentration of the treatment fluid.

[0024] In the learning-completed model generation method of the present invention, the object is preferably a substrate holding portion that rotates the learning object substrate, an arm that moves a nozzle that sprays the processing fluid toward the learning object substrate, a guard that receives the processing fluid scattered from the learning object substrate, or a valve that regulates the flow of the processing fluid. In the case where the object is the substrate holding portion, the physical quantity preferably represents a quantity related to the action and / or state of the substrate holding portion. In the case where the object is the arm, the physical quantity preferably represents a quantity related to the action and / or state of the arm. In the case where the object is the guard, the physical quantity preferably represents a quantity related to the action and / or state of the guard. In the case where the object is the valve, the physical quantity preferably represents a quantity related to the action and / or state of the valve.

[0025] In the method for generating a learned model of the present invention, the object preferably includes a blocking plate covering the upper surface of the learning object substrate with a gap. When the object is the blocking plate, the physical quantity preferably represents a quantity related to the action and / or state of the blocking plate.

[0026] In the learning-completed model generation method of the present invention, the feature quantity preferably further includes second feature quantity information. The second feature quantity information is preferably processing sequence information indicating the processing sequence of a predetermined number of the learning target substrates constituting one batch, or batch interval information indicating the time interval from the end of processing a batch to the start of processing a next batch.

[0027] According to another aspect of the present invention, an abnormality main cause estimation device estimates the main cause of abnormality of a processing target substrate after being processed by a processing fluid. The abnormality main cause estimation device includes a storage unit and an estimation unit. The storage unit stores a learning model constructed by machine learning of learning data. The estimation unit inputs input information to the learning model and obtains output information from the learning model. The learning data includes a feature quantity and abnormality main cause information. The abnormality main cause information indicates the main cause of abnormality of the learning target substrate after being processed by the processing fluid. The feature quantity includes first feature quantity information, the first feature quantity information is a feature of time transition of interval data in time series data, the time series data indicates a physical quantity of an object used by a substrate processing device that processes the learning target substrate using the processing fluid, and the first feature quantity information is represented by time. The input information includes first input information, the first input information indicates a feature of time transition of interval data in time series data, the time series data indicates a physical quantity of an object used by a substrate processing device that processes the processing target substrate using the processing fluid, and the first input information is represented by time. The output information indicates the main cause of abnormality of the processing target substrate after being processed by the processing fluid.

[0028] In the abnormal main cause inference device of the present invention, the abnormal main cause information preferably includes at least one of the first abnormal main cause information and the second abnormal main cause information. The first abnormal main cause information is preferably information related to the processing fluid when the substrate processing device processes the learning object substrate. The second abnormal main cause information is preferably information related to the parts of the substrate processing device that processes the learning object substrate. The output information preferably includes at least one of the first output information and the second output information. The first output information is preferably information related to the processing fluid when the substrate processing device processes the processing object substrate. The second output information is preferably information related to the parts of the substrate processing device that processes the processing object substrate.

[0029] In the abnormal main cause estimation device of the present invention, the first characteristic quantity information and the first input information each preferably include at least one of the first information, the second information, the third information, the fourth information, the fifth information and the sixth information. The first information is preferably information indicating the state of the physical quantity when the physical quantity increases toward the target value. The second information is preferably information indicating the overshoot of the physical quantity. The third information is preferably information indicating the change of the physical quantity. The fourth information is preferably information indicating the state of the physical quantity when the physical quantity decreases from the target value. The fifth information is preferably information indicating the overlap of the physical quantity of one of the two objects with the physical quantity of the other object when the substrate processing device uses at least two different objects. The sixth information is preferably information indicating the time interval between one of the objects and the other object on the time axis.

[0030] In the abnormality cause estimation device of the present invention, the object is preferably the treatment fluid. The physical quantity is preferably a physical quantity of the treatment fluid in a discharge pipe from which the treatment fluid is discharged.

[0031] In the abnormality cause estimation device of the present invention, the object is preferably the treatment fluid. The physical quantity preferably represents the flow rate of the treatment fluid, the temperature of the treatment fluid, or the concentration of the treatment fluid.

[0032] In the abnormality main cause estimation device of the present invention, the object used in the substrate processing device that processes the learning object substrate is preferably a substrate holding portion that rotates the learning object substrate, an arm that moves the nozzle that sprays the processing fluid to the learning object substrate, a guard plate that receives the processing fluid scattered from the learning object substrate, or a valve that regulates the flow of the processing fluid. The object used in the substrate processing device that processes the processing object substrate is preferably a substrate holding portion that rotates the processing object substrate, an arm that moves the nozzle that sprays the processing fluid to the processing object substrate, a guard plate that receives the processing fluid scattered from the processing object substrate, or a valve that regulates the flow of the processing fluid. In the case where the object is the substrate holding portion, the physical quantity preferably represents a quantity related to the action and / or state of the substrate holding portion. In the case where the object is the arm, the physical quantity preferably represents a quantity related to the action and / or state of the arm. In the case where the object is the guard plate, the physical quantity preferably represents a quantity related to the action and / or state of the guard plate. When the object is the valve, the physical quantity preferably represents a quantity related to an action and / or a state of the valve.

[0033] In the abnormality main cause estimation device of the present invention, the object used by the substrate processing device for processing the learning object substrate preferably includes a blocking plate covering the upper surface of the learning object substrate with a gap. The object used by the substrate processing device for processing the processing object substrate preferably includes a blocking plate covering the upper surface of the processing object substrate with a gap. In the case where the object is the blocking plate, the physical quantity preferably represents a quantity related to the action and / or state of the blocking plate.

[0034] In the abnormality main cause estimation device of the present invention, the characteristic quantity preferably further includes second characteristic quantity information. The second characteristic quantity information is preferably processing sequence information indicating the processing sequence of the predetermined number of the learning object substrates constituting one batch, or batch interval information indicating the time interval from the end of processing of a batch to the start of processing of the next batch. The input information preferably further includes second input information related to the processing object substrate. The second input information is preferably processing sequence information indicating the processing sequence of the predetermined number of the processing object substrates constituting one batch, or batch interval information indicating the time interval from the end of processing of a batch to the start of processing of the next batch.

[0035] According to still another aspect of the present invention, a substrate processing apparatus includes the abnormality cause estimating apparatus and a processing apparatus that processes a substrate.

[0036] According to another aspect of the present invention, the model completed by learning is constructed by machine learning of learning data, and a computer is made to function in a manner of estimating the main cause of abnormality of a processing target substrate after being processed by a processing fluid. The learning data includes a feature quantity and information on the main cause of abnormality. The information on the main cause of abnormality indicates the main cause of abnormality of the learning target substrate after being processed by the processing fluid. The feature quantity includes feature quantity information, and the feature quantity information indicates the time-lapse feature of interval data in time series data, and the time series data indicates the physical quantity of an object used by a substrate processing device that processes the learning target substrate using the processing fluid. The feature quantity information is represented by time. The computer is made to function in a manner of inputting input information and outputting output information. The information included in the input information indicates the time-lapse feature of interval data in time series data, and the time series data indicates the physical quantity of an object used by a substrate processing device that processes the processing target substrate using the processing fluid. The information is represented by time. The output information indicates the main cause of abnormality of the processing target substrate after being processed by the processing fluid.

[0037] According to another aspect of the present invention, a method for estimating a main cause of abnormality estimates a main cause of abnormality of a substrate to be processed after being processed by a processing fluid. The method for estimating a main cause of abnormality includes the following steps: obtaining input information; and inputting the input information into a model constructed by machine learning of learning data, and obtaining output information from the model. The learning data includes a feature quantity and main cause of abnormality information. The main cause of abnormality information indicates the main cause of abnormality of the substrate to be processed after being processed by the processing fluid. The feature quantity includes feature quantity information, the feature quantity information indicates the time-lapse feature of interval data in time series data, the time series data indicates the physical quantity of an object used by a substrate processing device that processes the substrate to be processed using the processing fluid, and the feature quantity information is represented by time. The input information includes information indicating the time-lapse feature of interval data in time series data, the time series data indicates the physical quantity of an object used by a substrate processing device that processes the substrate to be processed using the processing fluid, and the information is represented by time. The output information indicates the main cause of abnormality of the substrate to be processed after being processed by the processing fluid.

[0038] According to yet another aspect of the present invention, a learning method includes the following steps: acquiring learning data; and performing machine learning on the learning data. The learning data includes a feature quantity and information on a main cause of an abnormality. The information on the main cause of an abnormality indicates the main cause of an abnormality of a learning object substrate after being processed by a processing fluid. The feature quantity includes feature quantity information, the feature quantity information indicating a time-lapse feature of interval data in time series data, the time series data indicating a physical quantity of an object used by a substrate processing device that processes the learning object substrate using the processing fluid. The first feature quantity information is represented by time.

[0039] According to yet another aspect of the present invention, a learning device includes a storage unit and a learning unit. The storage unit stores learning data. The learning unit performs machine learning on the learning data. The learning data includes a feature quantity and information on a main cause of an abnormality. The information on the main cause of an abnormality indicates the main cause of an abnormality of a learning object substrate after being processed by a processing fluid. The feature quantity includes feature quantity information indicating a feature of a time lapse of interval data in time series data, and the time series data indicates a physical quantity of an object used by a substrate processing device that processes the learning object substrate using the processing fluid. The feature quantity information is represented by time.

[0040] According to yet another aspect of the present invention, a learning data production method includes the following steps: acquiring at least one interval data from time series data, the time series data representing a physical quantity of an object used in a substrate processing device that processes a learning object substrate using a processing fluid; calculating feature quantity information representing a feature of the time passage of the interval data; and controlling a storage unit in a manner that stores abnormality main cause information associated with the feature quantity information. The abnormality main cause information represents the main cause of the abnormality of the learning object substrate after being processed by the processing fluid. The feature quantity information is represented by time. The feature quantity information and the abnormality main cause information constitute learning data that is the object of machine learning.

[0041] [Effects of the Invention]

[0042] According to the present invention, the main cause of abnormality of a substrate to be processed after being processed by a processing fluid can be determined with high accuracy, and the time required to determine the main cause of the abnormality can be shortened. BRIEF DESCRIPTION OF THE DRAWINGS

[0043] Figure 1 It is a block diagram showing a substrate processing system according to an embodiment of the present invention.

[0044] Figure 2 It is a schematic plan view showing a substrate processing apparatus according to this embodiment.

[0045] Figure 3Schematic cross-sectional view of a substrate processing apparatus showing the present embodiment.

[0046] FIG. 4( a ) is a diagram showing a fluid supply unit according to the present embodiment, and FIG. 4( b ) is a diagram showing another fluid supply unit according to the present embodiment.

[0047] Figure 5 It is a figure which shows the substrate processing apparatus of this embodiment.

[0048] Figure 6 1 is a block diagram showing a control device of the substrate processing apparatus according to the present embodiment.

[0049] Figure 7 It is a flowchart showing the substrate processing method according to the present embodiment.

[0050] Figure 8 It is a block diagram showing the learning data creation device according to the present embodiment.

[0051] Fig. 9 This is a graph showing an example of time series data in this embodiment.

[0052] Fig.10 It is a diagram showing the characteristic quantities of this embodiment.

[0053] Figure 11(a) is a diagram showing an example of a method for calculating the first information contained in the first feature information of the present embodiment, Figure 11(b) is a diagram showing another example of a method for calculating the first information, and Figure 11(c) is a diagram showing yet another example of a method for calculating the first information.

[0054] Fig.12 It is a diagram showing an example of a method for calculating the second information included in the first feature amount information according to the present embodiment.

[0055] Fig.13 This is a diagram showing an example of a method for calculating the third information included in the first feature amount information according to the present embodiment.

[0056] Figure 14(a) is a diagram showing an example of a method for calculating the fourth information contained in the first feature information of the present embodiment, Figure 14(b) is a diagram showing another example of a method for calculating the fourth information, and Figure 14(c) is a diagram showing yet another example of a method for calculating the fourth information.

[0057] Fig.15 This is a diagram showing an example of a method for calculating fifth information included in first feature amount information according to the present embodiment.

[0058] Fig.16 It is a diagram showing an example of a method for calculating the sixth information included in the first feature amount information according to the present embodiment.

[0059] Fig.17 It is a diagram showing abnormality cause information according to this embodiment.

[0060] Fig.18 It is a flowchart showing the learning data creation method according to the present embodiment.

[0061] Fig.19 It is a block diagram showing the learning device according to this embodiment.

[0062] Fig. 20 It is a flowchart which shows the learning method of this embodiment.

[0063] Fig.21 1 is a block diagram showing the abnormality cause estimating device according to the present embodiment.

[0064] Fig. 22 It is a diagram showing input information in this embodiment.

[0065] Fig.23 It is a diagram showing output information according to this embodiment.

[0066] Fig.24 It is a flowchart which shows the abnormality main cause estimation method of this embodiment.

[0067] [Explanation of Symbols]

[0068] 1: Processing device

[0069] 3: Control device

[0070] 3A: Control Department

[0071] 3B, 4B, 5B, 6B: Storage

[0072] 3C, 4C, 5C, 6C: Communications Department

[0073] 3D, 4D, 5D, 6D: Input

[0074] 3E, 4E, 5E, 6E: Display unit

[0075] 3F: Flow meter group

[0076] 3G: Temperature sensor cluster

[0077] 3H: Sensor Cluster

[0078] 4: Processing fluid tank

[0079] 4A, 5A, 6A: Processing Department

[0080] 5: Fluid handling cabinet

[0081] 11: Chamber

[0082] 13: Rotating chuck

[0083] 15: Rotating motor

[0084] 17, 21, 25, 27, 295, 297, NZ: Nozzle

[0085] 19, 23: Nozzle moving part

[0086] 29: Fluid supply unit

[0087] 31: Unit Action

[0088] 33: Guard

[0089] 50: Processing fluid tank

[0090] 51: Circulation piping

[0091] 52: Upstream piping

[0092] 53: Individual piping

[0093] 54: Downstream piping

[0094] 55: Pump

[0095] 56: Pulsation Damper

[0096] 57: Filter

[0097] 58: Temperature regulator

[0098] 61: Data Production Department

[0099] 63, 75, 85: Storage control unit

[0100] 65, 77, 87: Display control unit

[0101] 71, 611, 811: Acquisition

[0102] 73: Learning Department

[0103] 81: Pre-processing unit

[0104] 83: Presumption Department

[0105] 100: Substrate processing system

[0106] 130: Chuck component

[0107] 131: Rotating base

[0108] 150: Motor body

[0109] 151: Axis

[0110] 191, 231, AM: Arm

[0111] 193, 233: Rotation axis

[0112] 195, 235: Nozzle moving mechanism

[0113] 200, 300: Substrate processing equipment

[0114] 291: Blocking plate

[0115] 293: Axis

[0116] 400: Learning data production device

[0117] 500: Learning device

[0118] 600: Abnormality main cause estimation device

[0119] 613, 813: Feature calculation unit

[0120] A1, A2: Area

[0121] AX: Axis of rotation

[0122] a~q: information

[0123] CD: Processing condition information

[0124] CR: Central Robot

[0125] FL, FL1, FL2: Flow

[0126] FW, FW1, FW2, FW3: Flow meter

[0127] G1~G6: Piping

[0128] GA: Inert gas

[0129] IDA, IDC: object information

[0130] IDB, IDD: physical quantity information

[0131] IF1: Input information

[0132] IF2: Output information

[0133] IN: intersection

[0134] IR: Indexer Robot

[0135] LM: Learning Complete Model

[0136] LP: Load Port

[0137] LQ1, LQ3, LQ4, LQ5, LQ21, LQ22: Treatment fluid

[0138] MN: Local Minimum

[0139] MX: Maximum

[0140] MXN: Mixed Division

[0141] P1, X1: First Information

[0142] P2, X2: Second information

[0143] P3, X3: Third Information

[0144] P4, X4: The fourth information

[0145] P5, X5: The fifth message

[0146] P6, X6: Sixth Information

[0147] P7: Discharge pipe

[0148] PA, XA: Processing order information

[0149] PB, XB: Batch interval information

[0150] PD1, XD1: First feature quantity information

[0151] PD2, XD2: Second feature quantity information

[0152] PD3, XD3: Third feature quantity information

[0153] PG1: Control Program

[0154] PG2: Learn the data creation process

[0155] PG3: Handler

[0156] PG4: Learning Procedures

[0157] PG5: Presumptive Procedure

[0158] PM: parameter information

[0159] Q101, Q201, Y101, Y201: First major cause information

[0160] Q102, Q202, Y102, Y202: Second main reason information

[0161] Q103, Q203, Y103, Y203: Third major cause information

[0162] Q104, Q204, Y104, Y204: Fourth major cause information

[0163] Q105, Q205, Y105, Y205: Fifth major cause information

[0164] Q106, Q206, Y106, Y206: Sixth major cause information

[0165] Q107, Q207, Y107, Y207: Seventh major cause information

[0166] Q108, Y108: Eighth major cause information

[0167] Q109, Y109: Ninth major cause information

[0168] QD1, YD1: Information on the main cause of the first abnormality

[0169] QD2, YD2: Second abnormality main cause information

[0170] RG1: First regulated range

[0171] RG2: Second regulated range

[0172] RP: Recipe Information

[0173] S1~S10, S21~S25, S31~S34, S41~S49: Steps

[0174] SN: Sensor

[0175] SNS: Detection Group

[0176] SP, SX, SX1, SX2: interval data

[0177] T11~T14、T21~T24、T31~T34、T41~T44、T51、T61:Time interval

[0178] TD, TD1, TD2: time series data

[0179] TD11~TD14: Individual timing data

[0180] TG: target value

[0181] TND: Learning Data

[0182] TS, TS1~TS7: Temperature sensor

[0183] TW: Tower

[0184] t1~t10: time

[0185] ta: When the ascent begins

[0186] tb: When rising, when the rising is suspected to be completed

[0187] tc, te: reference time

[0188] td: When the flow rate changes into the first specified range

[0189] tf: stable state

[0190] tg: when the descent starts

[0191] th: When descending, when descent is suspected to be completed

[0192] tx: When ejection control starts

[0193] ty: When the ejection control stops

[0194] V1~V6: Fluid supply part

[0195] V11, V12, V21, V22, V23, VB: valve

[0196] W: substrate

[0197] W1: Learning Object Substrate

[0198] W2: Processing target substrate

[0199] XD: Feature quantity

[0200] YD: Abnormal main cause information

[0201] θ1, θ2: Tilt angle DETAILED DESCRIPTION

[0202] Hereinafter, the embodiments of the present invention will be described with reference to the accompanying drawings. In addition, the same reference symbols are attached to the same or corresponding parts in the figures, and the description is not repeated. In addition, the X-axis, Y-axis and Z-axis are appropriately illustrated in the figures for easy understanding. The X-axis, Y-axis and Z-axis are orthogonal to each other, the X-axis and Y-axis are parallel to the horizontal direction, and the Z-axis is parallel to the vertical direction. In addition, "looking down" means viewing an object from the vertical direction.

[0203] Reference Figure 1 to Figure 22 , a substrate processing system 100 according to an embodiment of the present invention is described. First, referring to Figure 1 The substrate processing system 100 will be described.

[0204] Figure 1 is a diagram showing a substrate processing system 100. Figure 1 As shown, the substrate processing system 100 of the present embodiment includes a substrate processing apparatus 200 , a substrate processing apparatus 300 , a learning data creating apparatus 400 , a learning apparatus 500 , and an abnormality main cause estimating apparatus 600 .

[0205] The substrate processing apparatus 200 processes the learning target substrate using a processing fluid. The processing fluid is, for example, a processing liquid or a processing gas. The substrate processing apparatus 200 is a single-wafer type that processes the learning target substrates one by one. The learning target substrate is substantially disk-shaped.

[0206] The substrate processing apparatus 300 processes the processing target substrate using a processing fluid. The processing fluid is, for example, a processing liquid or a processing gas. The structure of the processing fluid used in the substrate processing apparatus 300 is the same as the structure of the processing fluid used in the substrate processing apparatus 200. The structure of the processing target substrate is the same as the structure of the learning target substrate. The substrate processing apparatus 300 is a single-chip type that processes the processing target substrate one by one. The processing target substrate is roughly disk-shaped.

[0207] In the following, the learning object substrate is sometimes described as "learning object substrate W1", and the processing object substrate is sometimes described as "processing object substrate W2". In addition, when there is no need to distinguish between the learning object substrate W1 and the processing object substrate W2 for description, the learning object substrate W1 and the processing object substrate W2 are sometimes described as "substrate W".

[0208] The substrate W is, for example, a semiconductor wafer, a substrate for a liquid crystal display device, a substrate for a plasma display, a substrate for a field emission display (FED), a substrate for an optical disk, a substrate for a magnetic disk, a substrate for an optical magneto-optical disk, a substrate for a mask, a ceramic substrate, or a substrate for a solar cell.

[0209] In the present specification, the processing fluid is not particularly limited as long as it is a fluid that contacts the substrate W. The processing liquid as the processing fluid is, for example, a chemical liquid or a rinse liquid.

[0210] The chemical liquid may be, for example, dilute hydrofluoric acid (DHF (Dilute HF)), hydrofluoric acid (HF), nitric fluoric acid (a mixture of hydrofluoric acid and nitric acid (HNO3)), buffered hydrofluoric acid (BHF (Buffered HF)), ammonium fluoride, a mixture of hydrofluoric acid and ethylene glycol (HFEG), phosphoric acid (H3PO4), sulfuric acid, acetic acid, nitric acid, hydrochloric acid, ammonia water, hydrogen peroxide water, organic acid (such as citric acid, oxalic acid), organic base (such as tetramethylammonium hydroxide (TMAH)), sulfuric acid hydrogen peroxide water mixture (SPM), ammonia hydrogen peroxide water mixture (SC1), hydrochloric acid hydrogen peroxide water mixture (SC2), isopropyl alcohol (IPA), surfactant or corrosion inhibitor.

[0211] The eluent is, for example, deionized water, carbonated water, ionized water, hydrogen water, ozone water, or hydrochloric acid water of a diluted concentration (eg, about 10 ppm to 100 ppm).

[0212] Furthermore, the processing gas as the processing fluid is, for example, a reaction gas or an inert gas that reacts with the liquid attached to the substrate W or the substrate W. The reaction gas is, for example, ozone gas, fluorine gas, gas containing hydrogen fluoride, or gas containing IPA. The inert gas is, for example, nitrogen gas, helium gas, or argon gas.

[0213] The substrate processing apparatus 200 outputs timing data TD1. The timing data TD1 indicates the physical quantity of an object used by the substrate processing apparatus 200. The object is, for example, a processing fluid or a part of the substrate processing apparatus 200. The part is not only a hardware part but also a software part. Details of the object and the physical quantity will be described below.

[0214] The learning data production device 400 produces learning data TND based on the time series data TD1. In addition, the learning data production device 400 outputs the learning data TND. The learning data TND is the object of machine learning by the learning device 500. The learning data TND includes feature quantities related to the processing of the learning object substrate W1, and abnormal main cause information related to the processing of the learning object substrate W1. The abnormal main cause information indicates the main cause of the abnormality of the learning object substrate W1 after being processed by the processing fluid. The feature quantity is an explanatory variable, and the abnormal main cause information is a target variable. The details of the feature quantity and the abnormal main cause information will be described below.

[0215] The learning device 500 generates a learned model LM by performing machine learning on the learning data TND. In addition, the learning device 500 outputs the learned model LM. The learned model LM estimates the main cause of the abnormality of the processing object substrate W2 after being processed by the substrate processing device 300 using the processing fluid. Specifically, the learned model LM is a computer program that performs operations from input to output. The computer program constituting the learned model LM includes multiple parameters (e.g., multiple weighting coefficients) for operation.

[0216] The substrate processing apparatus 300 outputs timing data TD2. The timing data TD2 indicates the physical quantity of an object used by the substrate processing apparatus 300. The object is, for example, a processing fluid or a part of the substrate processing apparatus 300. The part is not only a hardware part but also a software part. Details of the object and the physical quantity will be described below.

[0217] In the definition of the timing data TD2, the object used by the substrate processing apparatus 300 corresponds to the object used by the substrate processing apparatus 200. Therefore, the structure of the object used by the substrate processing apparatus 300 is the same as the structure of the object used by the substrate processing apparatus 200. Moreover, in the definition of the timing data TD2, the physical quantity of the object used by the substrate processing apparatus 300 corresponds to the physical quantity of the object used by the substrate processing apparatus 200. Therefore, the physical quantity of the object used by the substrate processing apparatus 300 is the same as the physical quantity of the object used by the substrate processing apparatus 200.

[0218] The abnormality cause estimating apparatus 600 estimates the cause of abnormality of the processing target substrate W2 processed by the substrate processing apparatus 300 using the processing fluid based on the time series data TD2 using the learned model LM.

[0219] Specifically, the abnormality cause estimating device 600 includes a pre-processing unit 81 and an estimating unit 83 .

[0220] The pre-processing unit 81 generates input information IF1 based on the timing data TD2. In addition, the pre-processing unit 81 outputs the input information IF1. The input information IF1 includes a feature quantity related to the processing of the processing target substrate W2. The feature quantity is an explanatory variable. The feature quantity of the input information IF1 is information corresponding to the feature quantity of the learning data TND. The details of the feature quantity will be described below.

[0221] The inference unit 83 inputs the input information IF1 to the learned model LM. In addition, the learned model LM infers the main cause of the abnormality of the processing object substrate W2 after being processed by the substrate processing device 300 using the processing fluid based on the feature quantity contained in the input information IF1, and outputs the output information IF2. The inference unit 83 obtains the output information IF2 from the learned model LM and outputs the output information IF2. The output information IF2 contains the main cause information of the abnormality related to the processing of the processing object substrate W2. The main cause information of the abnormality indicates the main cause of the abnormality of the processing object substrate W2 after being processed by the processing fluid. The main cause information of the abnormality is the target variable. The main cause information of the abnormality of the output information IF2 is information corresponding to the main cause information of the abnormality of the learning data TND. The details of the main cause information of the abnormality will be described below.

[0222] Above, as reference Figure 1As described, according to the present embodiment, the learning device 500 performs machine learning. Therefore, regularity can be found from the time series data TD1 which is very complicated and has a large number of analysis objects, and a highly accurate learning model LM can be produced. In addition, the abnormality main cause estimation device 600 inputs the input information IF1 including the feature quantity based on the time series data TD2 to the learned model LM, and outputs the output information IF2 including the abnormality main cause information from the learned model LM. Therefore, the main cause of the abnormality of the processing object substrate W2 after being processed by the processing fluid can be determined with high accuracy, and the time for determining the main cause of the abnormality can be shortened.

[0223] Here, in this specification, “object” means “object used in the substrate processing apparatus 200 ” or “object used in the substrate processing apparatus 300 ”. Also, “physical quantity” means “physical quantity of an object”. Details of “object” and “physical quantity” will be described below.

[0224] Next, refer to Figure 2 to Figure 7 The substrate processing apparatus 200 is described. The substrate W processed by the substrate processing apparatus 200 is a "learning target substrate W1". In addition, the structure and operation of the substrate processing apparatus 300 are the same as those of the substrate processing apparatus 200. Therefore, refer to Figure 2 to Figure 7 The description of the substrate processing apparatus 200 described above can be referred to the description of the substrate processing apparatus 300. At this time, the substrate W processed by the substrate processing apparatus 300 is referred to as a "processing target substrate W2".

[0225] Figure 2 2 is a schematic plan view showing a substrate processing apparatus 200. Figure 2 As shown, the substrate processing apparatus 200 includes a plurality of load ports LP, an indexer robot IR, a center robot CR, a plurality of processing devices 1, a control device 3, a plurality of processing fluid boxes 4, and a processing fluid cabinet 5. The control device 3 controls the load ports LP, the indexer robot IR, the center robot CR, and the processing devices 1.

[0226] Each loading port LP stacks and accommodates a plurality of substrates W. The indexer robot IR transports the substrates W between the loading port LP and the central robot CR. The central robot CR transports the substrates W between the indexer robot IR and the processing device 1. Each processing device 1 supplies a processing fluid to the substrate W to process the substrate W. Each processing fluid box 4 accommodates a fluid machine. The processing fluid cabinet 5 accommodates the processing fluid.

[0227] Specifically, the plurality of processing devices 1 form a plurality of towers TW (four towers TW in this embodiment) arranged in a manner surrounding the central robot CR when viewed from above. Each tower TW includes a plurality of processing devices 1 stacked up and down (three processing devices 1 in this embodiment). The plurality of processing fluid boxes 4 correspond to the plurality of towers TW, respectively. The processing fluid in the processing fluid cabinet 5 is supplied to all processing devices 1 contained in the tower TW corresponding to the processing fluid box 4 via any processing fluid box 4.

[0228] Figure 3 is a schematic cross-sectional view showing the processing device 1. Figure 3 As shown, the processing device 1 uses a processing fluid to process the substrate W. Specifically, the processing device 1 is a single-wafer type that processes the substrate W one by one.

[0229] Figure 3 In the figure, as an example of a processing fluid, a processing liquid LQ1, a mixed liquid of a processing liquid LQ21 and a processing liquid LQ22, a processing liquid LQ3, a processing liquid LQ4, a processing liquid LQ5 and an inert gas GA are recorded. The processing liquid LQ21 is different from the processing liquid LQ22. The processing liquids LQ1 to LQ5, the processing liquid LQ21 and the processing liquid LQ22 are, for example, the chemical liquids or rinse liquids exemplified above. The inert gas GA is, for example, the inert gas exemplified above. Moreover, the substrate processing device 200 further includes a fluid supply part V1, a fluid supply part V2, a fluid supply part V3, a fluid supply part V4, a fluid supply part V5, a fluid supply part V6, piping G1, piping G2, piping G3, piping G4, piping G5, piping G6, and a temperature sensor TS1, a temperature sensor TS2, a temperature sensor TS3, a temperature sensor TS4, a temperature sensor TS5 and a temperature sensor TS6 for each processing device 1.

[0230] The processing device 1 includes a chamber 11, a rotating chuck 13, a rotating motor 15, a nozzle 17, a nozzle moving part 19, a nozzle 21, a nozzle moving part 23, a nozzle 25, a nozzle 27, a fluid supply unit 29, a unit action part 31 and a plurality of guard plates 33 (four guard plates 33 in the present embodiment).

[0231] The chamber 11 has a substantially box shape. The chamber 11 accommodates the substrate W, the spin chuck 13, the spin motor 15, the nozzle 17, the nozzle moving part 19, the nozzle 21, the nozzle moving part 23, the nozzle 25, the nozzle 27, the fluid supply unit 29, the unit operation part 31, a plurality of guard plates 33, and the temperature sensors TS1 to TS6. In addition, the chamber 11 accommodates a portion of each of the pipes G1, G2, G3, G4, G5, and G6.

[0232] The spin chuck 13 rotates while holding the substrate W. The spin chuck 13 corresponds to an example of a “substrate holding unit.” Specifically, the spin chuck 13 rotates the substrate W around the rotation axis AX while holding the substrate W horizontally in the chamber 11 .

[0233] The rotary chuck 13 includes a plurality of chuck members 130 and a rotary base 131. The plurality of chuck members 130 are disposed on the rotary base 131. The plurality of chuck members 130 hold the substrate W in a horizontal position. The rotary base 131 is roughly disk-shaped and supports the plurality of chuck members 130 in a horizontal position. The rotary motor 15 rotates the rotary base 131 around the rotation axis AX. Therefore, the rotary base 131 rotates around the rotation axis AX. As a result, the substrate W held by the plurality of chuck members 130 disposed on the rotary base 131 rotates around the rotation axis AX. Specifically, the rotary motor 15 includes a motor body 150 and a shaft 151. The shaft 151 is coupled to the rotary base 131. In addition, the motor body 150 rotates the rotary base 131 by rotating the shaft 151.

[0234] The nozzle 17 ejects the processing liquid LQ1 toward the substrate W during the rotation of the substrate W. The pipe G1 supplies the processing liquid LQ1 to the nozzle 17. The fluid supply unit V1 adjusts the supply of the processing liquid LQ1 to the nozzle 17. The temperature sensor TS1 detects the temperature of the processing liquid LQ1 flowing in the pipe G1 and outputs a detection signal indicating the temperature. The temperature sensor TS1 includes, for example, a thermocouple.

[0235] The nozzle moving part 19 moves the nozzle 17 in a substantially vertical direction and a substantially horizontal direction. Specifically, the nozzle moving part 19 includes an arm 191, a rotating shaft 193, and a nozzle moving mechanism 195. The arm 191 extends in a substantially horizontal direction. The nozzle 17 is arranged at the front end of the arm 191. The arm 191 is coupled to the rotating shaft 193. The rotating shaft 193 extends in a substantially vertical direction. The nozzle moving mechanism 195 rotates the rotating shaft 193 around a rotating axis along the vertical direction, so that the arm 191 rotates along a substantially horizontal plane. As a result, the nozzle 17 moves along a substantially horizontal plane. Moreover, the nozzle moving mechanism 195 raises and lowers the rotating shaft 193 in a substantially vertical direction, so that the arm 191 raises and lowers. As a result, the nozzle 17 moves in a substantially vertical direction. The nozzle moving mechanism 195 includes, for example, a ball screw mechanism and an electric motor that provides driving force to the ball screw mechanism.

[0236] The nozzle 21 ejects a mixed liquid of the processing liquid LQ21 and the processing liquid LQ22 toward the substrate W during the rotation of the substrate W. The pipe G2 supplies the mixed liquid to the nozzle 21. The fluid supply unit V2 adjusts the supply of the mixed liquid to the nozzle 21. The temperature sensor TS2 detects the temperature of the mixed liquid flowing in the pipe G2 and outputs a detection signal indicating the temperature. The temperature sensor TS2 includes, for example, a thermocouple.

[0237] The nozzle moving part 23 moves the nozzle 21 in a substantially vertical direction and a substantially horizontal direction. Specifically, the nozzle moving part 23 includes an arm 231, a rotating shaft 233, and a nozzle moving mechanism 235. The arm 231 extends in a substantially horizontal direction. The nozzle 21 is arranged at the front end of the arm 231. The arm 231 is coupled to the rotating shaft 233. The rotating shaft 233 extends in a substantially vertical direction. The nozzle moving mechanism 235 rotates the rotating shaft 233 around a rotating axis along a substantially vertical direction, so that the arm 231 rotates along a substantially horizontal plane. As a result, the nozzle 21 moves along a substantially horizontal plane. Moreover, the nozzle moving mechanism 235 raises and lowers the rotating shaft 233 in a substantially vertical direction, so that the arm 231 raises and lowers. As a result, the nozzle 21 moves in a substantially vertical direction. The nozzle moving mechanism 235 includes, for example, a ball screw mechanism and an electric motor that provides driving force to the ball screw mechanism.

[0238] The nozzle 25 ejects the processing liquid LQ3 toward the substrate W during the rotation of the substrate W. The pipe G3 supplies the processing liquid LQ3 to the nozzle 25. The fluid supply unit V3 adjusts the supply of the processing liquid LQ3 to the nozzle 25. The temperature sensor TS3 detects the temperature of the processing liquid LQ3 flowing in the pipe G3 and outputs a detection signal indicating the temperature. The temperature sensor TS3 includes, for example, a thermocouple.

[0239] The nozzle 27 ejects the processing liquid LQ4 toward the chuck member 130 during the rotation of the substrate W. The pipe G4 supplies the processing liquid LQ4 to the nozzle 27. The fluid supply unit V4 adjusts the supply of the processing liquid LQ4 to the nozzle 27. The temperature sensor TS4 detects the temperature of the processing liquid LQ4 flowing in the pipe G4 and outputs a detection signal indicating the temperature. The temperature sensor TS4 includes, for example, a thermocouple.

[0240] The fluid supply unit 29 is located above the spin chuck 13. The fluid supply unit 29 includes a blocking plate 291, a support shaft 293, a nozzle 295, and a nozzle 297.

[0241] The blocking plate 291 is, for example, roughly disk-shaped. The blocking plate 291 is arranged so that the lower surface of the blocking plate 291 is roughly horizontal. The blocking plate 291 is arranged so that the central axis of the blocking plate 291 is located on the rotation axis AX. The lower surface of the blocking plate 291 faces the substrate W held by the rotation chuck 13. The blocking plate 291 is connected to the lower end of the support shaft 293 in a horizontal posture.

[0242] The unit operation section 31 causes the fluid supply unit 29 to rise or fall between the approach position and the retreat position. The approach position indicates a position where the blocking plate 291 is lowered and close to the upper surface of the substrate W with a predetermined interval. In the approach position, the blocking plate 291 covers the surface of the substrate W and blocks the upper surface of the substrate W. That is, in the approach position, the blocking plate 291 faces the surface of the substrate W and covers the upper surface of the substrate W. The retreat position indicates a position above the approach position, where the blocking plate 291 rises and is away from the substrate W. Figure 3 In the embodiment of the present invention, the blocking plate 291 is located at the retreat position. Moreover, the unit action part 31 is in the approach position, so that the fluid supply unit 29 rotates. For example, the unit action part 31 includes a ball screw mechanism and a lifting motor that provides driving force to the ball screw mechanism. The lifting motor is, for example, a servo motor. For example, the unit action part 31 includes a motor and a transmission mechanism that transmits the rotation of the motor to the fluid supply unit 29.

[0243] The nozzle 295 and the nozzle 297 of the fluid supply unit 29 are arranged inside the blocking plate 291 and the support shaft 293. The front end of the nozzle 295 and the front end of the nozzle 297 are exposed from the lower surface of the blocking plate 291.

[0244] When the fluid supply unit 29 is located at the approach position, the nozzle 295 ejects the processing liquid LQ5 toward the rotating substrate W. The nozzle 295 is connected to the pipe G5. The pipe G5 supplies the processing liquid LQ5 to the nozzle 295. The fluid supply unit V5 regulates the supply of the processing liquid LQ5 to the nozzle 295. The temperature sensor TS5 detects the temperature of the processing liquid LQ5 flowing in the pipe G5, and outputs a detection signal indicating the temperature. The temperature sensor TS5 includes, for example, a thermocouple.

[0245] When the fluid supply unit 29 is located at the approach position, the nozzle 297 ejects the inert gas GA toward the rotating substrate W. The nozzle 297 is connected to the pipe G6. The pipe G6 supplies the inert gas GA to the nozzle 297. The fluid supply unit V6 adjusts the supply of the inert gas GA to the nozzle 297. The temperature sensor TS6 detects the temperature of the inert gas GA flowing in the pipe G6 and outputs a detection signal indicating the temperature. The temperature sensor TS6 includes, for example, a thermocouple.

[0246] Each of the plurality of guard plates 33 has a substantially cylindrical shape. Each of the plurality of guard plates 33 receives the processing fluid (processing liquid LQ1, a mixed liquid of processing liquid LQ21 and processing liquid LQ22, processing liquid LQ3, or processing liquid LQ5) discharged from the substrate W. Each of the plurality of guard plates 33 can rise or fall.

[0247] FIG4(a) is a diagram showing the fluid supply unit V1. The structures of the fluid supply units V3 to V6 are the same as those of the fluid supply unit V1. Therefore, in the description of the fluid supply units V3 to V6, reference will be made to FIG4(a) as appropriate, and reference symbols for the parts of the fluid supply unit V1 will be used as appropriate.

[0248] As shown in FIG4(a), the fluid supply unit V1 includes a flow meter FW1, a flow control valve V11, and a supply valve V12. The flow meter FW1, the flow control valve V11, and the supply valve V12 are inserted in the pipe G1. The flow meter FW1 detects the flow of the treatment fluid ( Figure 3 In the example of the present invention, the flow rate of the treatment fluid LQ1 is outputted as a detection signal indicating the flow rate. The flow regulating valve V11 regulates the flow rate of the treatment fluid flowing in the piping G1. The flow regulating valve V11 is, for example, a motor needle valve. The supply valve V12 opens or blocks the piping G1, and switches the supply start and stop of the treatment fluid to the nozzle 17. The supply valve V12 is, for example, a relief valve.

[0249] FIG4(b) is a diagram showing the fluid supply section V2. As shown in FIG4(b), the fluid supply section V2 includes a valve V21, a valve V22, and a mixing section MXN. Valve V21 switches the start and stop of supply of the treatment liquid LQ21 to the mixing section MXN. Valve V22 switches the start and stop of supply of the treatment liquid LQ22 to the mixing section MXN. The mixing section MXN mixes the treatment liquid LQ21 with the treatment liquid LQ22, and supplies the mixed liquid to the piping G2 and the nozzle 21. Specifically, the mixing section MXN includes a mixing valve V23 and a flow meter FW2. The mixing valve V23 mixes the treatment liquid LQ21 with the treatment liquid LQ22. The flow meter FW2 detects the flow rate of the mixed liquid and outputs a detection signal. The treatment liquid LQ21 is, for example, sulfuric acid, and the treatment liquid LQ22 is, for example, hydrogen peroxide solution.

[0250] Below, sometimes Figure 3 to Figure 4(b) The nozzle 17, nozzle 21, nozzle 25, nozzle 27, nozzle 295, and nozzle 297 shown are collectively referred to as "nozzle NZ". The valve V11, valve V12, valve V21, valve V22, and valve V23 are sometimes collectively referred to as "valve VB". The arm 191 and arm 231 are sometimes collectively referred to as "arm AM".

[0251] Figure 5 2 is a diagram showing a substrate processing apparatus 200 . Figure 5 In order to simplify the drawings, the treatment fluid ( Figure 3In the example of the processing liquid LQ1, the components related to the processing fluid supplied to the piping G2 to the piping G6 are described. In addition, the structure of the components related to the processing fluid supplied to the piping G2 to the piping G6 is the same as the structure of the components related to the processing fluid supplied to the piping G1. Therefore, in the description of the components related to the processing fluid supplied to the piping G2 to the piping G6, appropriate reference is made to Figure 5 , reference symbols attached to parts related to the process fluid supplied to the pipe G1 are used appropriately.

[0252] like Figure 5 As shown, the substrate processing apparatus 200 includes a pipe G1 and a fluid supply unit V1 for each processing device 1 in each tower TW. The fluid supply unit V1 is accommodated in a processing fluid box 4 corresponding to the tower TW. A portion of each pipe G1 is accommodated in the chamber 11, and a portion of each pipe G1 is accommodated in the processing fluid box 4.

[0253] Furthermore, the substrate processing apparatus 200 includes a processing fluid tank 50, a circulation piping 51, a pump 55, a pulse damper 56, a filter 57, and a temperature regulator 58. The processing fluid tank 50, the pump 55, the pulse damper 56, the filter 57, and the temperature regulator 58 are accommodated in the processing fluid cabinet 5. A part of the circulation piping 51 is accommodated in the processing fluid cabinet 5, and a part of the circulation piping 51 is accommodated in the processing fluid box 4.

[0254] The treatment fluid tank 50 stores the treatment fluid ( Figure 3 The circulation pipe 51 includes an upstream pipe 52 extending downstream from the treatment fluid tank 50 , a plurality of individual pipes 53 branching from the upstream pipe 52 , and a downstream pipe 54 extending downstream from each individual pipe 53 to the treatment fluid tank 50 .

[0255] The upstream end of the upstream pipe 52 is connected to the treatment fluid tank 50. The downstream end of the downstream pipe 54 is connected to the treatment fluid tank 50. The upstream end of the upstream pipe 52 corresponds to the upstream end of the circulation pipe 51, and the downstream end of the downstream pipe 54 corresponds to the downstream end of the circulation pipe 51. Each individual pipe 53 extends from the downstream end of the upstream pipe 52 to the upstream end of the downstream pipe 54.

[0256] The plurality of individual pipes 53 correspond to the plurality of towers TW, respectively. Three pipes G1 corresponding to the three processing devices 1 included in one tower TW are connected to one individual pipe 53 .

[0257] The pump 55 delivers the treatment fluid in the treatment fluid tank 50 to the circulation pipe 51. The pulse damper 56 suppresses the pulsation of the treatment fluid delivered from the pump 55. The filter 57 removes foreign matter from the treatment fluid flowing in the circulation pipe 51. The temperature regulator 58 adjusts the temperature of the treatment fluid in the treatment fluid tank 50. The temperature regulator 58 is, for example, a heater that heats the treatment fluid.

[0258] The pump 55, the pulse damper 56, the filter 57 and the temperature regulator 58 are arranged in the upstream piping 52. The treatment fluid in the treatment fluid tank 50 is sent to the upstream piping 52 by the pump 55, and flows from the upstream piping 52 to a plurality of individual pipings 53. The treatment fluid in the individual piping 53 flows to the downstream piping 54, and returns to the treatment fluid tank 50 from the downstream piping 54. The treatment fluid is heated by the temperature regulator 58 and maintained at a predetermined temperature. Therefore, the temperature of the treatment fluid circulating in the circulation piping 51 is maintained at a predetermined temperature. In addition, the treatment fluid maintained at a predetermined temperature in the circulation piping 51 is supplied to the piping G1.

[0259] Furthermore, the substrate processing device 200 further includes a plurality of exhaust pipes P7, a plurality of temperature sensors TS7, and a plurality of flow meters FW3. The plurality of exhaust pipes P7 are respectively configured corresponding to the plurality of processing devices 1. The exhaust pipe P7 is connected to the processing device 1 to transfer the processing fluid ( Figure 3 In the example of the embodiment, the treatment fluid LQ1 is discharged to the outside of the treatment device 1. A flow meter FW3 is inserted in the discharge pipe P7. In addition, the flow meter FW3 detects the flow rate of the treatment fluid flowing in the discharge pipe P7 and outputs a detection signal indicating the flow rate. Moreover, the temperature sensor TS7 detects the temperature of the treatment fluid flowing in the discharge pipe P7 and outputs a detection signal indicating the temperature. The temperature sensor TS7 includes, for example, a thermocouple.

[0260] Next, refer to Figure 6 The control device 3 of the substrate processing apparatus 200 will be described. Figure 6 FIG. 3 is a block diagram showing the control device 3. The control device 3 is, for example, a computer. Figure 6 As shown, the control device 3 includes a control unit 3A, a storage unit 3B, a communication unit 3C, an input unit 3D, and a display unit 3E.

[0261] The control unit 3A includes a processor such as a central processing unit (CPU) and a graphics processing unit (GPU). The storage unit 3B includes a storage device that stores data and computer programs. The processor of the control unit 3A executes the computer program stored in the storage device of the storage unit 3B to control the various components of the substrate processing device 200.

[0262] For example, the storage unit 3B includes a main storage device such as a semiconductor memory and an auxiliary storage device such as a semiconductor memory and a hard disk drive. The storage unit 3B may also include a removable media such as an optical disk. The storage unit 3B is, for example, a non-transitory computer-readable storage medium.

[0263] Specifically, the storage unit 3B stores a control program PG1 and a plurality of processing condition information CD. The control unit 3A executes the control program PG1 to control each component of the substrate processing apparatus 200 based on at least one processing condition information CD. Each of the plurality of processing condition information CD is information indicating a processing condition of the substrate W. Each of the plurality of processing condition information CD includes a recipe information RP and a plurality of parameter information PM. The recipe information RP specifies the processing content and processing sequence of the substrate W. Each of the plurality of parameter information PM is a setting value for realizing processing according to the recipe information RP, and indicates a setting value for a component of the substrate processing apparatus 200. The recipe information RP and the parameter information PM are software components of the substrate processing apparatus 200.

[0264] The communication unit 3C is connected to a network and communicates with an external device. In this embodiment, the network includes, for example, the Internet, a local area network (LAN), a public telephone network, and a short-distance wireless network. The communication unit 3C is a communication device, for example, a network interface controller.

[0265] The input unit 3D is an input device for inputting various information to the control unit 3A. For example, the input unit 3D is a keyboard, a pointing device, or a touch panel.

[0266] The display unit 3E displays images and is, for example, a liquid crystal display or an organic electroluminescent display.

[0267] The control unit 3A obtains the timing data TD1 from the detection meter group SNS, and stores the timing data TD1 in the storage unit 3B. At this time, the control unit 3A stores the timing data TD1 in the storage unit 3B in association with the batch identification information, the substrate identification information, the processing sequence information (hereinafter referred to as "processing sequence information XA"), and the batch interval information (hereinafter referred to as "batch interval information XB"). The batch identification information is information for identifying a batch (for example, a batch number). A batch represents a processing unit of a substrate W. A batch is composed of a predetermined number of substrates W. The substrate identification information is information for identifying a substrate W. The processing sequence information XA is information indicating the processing sequence of a predetermined number of substrates W constituting a batch. The batch interval information XB is information indicating the time interval from the end of processing a batch to the start of processing the next batch.

[0268] The timing data TD1 indicates the physical quantity of the object used by the substrate processing apparatus 200. In addition, the detection meter group SNS is included in the substrate processing apparatus 200. Whenever the detection meter group SNS processes a substrate W, it detects the physical quantity of the object used by the substrate processing apparatus 200 during the period from the start of processing of the substrate W to the end of processing, and outputs the detection signal indicating the physical quantity to the control unit 3A. In addition, whenever the control unit 3A processes a substrate W, the physical quantity indicated by the detection signal output by the detection meter group SNS during the period from the start of processing of the substrate W to the end of processing is stored in the storage unit 3B as the timing data TD1 in association with time.

[0269] Each of the plurality of time series data TD1 includes at least one individual time series data TD. In this embodiment, each of the plurality of time series data TD1 includes one or more individual time series data TD. One individual time series data TD is time series data representing a physical quantity of one object.

[0270] In the present embodiment, the "object used by the substrate processing apparatus 200" is a "processing fluid" and / or a "part of the substrate processing apparatus 200". The "part" may include one component, or may include a plurality of components. In the case where the object used by the substrate processing apparatus 200 is a processing fluid, for example, the "physical quantity of the object used by the substrate processing apparatus 200" is "the flow rate of the processing fluid", "the temperature of the processing fluid" or "the concentration of the processing fluid". In the case where the object used by the substrate processing apparatus 200 is a part of the substrate processing apparatus 200, for example, the "physical quantity of the object used by the substrate processing apparatus 200" is "a physical quantity representing the action and / or state of the part".

[0271] Specifically, the detection meter group SNS includes a flow meter group 3F, a temperature sensor group 3G, and a sensor group 3H.

[0272] The flow meter group 3F includes a plurality of flow meters FW. Each of the plurality of flow meters FW detects the flow rate of the processing fluid during the period from the start of processing of the substrate W to the end of processing, and outputs a detection signal indicating the flow rate, each time a substrate W is processed. In addition, the control unit 3A causes the flow rate indicated by the detection signal output by each flow meter FW during the period from the start of processing of the substrate W to the end of processing to be stored in the storage unit 3B as time series data TD1 in association with time each time a substrate W is processed. At this time, one individual time series data TD is time series data indicating the flow rate of the processing fluid supplied to one nozzle NZ when the processing fluid is ejected.

[0273] The plurality of flow meters FW include a plurality of flow meters FW1, a plurality of flow meters FW2, and a plurality of flow meters FW3 ( Figure 3 4(a) and 4(b)). Furthermore, the plurality of flow meters FW may also include one or more other flow meters. In this case, the flow meter may be disposed at any position (eg, piping) of the substrate processing apparatus 200.

[0274] The temperature sensor group 3G includes a plurality of temperature sensors TS. Each of the plurality of temperature sensors TS detects the temperature of the processing fluid during the period from the start of processing of the substrate W to the end of processing, and outputs a detection signal indicating the temperature, each time a substrate W is processed. In addition, the control unit 3A causes the storage unit 3B to store the temperature indicated by the detection signal output from each temperature sensor TS during the period from the start of processing of the substrate W to the end of processing, in association with time, as time series data TD1. At this time, one individual time series data TD is time series data indicating the temperature of the processing fluid supplied to one nozzle NZ when the processing fluid is ejected.

[0275] The plurality of temperature sensors TS include a plurality of temperature sensors TS1, a plurality of temperature sensors TS2, a plurality of temperature sensors TS3, a plurality of temperature sensors TS4, a plurality of temperature sensors TS5, a plurality of temperature sensors TS6, and a plurality of temperature sensors TS7 ( Figure 3 ). In addition, the plurality of temperature sensors TS may also include one or more other temperature sensors. In this case, the temperature sensor may be disposed at any position of the substrate processing apparatus 200 (eg, a pipe, a processing fluid tank 50 or a chamber 11).

[0276] The sensor group 3H includes a plurality of sensors SN. Each of the plurality of sensors SN detects a physical quantity indicating the movement and / or state of a part used in the substrate processing apparatus 200 during a period from the start of the processing of the substrate W to the end of the processing, and outputs a detection signal indicating the physical quantity indicating the movement and / or state of the part. In addition, the control unit 3A causes the physical quantity indicating the movement and / or state of the part, which is indicated by the detection signal output from each sensor SN during a period from the start of the processing of the substrate W to the end of the processing, to be stored in the storage unit 3B as time series data TD1 in association with time, each time a substrate W is processed. At this time, one individual time series data TD is time series data indicating the physical quantity of the movement and / or state of one part.

[0277] The parts used in the substrate processing device 200 include, for example, a rotating chuck 13 for rotating the substrate W, an arm AM for moving a nozzle NZ for spraying a processing fluid onto the substrate W, a blocking plate 291 for covering the upper surface of the substrate W with a gap, a protective plate 33 for receiving the processing fluid scattered from the substrate W, or a valve VB for regulating the flow of the processing fluid.

[0278] In the case where the component used in the substrate processing apparatus 200 is the rotating chuck 13, the physical quantity of the rotating chuck 13 represents a quantity related to the movement and / or state of the rotating chuck 13. For example, the physical quantity represents the rotational speed of the rotating chuck 13. Therefore, the sensor SN detects the rotational speed of the rotating chuck 13 and outputs a detection signal representing the rotational speed. At this time, the sensor SN detects the rotational speed of the rotating chuck 13 optically, electrically or mechanically. The rotational speed of the rotating chuck 13 represents the rotational speed of the substrate W. In addition, the sensor SN can also detect the rotational speed of the rotating motor 15. At this time, the component used in the substrate processing apparatus 200 is the rotating motor 15, and the physical quantity of the rotating motor 15 is the rotational speed of the rotating motor 15.

[0279] In the case where the part used in the substrate processing device 200 is the arm AM, the physical quantity of the arm AM represents a quantity related to the movement and / or state of the arm AM. For example, the physical quantity is information representing the position or displacement of the arm AM when performing the scanning process of the nozzle NZ. The so-called scanning process is that the nozzle NZ moves along the radial direction of the substrate W at the ejection position of the processing fluid while processing the substrate W with the processing fluid. At this time, the sensor SN detects the position or displacement of the arm AM and outputs a detection signal representing the position or displacement. At this time, the sensor SN detects the position or displacement of the arm AM optically, electrically or mechanically.

[0280] When the part used in the substrate processing device 200 is a blocking plate 291, the physical quantity of the blocking plate 291 represents a quantity related to the action and / or state of the blocking plate 291. For example, the physical quantity is information indicating the position of the blocking plate 291 along the vertical direction, information indicating the displacement, or information indicating the rotation speed. At this time, the sensor SN detects the position, displacement, or rotation speed of the blocking plate 291 and outputs a detection signal indicating the position, displacement, or rotation speed. At this time, the sensor SN detects the position, displacement, or rotation speed of the blocking plate 291 optically, electrically, or mechanically.

[0281] When the part used in the substrate processing apparatus 200 is the guard plate 33, the physical quantity of the guard plate 33 indicates a quantity related to the action and / or state of the guard plate 33. For example, the physical quantity is information indicating the position or displacement of the guard plate 33. At this time, the sensor SN detects the position or displacement of the guard plate 33 and outputs a detection signal indicating the position or displacement. At this time, the sensor SN detects the position or displacement of the guard plate 33 optically, electrically, or mechanically.

[0282] When the component used in the substrate processing apparatus 200 is a valve VB, the physical quantity of the valve VB indicates a quantity related to the action and / or state of the valve VB. For example, the physical quantity is information indicating the opening of the valve VB. At this time, the sensor SN detects the opening of the valve VB and outputs a detection signal indicating the opening. At this time, the sensor SN detects the opening of the valve VB optically, electrically, or mechanically. The opening indicates the degree to which the valve VB is opened.

[0283] The plurality of sensors SN may also include a plurality of concentration sensors. In this case, each time a substrate W is processed, the plurality of concentration sensors directly or indirectly detect the concentration of the processing fluid during the period from the start of processing of the substrate W to the end of processing, and output a detection signal indicating the concentration. In addition, the control unit 3A causes the concentration indicated by the detection signal output from each concentration sensor during the period from the start of processing of the substrate W to the end of processing to be stored in the storage unit 3B as timing data TD1 in association with time each time a substrate W is processed. In this case, an individual timing data TD is timing data indicating the concentration of the processing fluid supplied to a nozzle NZ when the processing fluid is ejected. In addition, the concentration may also be indicated by the specific gravity of the processing fluid. In this case, the concentration sensor detects the specific gravity of the processing fluid. Moreover, the concentration sensor may be arranged in the piping, or may also be arranged in the processing fluid tank 50.

[0284] The control unit 3A controls the communication unit 3C to transmit the time series data TD1 to the learning data creation device 400, for example. As a result, the communication unit 3C transmits the time series data TD1 to the learning data creation device 400 via the network.

[0285] The control unit 3A controls the display unit 3E so as to display the time series data TD1, for example. As a result, the display unit 3E displays the time series data TD1.

[0286] Next, refer to Figure 2 , Figure 3 , Figure 6 and Figure 7 , a substrate processing method performed by the substrate processing apparatus 200 is described. Figure 7 FIG. 2 is a flow chart showing a substrate processing method according to the present embodiment. Figure 7 As shown, the substrate processing method includes steps S1 to S10. The substrate processing method is performed on each substrate W by the substrate processing apparatus 200.

[0287] like Figure 7 As shown, in step S1, Figure 2 The central robot CR of the substrate processing apparatus 200 shown in FIG. 1 carries the substrate W into the processing apparatus 1 . Figure 3 In the processing apparatus 1 shown, the substrate W is held by the chuck member 130 of the spin chuck 13 .

[0288] Next, in step S2, Figure 6 The control unit 3A of the control device 3 shown starts acquiring and storing the time series data TD1 from the detection meter group SNS.

[0289] Next, in step S3 , the spin chuck 13 starts rotating the substrate W.

[0290] Next, in step S4 , the nozzle 17 of the processing apparatus 1 discharges the processing liquid LQ1 onto the surface of the substrate W, thereby processing the substrate W.

[0291] Next, in step S5 , the nozzle 25 of the processing apparatus 1 discharges the processing liquid LQ3 toward the surface of the substrate W, thereby rinsing the processing liquid LQ1 from the substrate W.

[0292] Next, in step S6 , the nozzle 295 of the processing apparatus 1 discharges the processing liquid LQ5 onto the surface of the substrate W to replace the processing liquid LQ3 with the processing liquid LQ5 , thereby drying the substrate W.

[0293] Next, in step S7 , the nozzle 297 of the processing apparatus 1 sprays the inert gas GA toward the surface of the substrate W to dry the substrate W.

[0294] Next, in step S8 , the spin chuck 13 of the processing apparatus 1 stops rotating the substrate W, and the chuck member 130 releases the substrate W.

[0295] Next, in step S9 , the control unit 3A of the control device 3 stops acquiring and storing the time series data TD1 from the detection meter group SNS.

[0296] Next, in step S10 , the central robot CR of the substrate processing apparatus 200 carries the substrate W out of the processing apparatus 1 .

[0297] As described above, the substrate processing apparatus 200 executes steps S1 to S10 , thereby completing the processing of one substrate W.

[0298] also, Figure 2 to Figure 7 In the description of the substrate processing apparatus 200, the "substrate W" may be changed to "learning target substrate W1". Figure 2 to Figure 7 In the description, the "substrate processing apparatus 200" may be renamed as the "substrate processing apparatus 300", the "timing data TD1" may be renamed as the "timing data TD2", and the "substrate W" may be renamed as the "processing target substrate".

[0299] Next, refer to Figures 8 to 18 The learning data preparation device 400 is described. Figure 8 The following describes the learning data creation device 400. The learning data creation device 400 is, for example, a computer. Figure 8 4 is a block diagram showing the learning data preparation device 400. Figure 8 As shown, the learning data creating device 400 includes a processing unit 4A, a storage unit 4B, a communication unit 4C, an input unit 4D, and a display unit 4E.

[0300] The processing unit 4A includes a processor such as a CPU and a GPU. The storage unit 4B includes a storage device that stores data and computer programs. The processor of the processing unit 4A executes the computer program stored in the storage device of the storage unit 4B to perform various processes. For example, the storage unit 4B includes a main storage device and an auxiliary storage device, similar to the storage unit 3B, and may also include a removable medium. The storage unit 4B is, for example, a non-temporary computer-readable storage medium.

[0301] Specifically, the storage unit 4B stores the learning data creation program PG2. The processing unit 4A executes the learning data creation program PG2 and functions as a learning data creation unit 61, a storage control unit 63, and a display control unit 65. That is, the processing unit 4A includes the learning data creation unit 61, the storage control unit 63, and the display control unit 65. The learning data creation unit 61 includes an acquisition unit 611 and a feature quantity calculation unit 613.

[0302] The communication unit 4C is connected to a network and communicates with an external device. The communication unit 4C is a communication device, such as a network interface controller.

[0303] The input unit 4D is an input device for inputting various information to the processing unit 4A. For example, the input unit 4D is a keyboard, a pointing device, or a touch screen.

[0304] The display unit 4E displays images and is, for example, a liquid crystal display or an organic electroluminescent display.

[0305] Next, refer to Figure 8 The processing unit 4A is described. The learning data generating unit 61 of the processing unit 4A generates learning data TND based on the time series data TD1. The learning data TND is the object of machine learning. In addition, the storage control unit 63 controls the storage unit 4B in a manner to store the learning data TND. As a result, the storage unit 4B stores the learning data TND in association with the batch identification information and the substrate identification information. For example, one learning data TND is generated for at least one time series data TD1 obtained when processing a learning object substrate W1.

[0306] The learning data TND includes a feature quantity XD related to the processing of the learning object substrate W1, and abnormal main cause information YD related to the processing of the learning object substrate W1. The feature quantity XD includes first feature quantity information XD1. The feature quantity XD may also include second feature quantity information XD2 and / or third feature quantity information XD3. In this embodiment, the feature quantity XD includes first feature quantity information XD1, second feature quantity information XD2, and third feature quantity information XD3.

[0307] Specifically, the acquisition unit 611 of the learning data preparation unit 61 acquires a plurality of time series data TD1 from the substrate processing apparatus 200. At this time, the batch identification information, the substrate identification information, the processing order information XA, and the batch interval information XB are attached to each time series data TD1. For example, the acquisition unit 611 acquires a plurality of time series data TD1 from the substrate processing apparatus 200 via the network and the communication unit 4C.

[0308] In addition, the storage unit 4B stores each time series data TD1 in association with the lot identification information, the substrate identification information, the processing sequence information XA, and the lot interval information XB.

[0309] Furthermore, the acquisition unit 611 acquires at least one interval data from the time series data TD1. In this embodiment, the acquisition unit 611 acquires a plurality of interval data from the time series data TD1. The interval data is data of a time interval including a period during which a characteristic portion of the time series data appears, and is a part of the time series data TD1.

[0310] In addition, the feature amount calculation unit 613 calculates the first feature amount information XD1 based on the interval data for each interval data. The first feature amount information XD1 represents the time-lapse feature of the interval data. The first feature amount information XD1 is represented by time. The first feature amount information XD1 is equivalent to an example of "feature amount information". The details of the first feature amount information XD1 will be described below.

[0311] Then, the input unit 4D receives a plurality of abnormality cause information YD from the user.

[0312] In addition, the storage control unit 63 marks the first feature quantity information XD1 with the abnormal main cause information YD. Specifically, the storage control unit 63 controls the storage unit 4B in such a manner that the first feature quantity information XD1 is stored in association with the abnormal main cause information YD. As a result, the storage unit 4B stores the first feature quantity information XD1 in association with the abnormal main cause information YD. The abnormal main cause information YD indicates the main cause of the abnormality of the learning object substrate W1 after being processed by the processing fluid.

[0313] Furthermore, the storage control unit 63 marks the abnormal main cause information YD for the second feature quantity information XD2. Specifically, the storage control unit 63 controls the storage unit 4B in such a manner that the abnormal main cause information YD is associated with the second feature quantity information XD2 and stored. As a result, the storage unit 4B stores the abnormal main cause information YD in association with the second feature quantity information XD2. The second feature quantity information XD2 is the processing sequence information XA indicating the processing sequence of a predetermined number of learning object substrates W1 constituting a batch, or the batch interval information XB indicating the time interval from the end of the processing of a batch to the start of the processing of the next batch.

[0314] Furthermore, the storage control unit 63 marks the abnormal main cause information YD on the third characteristic quantity information XD3. Specifically, the storage control unit 63 controls the storage unit 4B in such a manner that the abnormal main cause information YD is associated with the third characteristic quantity information XD3 and stored. As a result, the storage unit 4B stores the abnormal main cause information YD in association with the third characteristic quantity information XD3. The third characteristic quantity information XD3 is interval data, object information (hereinafter referred to as "object information IDA") or physical quantity information (hereinafter referred to as "physical quantity information IDB"). The interval data is the interval data acquired by the acquisition unit 611 from the time series data TD1.

[0315] The object information IDA is information for identifying an object used in the substrate processing apparatus 200. When the object is a "processing fluid", the object information IDA is information for identifying the "processing fluid". When the object is a "substrate processing apparatus component", the object information IDA is information for identifying the "component".

[0316] The physical quantity information IDB is information for identifying the category of the physical quantity of the object used by the substrate processing apparatus 200. When the object is a "processing fluid", the physical quantity information IDB is information indicating that the category of the physical quantity is "the flow rate of the processing fluid", information indicating that the category of the physical quantity is "the temperature of the processing fluid", or information indicating that the category of the physical quantity is "the concentration of the processing fluid". When the object is a "part", the physical quantity information IDB is information indicating that the category of the physical quantity is "a physical quantity related to the movement and / or state of the part". For example, when the object is a "part", the physical quantity information IDB is information indicating that the category of the physical quantity is "the rotation speed of the part", "the position of the part", "the displacement of the part", or "the opening degree of the part".

[0317] Above, as reference Figure 8 As described above, the storage unit 4B stores the abnormality cause information YD in association with the first feature amount information XD1 , the second feature amount information XD2 , and the third feature amount information XD3 .

[0318] That is, the storage control unit 63 marks the abnormal main cause information YD for the feature value XD. Specifically, the storage control unit 63 controls the storage unit 4B in a manner of storing the feature value XD in association with the abnormal main cause information YD. As a result, the storage unit 4B stores the feature value XD and the abnormal main cause information YD as the learning data TND. That is, the feature value XD and the abnormal main cause information YD constitute the learning data TND. For example, one learning data TND includes the feature value XD calculated based on at least one time series data TD1 obtained when processing a learning object substrate W1, and the abnormal main cause information YD associated with the feature value XD.

[0319] The abnormality main cause information YD includes at least one of the first abnormality main cause information YD1 and the second abnormality main cause information YD2. In the present embodiment, the abnormality main cause information YD includes the first abnormality main cause information YD1 and the second abnormality main cause information YD2.

[0320] The first abnormal main cause information YD1 is information related to the processing fluid when the substrate processing device 200 processes the learning object substrate W1, and is information indicating the main cause of the abnormality of the learning object substrate W1 after being processed by the processing fluid. The second abnormal main cause information YD2 is information related to the parts of the substrate processing device 200, and is information indicating the main cause of the abnormality of the learning object substrate W1 after being processed by the processing fluid. The input unit 4D receives a plurality of first abnormal main cause information YD1 and a plurality of second abnormal main cause information YD2 from the user. In addition, the storage control unit 63 marks the first abnormal main cause information YD1 and the second abnormal main cause information YD2 on the first feature quantity information XD1 to the third feature quantity information XD3. Specifically, the storage control unit 63 controls the storage unit 4B in a manner that the first abnormal main cause information YD1 and the second abnormal main cause information YD2 are associated with the first feature quantity information XD1 to the third feature quantity information XD3 and stored. As a result, the storage unit 4B stores the first abnormality cause information YD1 and the second abnormality cause information YD2 in association with the first feature amount information XD1 to the third feature amount information XD3.

[0321] Furthermore, the display control unit 65 controls the display unit 4E so as to display the time series data TD1 or the learning data TND. As a result, the display unit 4E displays the time series data TD1 or the learning data TND.

[0322] Next, refer to Figure 8 and Fig. 9 The time series data TD1 is described. In this description, an example is given in which the physical quantity of the object used in the substrate processing apparatus 200 is the "flow rate of the processing fluid". The "processing fluid" includes the processing liquid LQ1, the processing liquid LQ3, the processing liquid LQ5, and the inert gas GA.

[0323] Fig. 9 It is a graph showing an example of time series data TD1. Fig. 9 In the figure, the horizontal axis represents time and the vertical axis represents the flow rate of the processed fluid.

[0324] like Fig. 9As shown, the timing data TD1 includes individual timing data TD11, individual timing data TD12, individual timing data TD13, and individual timing data TD14 as individual timing data TD. The individual timing data TD11 is the flow rate of the processing liquid LQ1 supplied to the nozzle 17 when the processing liquid LQ1 is ejected, and represents the flow rate detected by the flow meter FW1 of the fluid supply part V1. The individual timing data TD12 is the flow rate of the processing liquid LQ3 supplied to the nozzle 25 when the processing liquid LQ3 is ejected, and represents the flow rate detected by the flow meter FW1 of the fluid supply part V3. The individual timing data TD13 is the flow rate of the processing liquid LQ5 supplied to the nozzle 295 when the processing liquid LQ5 is ejected, and represents the flow rate detected by the flow meter FW1 of the fluid supply part V5. The individual time series data TD14 is the flow rate of the inert gas GA supplied to the nozzle 297 when the inert gas GA is ejected, and indicates the flow rate detected by the flow meter FW1 of the fluid supply unit V6 .

[0325] like Figure 8 and Fig. 9 As shown, the acquisition unit 611 acquires the interval data of each time interval T11 to time interval T14, the interval data of each time interval T21 to time interval T24, the interval data of each time interval T31 to time interval T34, the interval data of each time interval T41 to time interval T44, and the interval data of each time interval T51 and time interval T61 from the time series data TD1.

[0326] Hereinafter, “section data” may be collectively referred to as “section data SX”.

[0327] The time interval T11 to the time interval T14, the time interval T21 to the time interval T24, the time interval T31 to the time interval T34, the time interval T41 to the time interval T44, the time interval T51, and the time interval T61 are intervals including a characteristic portion of the time transition of the time series data TD1. For example, the time interval T11 to the time interval T14, the time interval T21 to the time interval T24, the time interval T31 to the time interval T34, the time interval T41 to the time interval T44, the time interval T51, and the time interval T61 are the same in the plurality of time series data TD1 for each of the plurality of learning target substrates W1. For example, the time interval T11 to the time interval T14, the time interval T21 to the time interval T24, the time interval T31 to the time interval T34, the time interval T41 to the time interval T44, the time interval T51, and the time interval T61 are the same in all the time series data TD1 for all the learning target substrates W1.

[0328] Time interval T11 to time interval T14 represent the period during which the first characteristic portion of the time transition of the timing data TD1 (individual timing data TD) may appear. The first characteristic portion is the on delay portion. The on delay portion is the portion of the timing data TD1 (individual timing data TD) that represents the flow rate from the start of the ejection of the treatment fluid and the increase in flow rate until the target value is reached. The on delay portion represents the delay until the flow rate of the treatment fluid reaches the target value when the flow rate of the treatment fluid increases toward the target value. In addition, Fig. 9 In the figure, the "start of ejection control" of the processing fluid is recorded as "processing liquid LQ1 is turned on", "processing liquid LQ3 is turned on", "processing liquid LQ5 is turned on" and "inert gas GA is turned on". "Start of ejection control" means when the control unit 3A outputs an opening instruction signal to the supply valve V12 (Figure 4 (a)). The opening instruction signal is a signal (valve connection signal) that instructs the supply valve V12 to open. In the following, the start of ejection control is sometimes recorded as "ejection control start time tx".

[0329] The time interval T21 to the time interval T24 represent the period during which the second characteristic part of the time transition of the time series data TD1 (individual time series data TD) may appear. The second characteristic part is the overshoot part. The overshoot part is the part in the time series data TD1 (individual time series data TD) where the flow rate increases from the start of the ejection of the treatment fluid and exceeds the target value.

[0330] Time intervals T31 to T34 represent periods during which the third characteristic portion of the temporal transition of the time series data TD1 (individual time series data TD) may appear. The third characteristic portion is a variable portion. The variable portion is a portion in the time series data TD1 (individual time series data TD) where the flow rate fluctuates around the target value.

[0331] Time interval T41 to time interval T44 represent the period during which the fourth characteristic portion of the time transition of the timing data TD1 (individual timing data TD) may appear. The fourth characteristic portion is the off delay portion. The off delay portion is the portion of the timing data TD1 (individual timing data TD) that represents the flow rate from when the ejection of the treatment fluid stops and the flow rate decreases until it reaches the lower limit target value (e.g., zero). The off delay portion represents the delay until the flow rate of the treatment fluid reaches the lower limit target value when the flow rate of the treatment fluid decreases toward the lower limit target value (e.g., zero). In addition, Fig. 9In the figure, "when the ejection control is stopped" of the processing fluid is recorded as "the processing liquid LQ1 is turned off", "the processing liquid LQ3 is turned off", "the processing liquid LQ5 is turned off" and "the inert gas GA is turned off". "When the ejection control is stopped" means when the control unit 3A outputs a clogging indication signal to the supply valve V12 (Figure 4 (a)). The clogging indication signal is a signal (valve closing signal) indicating the closing of the supply valve V12. In the following, the ejection control is sometimes recorded as "when the ejection control is stopped ty".

[0332] The time interval T51 represents the period during which the fifth characteristic part of the time transition of the time series data TD1 (individual time series data TD11, individual time series data TD12) may appear. The fifth characteristic part is the overlapping part. The overlapping part is the temporal repetition part of the adjacent individual time series data TD11 and the individual time series data TD12 in the time series data TD1. That is, the overlapping part is the temporal repetition part of the flow rates of two different treatment fluids in the time series data TD1.

[0333] The time interval T61 represents the period during which the sixth characteristic part of the time transition of the time series data TD1 (individual time series data TD12, individual time series data TD13) may appear. The sixth characteristic part is the intermittent part. The intermittent part is the part in the time series data TD1 where the interval between the adjacent individual time series data TD12 and the individual time series data TD13 is spaced apart in time. That is, the intermittent part is the part in the time series data TD1 where the interval between two different treatment fluids is spaced apart in time.

[0334] Next, the on-delay part, overshoot part, fluctuation part, off-delay part, overlap part and intermittent part are generally defined. In the following definition, "object" means "object used in the substrate processing apparatus 200", and "physical quantity" means "physical quantity of the object".

[0335] That is, the on-delay portion is the portion of the time series data TD1 (individual time series data TD) that represents the physical quantity of the object from the start of use of the object or the change of the use state until the physical quantity of the object increases to the target value. The on-delay portion represents the delay until the physical quantity of the object reaches the target value when the physical quantity of the object increases toward the target value.

[0336] The overshoot amount portion is a portion in the time series data TD1 (individual time series data TD) in which the physical quantity of the object increases after the use of the object is started or the use state is changed, and the overshoot amount portion exceeds the target value.

[0337] The fluctuating portion is a portion in which the physical quantity of the object fluctuates up and down around the target value in the time series data TD1 (individual time series data TD).

[0338] The shutdown delay portion is a portion of the timing data TD1 (individual timing data TD) that indicates the physical quantity of the object from the time when the use of the object stops or the use state changes and the physical quantity of the object decreases until it reaches the lower limit target value. The shutdown delay portion indicates the delay until the physical quantity of the object reaches the lower limit target value when the physical quantity of the object decreases toward the lower limit target value.

[0339] The overlapping portion is a portion in which the physical quantities of two different objects overlap in time in the time series data TD1.

[0340] The intermittent portion is a portion in which the interval between two different objects in the time series data TD1 is spaced apart in terms of time.

[0341] Next, refer to Fig.10 The feature amount XD will be described. Fig.10 is a graph showing the characteristic quantity XD. Fig.10 As shown, the first feature quantity information XD1 of the feature quantity XD includes at least one of the first information X1, the second information X2, the third information X3, the fourth information X4, the fifth information X5 and the sixth information X6. Fig.10 In the example, the first feature amount information XD1 includes first information X1, second information X2, third information X3, fourth information X4, fifth information X5 and sixth information X6.

[0342] The first information X1 indicates the time-lapse characteristics of the on-delay portion of the time series data TD1. That is, the first information X1 is information indicating the flow rate state when the flow rate of the treatment fluid increases toward the target value.

[0343] The first information X1 includes at least one of information a, information b, information c, and information d. Fig.10 In the example of , the first information X1 includes information a to information d. Information a to information c are represented by time, and information d is represented by angle. Details of information a to information d will be described below.

[0344] The second information X2 indicates the characteristics of the temporal transition of the overshoot portion of the time series data TD1. That is, the second information X2 is information indicating the overshoot of the flow rate of the process fluid.

[0345] The second information X2 includes at least one of the information e and the information f. Fig.10 In the example of , the second information X2 includes information e and information f. Information e is represented by time, and information f is represented by flow. Details of information e and information f will be described below.

[0346] The third information X3 indicates the characteristics of the temporal transition of the variation portion of the time series data TD1. That is, the third information X3 is information indicating the variation (temporal variation) of the flow rate of the treatment fluid.

[0347] The third information X3 includes at least one of the information g, the information h and the information i. Fig.10 In the example of , the third information X3 includes information g to information i. Information g is represented by time, information h is represented by area, and information i is represented by period. Details of information g to information i will be described below.

[0348] The fourth information X4 indicates the time-lapse characteristics of the OFF delay portion of the time series data TD1. That is, the fourth information X4 is information indicating the flow rate state when the flow rate of the process fluid decreases from the target value.

[0349] The fourth information X4 includes at least one of information j, information k, information l and information m. Fig.10 In the example of , the fourth information X4 includes information j to information m. Information j to information l are represented by time, and information m is represented by angle. Details of information j to information m will be described below.

[0350] The fifth information X5 indicates the time-lapse characteristics of the overlapping portion of the time series data TD1. That is, the fifth information X5 is information indicating the overlap of the flow rate of one of the two processing fluids with the flow rate of the other processing fluid when the substrate processing apparatus 200 uses at least two different processing fluids.

[0351] The fifth information X5 includes at least one of information n, information o, and information p. Fig.10 In the example of , the fifth information X5 includes information n to information p. Information n is represented by time, information o is represented by area, and information p is represented by flow rate. Details of information n to information p will be described below.

[0352] The sixth information X6 indicates the time-lapse characteristics of the intermittent portion of the time series data TD1. That is, the sixth information X6 is information indicating the time interval between one of the two processing fluids and the other processing fluid on the time axis when the substrate processing apparatus 200 uses at least two different processing fluids.

[0353] The sixth information X6 includes information q. The information q is represented by time. The details of the information q will be described below.

[0354] above, Fig.10 In the description of , the first information X1 to the sixth information X6 are described by citing “processing fluid” as an example of “object” and “flow rate” as an example of “physical quantity”. However, the first information X1 to the sixth information X6 can be defined in a generalized manner as follows. That is, Fig.10In the description of the first information X1 to the sixth information X6, "processing fluid" may be replaced by "object" and "flow rate" may be replaced by "physical quantity". In addition, the lot identification information and the substrate identification information are associated with the feature quantity XD.

[0355] Furthermore, the second feature amount information XD2 includes at least one of the processing order information XA and the batch interval information XB. Fig.10 In the example of , the second feature amount information XD2 includes the processing order information XA and the batch interval information XB.

[0356] Furthermore, the third feature quantity information XD3 includes at least one of the section data SX, the object information IDA, and the physical quantity information IDB. Fig.10 In the example of , the third feature quantity information XD3 includes section data SX, object information IDA, and physical quantity information IDB.

[0357] Next, refer to Figure 11(a) to Figure 16 , the feature quantity calculation unit 613 ( Figure 8 ) Information a~Information q( Fig.10 ) is explained below. Figure 11(a) to Figure 16 In the figure, the horizontal axis represents time and the vertical axis represents the flow rate of the processed fluid.

[0358] Figure 11(a) to Figure 11(c) It is a diagram showing an example of a method of calculating the first information X1 included in the first feature amount information XD1.

[0359] As shown in FIG. 11( a ), the feature quantity calculation unit 613 analyzes the interval data SX including the on-delay portion and calculates the time t1. The time t1 is information a of the first information X1. The time t1 represents the period from the start time tx of the ejection control of the treatment fluid to the start time ta of the increase of the flow rate of the treatment fluid. The start time ta of the increase represents when the flow rate reaches U% of the target value TG. "U%" is, for example, a value below 10%.

[0360] As shown in FIG11(b), the feature quantity calculation unit 613 analyzes the interval data SX including the connection delay portion and calculates the time t2. The time t2 is the information b of the first information X1. The time t2 represents the period from the start of the rise of the flow rate of the treatment fluid ta to the rise of the flow rate of the treatment fluid tb. The rise time tb represents when the flow rate approaches the target value TG. Specifically, the rise time tb represents when the flow rate reaches V% of the target value TG. V% is greater than U%. "V%" represents, for example, a value greater than 70% and less than 100%. In addition, the rise time tb can also be recorded as "the rise is suspected to be completed tb".

[0361] Furthermore, the feature quantity calculation unit 613 analyzes the interval data SX including the on-delay portion, and calculates the inclination angle θ1 indicating the rising inclination of the flow rate. The inclination angle θ1 is information d of the first information X1. The inclination angle θ1 is, for example, the inclination angle of a straight line connecting the flow value at the start of the rise ta of the treatment fluid and the flow value at the rise tb, or the inclination angle of an approximate straight line indicating the flow value between the start of the rise ta and the rise tb of the treatment fluid.

[0362] As shown in Fig. 11(c), the feature quantity calculation unit 613 analyzes the interval data SX including the on-delay portion and calculates the time t3. The time t3 is the information c of the first information X1. The time t3 represents the period from the start time tx of the ejection control of the treatment fluid to the rise time tb of the flow rate of the treatment fluid.

[0363] Above, as reference Figure 11(a) to Figure 11(c) As described above, the time t1 to the time t3 and the inclination angle θ1 are indices indicating the delay until the flow rate of the processing fluid reaches the target value TG when the flow rate of the processing fluid increases toward the target value TG.

[0364] Fig.12 : is a diagram showing an example of a method for calculating the second information X2 included in the first feature amount information XD1. Fig.12 As shown, the characteristic quantity calculation unit 613 analyzes the interval data SX including the overshoot portion and calculates the time t4. The time t4 is the information e of the second information X2. The time t4 represents the period from the reference time tc to the time td when the flow rate changes due to the overshoot of the flow rate enters the first specified range RG1. The reference time tc is set to any time from the start time tx of the ejection control to the time when the flow rate reaches the maximum value MX. The first specified range RG1 includes the target value TG of the flow rate.

[0365] Fig.12 In the example, the reference time tc is when the flow rate reaches the maximum value MX. Moreover, the upper limit value of the first prescribed range RG1 is X1% of the target value TG of the flow rate, and the lower limit value of the first prescribed range RG1 is X2% of the target value TG of the flow rate. The specific values ​​of X1% and X2% are determined, for example, based on experiments and / or experience. When the change in flow rate enters the first prescribed range RG1, td is, for example, when the local minimum value MN of the flow rate enters the first prescribed range RG1. The so-called local minimum value MN is the minimum value of a local time interval. When the change in flow rate enters the first prescribed range RG1, td may also be, for example, when the local maximum value of the flow rate enters the first prescribed range RG1. The so-called local maximum value is the maximum value of a local time interval.

[0366] For example, time t4 indicates a period from when the flow rate exceeds the overshoot amount to when the flow rate continues to be within the first predetermined range RG1.

[0367] Then, the characteristic amount calculation unit 613 analyzes the section data SX including the overshoot amount portion and calculates the maximum value MX of the flow rate. The maximum value MX of the flow rate is the information f of the second information X2.

[0368] Above, as reference Fig.12 As described above, the time t4 and the maximum value MX are indicators indicating the degree of overshoot of the flow rate of the process fluid. The overshoot is a phenomenon that the flow rate exceeds the first predetermined range RG1 for the first time and exceeds the first predetermined range RG1.

[0369] Fig.13 : is a diagram showing an example of a method for calculating the third information X3 included in the first feature amount information XD1. Fig.13 As shown, the feature quantity calculation unit 613 analyzes the interval data SX including the change portion of the flow rate and calculates the time t5. The time t5 is the information g of the third information X3. The time t5 represents the period from the reference time te to the time tf when the flow rate changes. The stable state represents the state in which the change of the flow rate continues to enter the second prescribed range RG2 (within the range of Y1% to Y2%). The specific values ​​of Y1% and Y2% are determined, for example, based on experiments and / or experience. The feature quantity calculation unit 613 determines that the change of the flow rate is in a stable state, for example, when the local maximum value of the flow rate enters the second prescribed range RG2 continuously for a prescribed number of times, when the local minimum value of the flow rate enters the second prescribed range RG2 continuously for a prescribed number of times, or when the local maximum value and the local minimum value of the flow rate enter the second prescribed range RG2 continuously for a prescribed number of times.

[0370] The second predetermined range RG2 includes the target value TG. The second predetermined range RG2 is the same as the first predetermined range RG1 or is a narrower range than the first predetermined range RG1. The reference time te is set to any time from the start time tx of the discharge control of the process fluid to the time when the predetermined time has passed.

[0371] When an overshoot of the flow rate occurs, for example, the reference time te is when a certain time has passed since the overshoot converged or the flow rate reached the maximum value MX. When an overshoot of the flow rate does not occur, for example, the reference time te is when the ejection control starts tx, when the flow rate starts to rise ta, or when the flow rate rises tb.

[0372] Furthermore, the characteristic quantity calculation unit 613 analyzes the interval data SX including the flow rate variation portion and calculates the area A1 related to the flow rate. The area A1 related to the flow rate is the information h of the third information X3. The area A1 represents the total value of the area of ​​the portion surrounded by the straight line representing the target flow rate TG and the line representing the flow rate in the interval data SX during the period from the reference time te to the stable state of the flow rate variation (for example, Fig.13 The area A1 may be the total value of the area of ​​the oblique line portion of the flow rate during a predetermined period. The stable state is the same as the case of calculating the time t5. In addition, the area A1 may also represent the total value of the area of ​​the portion surrounded by the straight line representing the target flow rate TG and the line representing the flow rate in the interval data SX.

[0373] Furthermore, the characteristic amount calculation unit 613 analyzes the interval data SX including the variable portion of the flow rate and calculates the cycle T of the flow rate. The cycle T of the flow rate is information i of the third information X3. The cycle T represents the cycle of the flow rate change from the reference time te to the stable state. The stable state is the same as the case of calculating the time t5. In addition, the cycle T may also represent the cycle of the flow rate change in a predetermined period.

[0374] Furthermore, the inverse of the period T of the flow rate, that is, the frequency fa of the flow rate may also be the information i of the first information X1.

[0375] Above, as reference Fig.13 As described above, time t5, area A1, period T, and frequency fa are indices indicating the degree of fluctuation of the flow rate of the processing fluid. That is, time t5, area A1, period T, and frequency fa are indices indicating the degree of stability of the flow rate of the processing fluid.

[0376] Figure 14(a) to Figure 14(c) It is a diagram showing an example of a calculation method of the fourth information X4 included in the first feature amount information XD1.

[0377] As shown in FIG. 14( a), the feature quantity calculation unit 613 analyzes the interval data SX including the shutdown delay portion and calculates the time t6. The time t6 is information j of the fourth information X4. The time t6 represents the period from the time ty when the ejection control of the treatment fluid stops to the time tg when the flow rate of the treatment fluid starts to decrease. The time tg when the decrease starts represents when the flow rate reaches V% of the target value TG. "V%" represents, for example, a value greater than 70% and less than 100%.

[0378] As shown in Figure 14(b), the feature quantity calculation unit 613 analyzes the interval data SX including the shutdown delay portion and calculates the time t7. Time t7 is information k of the fourth information X4. Time t7 represents the period from the start of the decrease in the flow rate of the treatment fluid tg to the decrease in the flow rate of the treatment fluid th. The decrease time th indicates when the flow rate approaches the lower limit target value (for example, zero). Specifically, the decrease time th indicates when the flow rate reaches U% of the target value TG. "U%" represents, for example, a value below 10%. In addition, the decrease time th can also be recorded as "the suspected completion time th of the decrease".

[0379] Furthermore, the characteristic amount calculation unit 613 analyzes the interval data SX including the shutdown delay portion, and calculates the inclination angle θ2 indicating the downward inclination of the flow rate. The inclination angle θ2 is information m of the fourth information X4. The inclination angle θ2 is, for example, the inclination angle of a straight line connecting the flow value at the start of the decline tg of the treatment fluid and the flow value at the decline th, or the inclination angle of an approximate straight line indicating the flow value between the start of the decline tg and the decline th of the treatment fluid.

[0380] As shown in Fig. 14(c), the feature quantity calculation unit 613 analyzes the interval data SX including the shutdown delay portion and calculates the time t8. The time t8 is information 1 of the fourth information X4. The time t8 represents the period from the time ty when the ejection control of the treatment fluid stops to the time th when the flow rate of the treatment fluid decreases.

[0381] Above, as reference Figure 14(a) to Figure 14(c) As described above, the time t6 to the time t8 and the inclination angle θ2 are indicators showing the delay until the flow rate of the processing fluid reaches the lower limit target value when the flow rate of the processing fluid decreases toward the lower limit target value (for example, zero).

[0382] Fig.15 : is a diagram showing an example of a method for calculating the fifth information X5 included in the first feature information XD1. Fig.15 As shown, the feature quantity calculation unit 613 analyzes the interval data SX1 and the interval data SX2 including the overlapping parts, and calculates the time t9. Time t9 is the information n of the fifth information X5. The interval data SX1 is a part of one of the individual time series data TD adjacent to each other (the back end part in time), and the interval data SX2 is a part of another individual time series data TD (the back end part in time). Time t9 represents the period of the part in which the interval data SX1 and the interval data SX2 repeat. In other words, time t9 represents the period in which the flow rate shown by the interval data SX1 overlaps with the flow rate shown by the interval data SX2. In other words, time t9 represents the period in which the flow rate of one treatment fluid overlaps with the flow rate of another treatment fluid among different treatment fluids.

[0383] For example, time t9 represents the period from the rising start time ta of the flow rate shown in the interval data SX2 to the falling time th of the flow rate shown in the interval data SX1 in the overlapping part of the time series data TD1. That is, time t9 represents the time from the rising start time ta of the flow rate of one of the different treatment fluids to the falling time th of the flow rate of the other treatment fluid in the overlapping part of the time series data TD1.

[0384] Furthermore, the feature quantity calculation unit 613 analyzes the interval data SX1 and the interval data SX2 including the overlapping portion, and calculates the area A2. The area A2 is the information o of the fifth information X5. The area A2 represents the area of ​​the overlapping portion of the interval data SX1 and the interval data SX2. That is, the area A2 represents the area of ​​the overlapping portion of the flow rate of one treatment fluid and the flow rate of another treatment fluid in the different treatment fluids.

[0385] For example, area A2 represents the area of ​​the overlapping part of the time series data TD1, which is surrounded by the line representing the flow rate FL2 at ta when the flow rate shown in the interval data SX2 starts to rise to the intersection IN (flow rate FL), the line representing the flow rate FL1 at th when the flow rate shown in the interval data SX1 drops to the intersection IN (flow rate FL), and the line from the flow rate FL2 at ta when the flow rate starts to rise to the flow rate FL1 at th when the flow rate drops.

[0386] Furthermore, the characteristic amount calculation unit 613 analyzes the interval data SX1 and the interval data SX2 including the overlapping portion, and calculates the flow rate FL at the intersection IN. The flow rate FL at the intersection IN is the information p of the fifth information X5. The flow rate FL indicates the flow rate at the following intersection IN, which is the intersection of the line indicating the flow rate shown in the interval data SX1 and the line indicating the flow rate shown in the interval data SX2. That is, the flow rate FL indicates the flow rate at the following intersection IN, which is the intersection of the line indicating the flow rate of one of the different treatment fluids and the line indicating the flow rate of the other treatment fluid.

[0387] Above, as reference Fig.15 As described above, the time t9, the area A2, and the flow rate FL at the intersection IN are indicators showing the degree of overlap of the flow rates of the process fluids.

[0388] Fig.16 : is a diagram showing an example of a method for calculating the sixth information X6 included in the first feature amount information XD1. Fig.16 As shown, the feature quantity calculation unit 613 analyzes the interval data SX1 and the interval data SX2 including the intermittent part and calculates the time t10. The time t10 is the information q of the fifth information X5. The time t10 represents the time interval between the flow rate of the treatment fluid shown in the interval data SX1 and the flow rate of the treatment fluid shown in the interval data SX2. That is, the time t10 represents the time interval between one treatment fluid and another treatment fluid in the different treatment fluids on the time axis.

[0389] For example, time t10 represents the time from the decreasing time th of the flow rate shown in the interval data SX1 to the increasing start time ta of the flow rate shown in the interval data SX2 in the intermittent part of the time series data TD1. That is, it represents the time from the decreasing time th of the flow rate of one of the different processing fluids to the increasing start time ta of the flow rate of another processing fluid.

[0390] Above, as reference Fig.16 As described above, the time t10 is an index showing the degree of intermittency of the flow rate of the treatment fluid.

[0391] Here, Figure 11(a) to Figure 16 In the description, "flow rate" is exemplified as "physical quantity of the object", and information a to information p (time t1 to time t10, inclination angle θ1, inclination angle θ2, maximum value MX, area A1, area A2, period T, and flow rate FL) of the first feature quantity information XD1 are described. However, the information a to information p of the first feature quantity information XD1 can be defined in a generalized manner as follows.

[0392] Right now, Figure 11(a) to Figure 16 In the description of the first characteristic quantity information XD1, "flow rate" is changed to "physical quantity", "processing liquid" is changed to "object", "when ejection control starts" is changed to "when use control starts", and "when ejection control stops" is changed to "when use control stops". At this time, "when use control starts" means when the control unit 3A sends a use start indication signal to the part output object of the substrate processing apparatus 200. The use start indication signal indicates a signal indicating the start of use of the object. "When use control stops" means when the control unit 3A sends a use stop indication signal to the part output object of the substrate processing apparatus 200. The use stop indication signal indicates a signal indicating the stop of use of the object.

[0393] Above, as reference Figures 8 to 16 As described, according to the present embodiment, the first feature quantity information XD1 is represented by time. Therefore, the time-lapse feature of the interval data SX in the time series data TD1 can be easily and accurately extracted. As a result, by performing machine learning on the learning data TND containing the first feature quantity information XD1, a learning model LM with high estimation accuracy can be generated.

[0394] In particular, in this embodiment, the first feature quantity information XD1 representing the characteristics of the passage of time is extracted from the interval data SX which is a part of the time series data TD1. Therefore, the characteristic part contained in the interval data SX becomes one. As a result, it becomes easier to extract the characteristic part from the time series data TD1. Moreover, it is easy to mark the abnormal main cause information YD for the first feature quantity information XD1.

[0395] In other words, in the present embodiment, the acquisition unit 611 acquires the section data SX from the time series data TD1 so that the characteristic portion over time becomes one.

[0396] Furthermore, in this embodiment, if the timing data TD1 is directly input as raw data from the substrate processing apparatus 200 to the learning data generating apparatus 400, the learning data generating apparatus 400 calculates the feature value XD. Therefore, the user of the substrate processing apparatus 200 is not required to analyze and process the timing data TD1. As a result, the burden on the user can be reduced.

[0397] Next, refer to Fig.17 The abnormality main cause information YD is explained. Fig.17 It is a diagram showing abnormality main cause information YD.

[0398] First, the first abnormality main cause information YD1 of the abnormality main cause information YD will be described. Fig.17 As shown, the first abnormality main cause information YD1 includes at least one main cause information from the first main cause information Y101 to the ninth main cause information Y109. Fig.17 In the example of , the first abnormality main cause information YD1 includes first to ninth abnormality main cause information Y101 to Y109.

[0399] The first factor information Y101 is information indicating the degree of poor adhesion of the eluent to the chuck member 130 or the learning target substrate W1 due to a nozzle NZ positional defect.

[0400] The second factor information Y102 is information indicating the extent to which the processing liquid is scattered due to interference between the processing liquid ejected from the nozzle NZ to be switched and the processing liquid ejected from the other nozzles NZ when the nozzle NZ to be switched is switched to another nozzle NZ.

[0401] The third major cause information Y103 is information indicating the degree of instability of the flow rate of the process fluid (eg, chemical liquid or reaction gas).

[0402] The fourth major cause information Y104 is information indicating the degree of change in the concentration of the process fluid (eg, chemical liquid or reaction gas).

[0403] The fifth factor information Y105 is information indicating the degree of instability of the flow rate of the mixed fluid when a plurality of treatment fluids are mixed using a mixing valve such as the mixing valve V23.

[0404] The sixth major factor information Y106 is information indicating the extent to which the learning target substrate W1 is not covered with the processing liquid when the processing liquid ejected onto the learning target substrate W1 spreads on the surface of the learning target substrate W1 .

[0405] The seventh main cause information Y107 is information indicating the amount of the processing liquid that falls when the processing liquid attached to the outer surface of the nozzle NZ falls to the surface of the learning object substrate W1. For example, sometimes when the nozzle NZ to be switched is switched to another nozzle NZ, the processing liquid ejected from the nozzle NZ to be switched interferes with the processing liquid ejected from the other nozzle NZ, and the processing liquid is scattered. At this time, there is a possibility that the scattered processing liquid adheres to the outer surface of the nozzle NZ.

[0406] The eighth main cause information Y108 is information indicating the degree of mixing of the treatment fluid ejected first into the treatment fluid ejected later on the learning object substrate W1. For example, in the case where an intermittent portion must be ensured, the mixing of the treatment fluid ejected first into the treatment fluid ejected later may affect the processing of the learning object substrate W1. Moreover, in the case where an overlapping portion must be ensured, for example, the excessive mixing of the treatment fluid ejected first into the treatment fluid ejected later may also affect the processing of the learning object substrate W1.

[0407] The ninth main cause information Y109 is information indicating the degree to which other processing fluids are mixed into the processing fluid of the recycling object when the processing fluid is recycled after processing the learning object substrate W1. For example, in the case where the intermittent part must be ensured, if the processing fluid ejected first is mixed into the processing fluid ejected later, then when the processing fluid ejected later is recycled, it is possible to affect the processing of the learning object substrate W1. Moreover, for example, in the case where the overlapping part must be ensured, if the processing fluid ejected first is excessively mixed into the processing fluid ejected later, then when the processing fluid ejected later is recycled, it is possible to affect the processing of the learning object substrate W1.

[0408] Above, as reference Fig.17 As described above, the first main cause information Y101 to the ninth main cause information Y109 are information related to the processing fluid when processing the learning object substrate W1, and indicate the main cause of the abnormality of the learning object substrate W1 after being processed by the processing fluid (hereinafter sometimes recorded as "the main cause of abnormality"). Moreover, each of the first main cause information Y101 to the ninth main cause information Y109 can indicate the "degree" in two levels, such as "there is a problem" or "there is no problem", or can indicate the "degree" in three or more levels.

[0409] Next, the second abnormality main cause information YD2 of the abnormality main cause information YD will be described. Fig.17 As shown, the second abnormality main cause information YD2 includes at least one main cause information from the first main cause information Y201 to the seventh main cause information Y207. Fig.17In the example of , the second abnormality main cause information YD2 includes the first main cause information Y201 to the seventh main cause information Y207.

[0410] The first factor information Y201 is information indicating the degree of malfunction of a flow control valve such as the flow control valve V11. The flow control valve is, for example, a motor needle valve.

[0411] The second main cause information Y202 is information indicating the degree of malfunction of valves such as valve V12, valve V21, valve V22, and valve V23.

[0412] The third major cause information Y203 is information indicating the degree of malfunction of a pump such as the pump 55. The third major cause information Y203 may also include information indicating the degree of malfunction of a pulsation dampener such as the pulsation dampener 56. Malfunction of a pump or a pulsation dampener may cause pulsation of the treatment liquid.

[0413] The fourth major factor information Y204 is information indicating the degree of variation in the source pressure of the processing fluid supplied to the nozzle NZ.

[0414] The fifth main cause information Y205 is information indicating that the flow meter FW is malfunctioning.

[0415] The sixth major cause information Y206 is information indicating the degree of inappropriateness of the recipe information RP.

[0416] The seventh main cause information Y207 is information indicating the degree of setting error of the parameter information PM.

[0417] Above, as reference Fig.17 As described above, the first main cause information Y201 to the seventh main cause information Y207 are information related to the parts of the substrate processing apparatus 200, and indicate the main causes of abnormalities of the learning target substrate W1 after being processed by the processing fluid (hereinafter sometimes referred to as "main causes of abnormalities"). In addition, each of the first main cause information Y201 to the seventh main cause information Y207 can indicate the "degree" in two levels, such as "there is a problem" or "there is no problem", or can indicate the "degree" in three or more levels.

[0418] Next, refer to Fig.10 and Fig.17 , an example of the correlation between the first abnormality main cause information YD1, the second abnormality main cause information YD2, and the first feature amount information XD1 is described.

[0419] For example, it is possible that the main abnormal cause indicated by at least one of the first main cause information Y201, the second main cause information Y202, and the fifth main cause information Y205 of the second main abnormal cause information YD2 causes a larger or smaller state of the on-delay and / or off-delay of the process fluid. On the other hand, the on-delay of the process fluid is indicated by the first information X1 of the first feature quantity information XD1, and the off-delay of the process fluid is indicated by the fourth information X4 of the first feature quantity information XD1. Therefore, there is a correlation between the first feature quantity information XD1 and the second main abnormal cause information YD2.

[0420] Furthermore, the abnormal main cause indicated by at least one of the first main cause information Y201, the second main cause information Y202, and the fifth main cause information Y205 of the second abnormal main cause information YD2 may cause the abnormal main cause indicated by at least one of the sixth main cause information Y106, the second main cause information Y102, and the seventh main cause information Y107 of the first abnormal main cause information YD1. Therefore, the first abnormal main cause information YD1 is also related to the first characteristic quantity information XD1 related to the second abnormal main cause information YD2.

[0421] For example, the overshoot of the process fluid may be generated due to at least one of the first main cause information Y201 and the second main cause information Y202 of the second main cause information YD2. On the other hand, the overshoot of the process fluid is shown by the second information X2 of the first feature information XD1. Therefore, there is a correlation between the first feature information XD1 and the second main cause information YD2.

[0422] Moreover, the main cause of abnormality indicated in at least one of the first main cause information Y201 and the second main cause information Y202 of the second main cause information YD2 may cause the main cause of abnormality indicated in at least one of the first main cause information Y101 and the second main cause information Y102 of the first main cause information YD1. Therefore, the first main cause information YD1 is also related to the first characteristic quantity information XD1 related to the second main cause information YD2.

[0423] For example, it is possible that the change in the process fluid becomes larger due to the abnormal main cause indicated by at least one of the third main cause information Y203 and the fourth main cause information Y204 of the second abnormal main cause information YD2. On the other hand, the change in the process fluid is represented by the third information X3 of the first feature quantity information XD1. Therefore, there is a correlation between the first feature quantity information XD1 and the second abnormal main cause information YD2.

[0424] Furthermore, the main abnormal cause indicated by at least one of the third main cause information Y203 and the fourth main cause information Y204 of the second main abnormal cause information YD2 may cause the main abnormal cause indicated by at least one of the third main cause information Y103, the fourth main cause information Y104 and the fifth main cause information Y105 of the first main abnormal cause information YD1. Therefore, the first main abnormal cause information YD1 is also related to the first characteristic quantity information XD1 related to the second main abnormal cause information YD2.

[0425] For example, it is possible that the state where the overlap between the processing fluids becomes larger or smaller, or the state where the interval between the processing fluids becomes larger or smaller, is caused by the abnormal main cause indicated by at least one of the sixth main cause information Y206, the seventh main cause information Y207, the first main cause information Y201, and the second main cause information Y202 of the second abnormal main cause information YD2. On the other hand, the overlap between the processing fluids is represented by the fifth information X5 of the first feature quantity information XD1, and the interval between the processing fluids is represented by the sixth information X6 of the first feature quantity information XD1. Therefore, there is a correlation between the first feature quantity information XD1 and the second abnormal main cause information YD2.

[0426] Furthermore, the abnormal main cause indicated by at least one of the sixth main cause information Y206, the seventh main cause information Y207, the first main cause information Y201, and the second main cause information Y202 of the second abnormal main cause information YD2 may cause the abnormal main cause indicated by at least one of the sixth main cause information Y106, the second main cause information Y102, the eighth main cause information Y108, and the ninth main cause information Y109 of the first abnormal main cause information YD1. Therefore, the first abnormal main cause information YD1 is also related to the first characteristic quantity information XD1 related to the second abnormal main cause information YD2.

[0427] Next, refer to Fig.10 and Fig.17 , an example of the correlation between the abnormality main cause information YD and the second feature quantity information XD2 is described.

[0428] For example, the usage period of the parts of the substrate processing device 200 and the total usage time from when they were new vary depending on the processing sequence and / or batch interval within a batch. As a result, the main cause of the abnormality shown in the main cause information YD may be caused. On the other hand, the second feature quantity information XD2 includes the processing sequence information XA and the batch interval information XB. Therefore, there is a correlation between the main cause information YD and the second feature quantity information XD2.

[0429] Next, refer to Fig.10 and Fig.17 , an example of the correlation between the abnormality main cause information YD and the third feature quantity information XD3 is described.

[0430] For example, the interval data SX included in the third feature quantity information XD3 includes a time-lapse feature. In addition, the main abnormality cause shown in the abnormality main cause information YD is expressed as a time-lapse feature of the interval data SX. Therefore, there is a correlation between the abnormality main cause information YD and the third feature quantity information XD3.

[0431] Next, refer to Figure 8 and Fig.18 , the learning data production method performed by the learning data production device 400 is explained. Fig.18 FIG. 4 is a flowchart showing the method for preparing learning data according to the present embodiment. Fig.18 As shown, the learning data preparation method includes steps S21 to S24.

[0432] like Figure 8 and Fig.18 As shown, in step S21 , the acquisition unit 611 of the learning data creating apparatus 400 acquires the time series data TD1 from the substrate processing apparatus 200 .

[0433] Next, in step S22, the acquisition unit 611 acquires at least one section data SX from the time series data TD1. In this embodiment, the acquisition unit 611 acquires a plurality of section data SX from the time series data TD1.

[0434] Next, in step S23, the feature quantity calculation unit 613 of the learning data production device 400 calculates the first feature quantity information XD1 based on the plurality of interval data SX. Specifically, the feature quantity calculation unit 613 calculates the first information X1, the second information X2, the third information X3, the fourth information X4, the fifth information X5, or the sixth information X6 constituting the first feature quantity information XD1 for each interval data SX based on each interval data SX.

[0435] Next, in step S24, the acquisition unit 611 of the learning data creation device 400 receives the abnormality factor information YD via the input unit 4D.

[0436] Next, in step S25, the storage control unit 63 of the learning data production device 400 controls the storage unit 4B in such a manner that the abnormality main cause information YD is associated with the first feature quantity information XD1, the second feature quantity information XD2, and the third feature quantity information XD3 and stored. As a result, the storage unit 4B stores the abnormality main cause information YD in association with the first feature quantity information XD1, the second feature quantity information XD2, and the third feature quantity information XD3. In other words, the abnormality main cause information YD is labeled with the first feature quantity information XD1, the second feature quantity information XD2, and the third feature quantity information XD3, thereby producing the learning data TND.

[0437] As described above, by executing steps S21 to S24 by the learning data creating device 400 , creation of one learning data TND corresponding to one learning target substrate W1 is completed.

[0438] Next, refer to Fig.19 The learning device 500 will be described. The learning device 500 is, for example, a computer. Fig.19 is a block diagram showing the learning device 500. Fig.19 As shown, the learning device 500 includes a processing unit 5A, a storage unit 5B, a communication unit 5C, an input unit 5D, and a display unit 5E.

[0439] The processing unit 5A includes a processor such as a CPU and a GPU. The storage unit 5B includes a storage device that stores data and computer programs. The processor of the processing unit 5A executes the computer program stored in the storage device of the storage unit 5B to perform various processes. For example, the storage unit 5B includes a main storage device and an auxiliary storage device, similar to the storage unit 3B, and may also include a removable medium. The storage unit 5B is, for example, a non-temporary computer-readable storage medium.

[0440] Specifically, storage unit 5B stores processing program PG3. Processing unit 5A executes processing program PG3 and functions as acquisition unit 71, learning unit 73, storage control unit 75, and display control unit 77. That is, processing unit 5A includes acquisition unit 71, learning unit 73, storage control unit 75, and display control unit 77.

[0441] The communication unit 5C is connected to a network and communicates with an external device. The communication unit 5C is a communication device, such as a network interface controller.

[0442] The input unit 5D is an input device for inputting various information to the processing unit 5A. For example, the input unit 5D is a keyboard, a pointing device, or a touch screen.

[0443] The display unit 5E displays images and is, for example, a liquid crystal display or an organic electroluminescent display.

[0444] Next, refer to Fig.19 The processing unit 5A will be described. The acquisition unit 71 of the processing unit 5A acquires a plurality of learning data TND from the learning data creation device 400. For example, the acquisition unit 71 acquires a plurality of learning data TND from the learning data creation device 400 via the network and the communication unit 5C.

[0445] The storage control unit 75 controls the storage unit 5B so as to store each learning data TND. As a result, the storage unit 5B stores each learning data TND.

[0446] The storage unit 5B stores a learning program PG4. The learning program PG4 is a program for executing a machine learning algorithm for finding a certain rule from a plurality of learning data TND and generating a learned model LM expressing the found rule.

[0447] The machine learning algorithm is not particularly limited as long as it is supervised learning, and may be, for example, a decision tree, a nearest neighbor method, a naive Bayes classifier, a support vector machine, or a neural network. Therefore, the learned model LM includes a decision tree, a nearest neighbor method, a naive Bayes classifier, a support vector machine, or a neural network. In the machine learning for generating the learned model LM, the error back propagation method may also be used.

[0448] For example, a neural network includes an input layer, a single or multiple intermediate layers, and an output layer. Specifically, the neural network is a deep neural network (DNN), a recurrent neural network (RNN), or a convolutional neural network (CNN) for deep learning. For example, a deep neural network includes an input layer, multiple intermediate layers, and an output layer.

[0449] The learning unit 73 performs machine learning on a plurality of learning data TND based on the learning program PG4. As a result, a certain rule is found from the plurality of learning data TND, and a learned model LM is generated. That is, the learned model LM is constructed by performing machine learning on the learning data TND. The storage unit 5B stores the learned model LM.

[0450] Specifically, the learning unit 73 finds a certain rule between the feature quantity XD and the abnormality cause information YD included in the learning data TND, and generates a learned model LM. The feature quantity XD is an explanatory variable, and the abnormality cause information YD is a target variable.

[0451] More specifically, the learning unit 73 performs machine learning on a plurality of learning data TND based on the learning program PG4, thereby calculating a plurality of learned parameters, and generates a learned model LM including one or more functions that apply the plurality of learned parameters. The learned parameters are parameters (coefficients) obtained based on the results of machine learning using the plurality of learning data TND.

[0452] The learned model LM allows the computer to function in a manner that inputs input information (hereinafter sometimes recorded as "input information IF1") and outputs output information (hereinafter sometimes recorded as "output information IF2"). In other words, the learned model LM inputs input information IF1 and outputs output information IF2. Specifically, the learned model LM allows the computer to function in a manner that estimates the main cause of the abnormality of the processing target substrate W2 after being processed by the processing fluid. In other words, the learned model LM estimates the main cause of the abnormality of the processing target substrate W2 after being processed by the processing fluid.

[0453] Above, as reference Fig.19 As described, according to the present embodiment, the learning device 500 performs machine learning on the learning data TND including the first feature quantity information XD1, which expresses the time-lapse characteristics of the interval data SX of the time series data TD1 with high time accuracy. As a result, a learning model LM with high estimation accuracy of the main cause of the abnormality of the processing target substrate W2 can be generated.

[0454] In particular, the main causes of abnormality shown in the first main cause information YD1 (first main cause information Y201 to seventh main cause information Y207) and the second main cause information YD2 (first main cause information Y101 to ninth main cause information Y109) are: Fig.17 ) may cause particle defects on the substrate W. Therefore, the main cause of the abnormality when particle defects occur on the processing target substrate W2 can be estimated by using the learned model LM generated by performing machine learning on the abnormality main cause information YD (the first abnormality main cause information YD1 and the second abnormality main cause information YD2) as the target variable equivalent to the output of the learned model LM. The so-called particle defect is a defect of the substrate W caused by particles such as particles adhering to the substrate W.

[0455] In addition, in this embodiment, the first feature quantity information XD1 as the object of machine learning includes at least one of the first information X1 to the sixth information X6 ( Fig.10 ). Therefore, machine learning can be effectively performed on the characteristic part of the time passage of the time series data TD1.

[0456] Furthermore, in this embodiment, the second feature amount information XD2 as the object of machine learning includes at least one of the processing order information XA and the batch interval information XB ( Fig.10 ). In addition, for example, the usage period of the parts of the substrate processing device 200 and the total usage time from when they were new vary depending on the processing sequence and / or batch interval within the batch. As a result, the main cause of the abnormality shown in the main cause information YD may be caused. Therefore, by performing machine learning on the processing sequence information XA and / or the batch interval information XB, the estimation accuracy of the main cause of the abnormality of the learned model LM can be further improved.

[0457] Furthermore, in this embodiment, the third feature quantity information XD3 as the object of machine learning includes the interval data SX. Therefore, machine learning can be performed on the original data including the time-lapse feature of the time series data TD1. As a result, the estimation accuracy of the main cause of abnormality of the model LM after learning can be further improved. In particular, the interval data SX is part of the time series data TD1, so compared with the case where machine learning is performed on all the time series data TD1, the execution time of machine learning can be shortened.

[0458] Furthermore, the display control unit 77 controls the display unit 5E so as to display the learning data TND. As a result, the display unit 5E displays the learning data TND.

[0459] Next, refer to Fig.19 and Fig. 20 The learning method executed by the learning device 500 is described. Fig. 20 FIG. 4 is a flow chart showing the learning method of this embodiment. Fig. 20 As shown, the learning method includes steps S31 to S34.

[0460] like Fig.19 and Fig. 20 As shown, in step S31 , the acquisition unit 71 of the learning device 500 acquires a plurality of learning data TND from the learning data creating device 400 .

[0461] Next, in step S32 , the learning unit 73 performs machine learning on the plurality of learning data TND based on the learning program PG4 .

[0462] Next, in step S33, the learning unit 73 determines whether a learning end condition is satisfied. The learning end condition is a condition set in advance to end machine learning. The learning end condition is, for example, that the number of repetitions reaches a predetermined number of times.

[0463] If a negative determination is made in step S33, the process proceeds to step S31. As a result, machine learning is repeated.

[0464] On the other hand, in the case of an affirmative determination in step S33 , the process proceeds to step S34 .

[0465] In step S34, the learning unit 73 outputs a model (one or more functions) to which the latest plurality of parameters (coefficients), that is, a plurality of learned parameters (coefficients) are applied as the learned model LM. Next, the storage unit 5B stores the learned model LM.

[0466] As described above, the learning device 500 executes steps S31 to S34 to generate a learned model LM.

[0467] Next, refer to Fig.21 The following describes the abnormality cause estimation device 600. The abnormality cause estimation device 600 is, for example, a computer. Fig.21 FIG. 6 is a block diagram showing the abnormality main cause estimation device 600. Fig.21 As shown, abnormality cause estimation device 600 includes a processing unit 6A, a storage unit 6B, a communication unit 6C, an input unit 6D, and a display unit 6E.

[0468] The processing unit 6A includes a processor such as a CPU and a GPU. The storage unit 6B includes a storage device that stores data and computer programs. The processor of the processing unit 6A executes the computer program stored in the storage device of the storage unit 6B to perform various processes. For example, the storage unit 6B includes a main storage device and an auxiliary storage device, similar to the storage unit 3B, and may also include a removable medium. The storage unit 6B is, for example, a non-temporary computer-readable storage medium.

[0469] Specifically, the storage unit 6B stores the estimation program PG5. The processing unit 6A executes the estimation program PG5 and functions as a pre-processing unit 81, an estimation unit 83, a storage control unit 85, and a display control unit 87. That is, the processing unit 6A includes the pre-processing unit 81, the estimation unit 83, the storage control unit 85, and the display control unit 87. The pre-processing unit 81 includes an acquisition unit 811 and a feature quantity calculation unit 813.

[0470] The communication unit 6C is connected to a network and communicates with an external device. The communication unit 6C is a communication device, such as a network interface controller.

[0471] The input unit 6D is an input device for inputting various information to the processing unit 6A. For example, the input unit 6D is a keyboard, a pointing device, or a touch screen.

[0472] The display unit 6E displays images and is, for example, a liquid crystal display or an organic electroluminescent display.

[0473] Next, refer to Fig.21The processing unit 6A is described. The pre-processing unit 81 of the processing unit 6A produces input information IF1 based on the timing data TD2. The timing data TD2 represents the physical quantity of the object used by the substrate processing apparatus 300. The timing data TD2 includes at least one individual timing data TD. The content of the individual timing data TD is the same as the content of the individual timing data TD of the timing data TD1.

[0474] In addition, the storage control unit 85 controls the storage unit 6B to store the input information IF1. As a result, the storage unit 6B stores the input information IF1 in association with the lot identification information and the substrate identification information. For example, one piece of input information IF1 is generated for at least one time series data TD2 acquired when processing one processing target substrate W2.

[0475] The input information IF1 is an explanatory variable. That is, the input information IF1 is the information input to the learned model LM. The content of the input information IF1 and the feature quantity XD( Figure 8 ) have the same content.

[0476] The input information IF1 includes the first feature quantity information PD1. The input information IF1 may also include the second feature quantity information PD2 and / or the third feature quantity information PD3. In the present embodiment, the input information IF1 includes the first feature quantity information PD1, the second feature quantity information PD2 and the third feature quantity information PD3. The first feature quantity information PD1 is equivalent to an example of "first input information". The second feature quantity information PD2 is equivalent to an example of "second input information". The third feature quantity information PD3 is equivalent to an example of "third input information".

[0477] Specifically, the acquisition unit 811 of the pre-processing unit 81 acquires a plurality of time series data TD2 from the substrate processing apparatus 300. At this time, the batch identification information, the substrate identification information, the processing sequence information PA, and the batch interval information PB are attached to each time series data TD2. For example, the acquisition unit 811 acquires a plurality of time series data TD2 from the substrate processing apparatus 300 via the network and the communication unit 6C. The definitions of the batch identification information, the substrate identification information, the processing sequence information PA, and the batch interval information PB are respectively the same as the definitions of the batch identification information, the substrate identification information, the processing sequence information XA, and the batch interval information XB attached to the time series data TD1.

[0478] In addition, the storage unit 6B stores each time series data TD2 in association with the lot identification information, the substrate identification information, the processing order information XA, and the lot interval information XB.

[0479] Furthermore, the acquisition unit 811 acquires at least one interval data from the time series data TD2. In this embodiment, the acquisition unit 811 acquires multiple interval data from the time series data TD2. The interval data is data of a time interval during which a characteristic portion of the time series data appears, and is part of the time series data TD1. Hereinafter, the interval data is sometimes collectively referred to as "interval data SP".

[0480] In addition, the feature quantity calculation unit 813 calculates the first feature quantity information PD1 for each interval data SP based on the interval data SP. The first feature quantity information PD1 represents the time-dependent characteristics of the interval data SP. The first feature quantity information PD1 is represented by time. The operation of the feature quantity calculation unit 813 is similar to that of the feature quantity calculation unit 613 ( Figure 8 ) has the same action as

[0481] The second feature amount information PD2 is processing order information PA indicating the processing order of a predetermined number of processing target substrates W2 constituting one batch, or batch interval information PB indicating the time interval from the end of processing of a batch to the start of processing of the next batch.

[0482] The third feature quantity information PD3 is section data SP, object information (hereinafter referred to as “object information IDC”) or physical quantity information (hereinafter referred to as “physical quantity information IDD”). The section data SP is section data acquired by the acquisition unit 811 from the time series data TD2.

[0483] The object information IDC is information for identifying an object used in the substrate processing apparatus 300 .

[0484] The physical quantity information IDD is information for identifying the type of the physical quantity of an object used in the substrate processing apparatus 300 .

[0485] Above, as reference Fig.21 As described above, the storage unit 6B stores the first feature amount information PD1 , the second feature amount information PD2 , and the third feature amount information PD3 as the input information IF1 .

[0486] Furthermore, the storage unit 6B stores the learned model LM.

[0487] The estimation unit 83 inputs the input information IF1 to the learned model LM, and obtains the output information IF2 from the learned model LM. The output information IF2 indicates the main cause of the abnormality of the processing target substrate W2 after being processed by the processing fluid. The output information IF2 is the target variable. That is, the output information IF2 is the information output from the learned model LM. The content of the output information IF2 is consistent with the abnormality main cause information YD( Figure 8 ) have the same content.

[0488] Specifically, the output information IF2 includes at least one of the first abnormal main cause information QD1 and the second abnormal main cause information QD2. In the present embodiment, the output information IF2 includes the first abnormal main cause information QD1 and the second abnormal main cause information QD2. The first abnormal main cause information QD1 is equivalent to an example of "first output information", and the second abnormal main cause information QD2 is equivalent to an example of "second output information".

[0489] The first abnormality main cause information QD1 is information related to the processing fluid when the substrate processing apparatus 300 processes the processing target substrate W2, and is information indicating the main cause of the abnormality of the processing target substrate W2 after being processed by the processing fluid. The second abnormality main cause information QD2 is information related to the parts of the substrate processing apparatus 300, and is information indicating the main cause of the abnormality of the processing target substrate W2 after being processed by the processing fluid.

[0490] The display control unit 65 controls the display unit 6E so as to display the output information IF2. As a result, the display unit 6E displays the output information IF2.

[0491] Above, as reference Fig.21 As described, according to the present embodiment, the learned model LM is generated by machine learning the learning data TND including the first feature quantity information XD1, which uses time to accurately represent the time-lapse characteristics of the interval data SX in the time series data TD1. As a result, by using the learned model LM, the main cause of the abnormality of the processing target substrate W2 can be estimated with high accuracy.

[0492] Furthermore, in this embodiment, if the time series data TD2 is directly input as raw data from the substrate processing apparatus 300 to the abnormality main cause estimation apparatus 600, the pre-processing unit 81 calculates the input information IF1. Therefore, the user of the substrate processing apparatus 300 is not required to analyze and process the time series data TD2. As a result, the burden on the user can be reduced.

[0493] Next, refer to Fig. 22 The input information IF1 will be described. Fig. 22 is a diagram showing input information IF1. Fig. 22 As shown, the first feature amount information PD1 of the input information IF1 includes at least one of the first information P1 to the sixth information P6. Fig. 22 In the example of , the first feature amount information PD1 includes first information P1 to sixth information P6.

[0494] The first information P1 is information indicating the state of the physical quantity when the physical quantity increases toward the target value. The first information P1 includes at least one of information a to information d. The second information P2 is information indicating the overshoot of the physical quantity. The second information P2 includes at least one of information e and information f. The third information P3 is information indicating the change of the physical quantity. The third information P3 includes at least one of information g to information i.

[0495] The fourth information P4 is information indicating the state of the physical quantity when the physical quantity decreases from the target value. The fourth information P4 includes at least one of the information j to the information m. The fifth information P5 is information indicating the overlap of the physical quantity of one of the two objects with the physical quantity of the other object when the substrate processing apparatus 300 uses two different objects. The fifth information P5 includes at least one of the information n to the information p. The sixth information P6 is information indicating the time interval between one of the two objects and the other object on the time axis when the substrate processing apparatus 300 uses at least two different objects. The sixth information P6 includes information q.

[0496] Fig. 22 In the example of , the first information P1 includes information a to information d. The second information P2 includes information e and information f. The third information P3 includes information g to information i. The fourth information P4 includes information j to information m. The fifth information P5 includes information n to information p.

[0497] The second feature amount information PD2 includes at least one of the processing order information PA and the batch interval information PB. Fig. 22 In the example of , the second feature amount information PD2 includes processing order information PA and batch interval information PB.

[0498] The third feature quantity information PD3 includes at least one of the section data SP, the object information IDC, and the physical quantity information IDD. Fig. 22 In the example of , the third feature quantity information PD3 includes section data SP, object information IDC, and physical quantity information IDD.

[0499] Here, the content of the first feature quantity information PD1 is the same as the content of the first feature quantity information XD1. Specifically, the content of the first information P1 to the sixth information P6 is the same as the content of the first information X1 to the sixth information X6. The content of information a to information q of the first information P1 to the sixth information P6 is the same as the content of information a to information q of the first information X1 to the sixth information X6. Moreover, the content of the second feature quantity information PD2 is the same as the content of the second feature quantity information XD2. Specifically, the content of the processing sequence information PA and the batch interval information PB is the same as the content of the processing sequence information XA and the batch interval information XB. Furthermore, the content of the third feature quantity information PD3 is the same as the content of the third feature quantity information XD3. Specifically, the content of the interval data SP, the object information IDC, and the physical quantity information IDD is the same as the content of the interval data SX, the object information IDA, and the physical quantity information IDB. Moreover, the content of the timing data TD2 output by the substrate processing apparatus 300 is the same as the content of the timing data TD1 output by the substrate processing apparatus 200. Furthermore, the content of the interval data SP is the same as the content of the interval data SX.

[0500] Therefore, by describing the first feature quantity information XD1, the first information X1 to the sixth information X6, the information a to the information q of the first information X1 to the sixth information X6, the second feature quantity information XD2, the processing sequence information XA, the batch interval information XB, the third feature quantity information XD3, the interval data SX, the object information IDA, the physical quantity information IDB, the time series data TD1 and the interval data SX ( Figure 8 to Figure 17 ), appropriately renaming the acquisition unit 611 as the acquisition unit 811, renaming the feature quantity calculation unit 613 as the feature quantity calculation unit 813, renaming the storage control unit 63 as the storage control unit 85, renaming the storage unit 4B as the storage unit 6B, renaming the substrate processing device 200 as the substrate processing device 300, and renaming the learning object substrate W1 as the processing object substrate W2, thereby replacing the first feature quantity information PD1, the first information P1 to the sixth information P6, the information a to the information q of the first information P1 to the sixth information P6, the processing sequence information PA, the batch interval information PB, the interval data SP, the object information IDC, the physical quantity information IDD, the time series data TD2, the interval data SP, the acquisition unit 811, the feature quantity calculation unit 813, the storage control unit 85 and the storage unit 6B.

[0501] Next, refer to Fig.23 The output information IF2 will be described. Fig.23 is a diagram showing output information IF2. Fig.23 As shown, the first abnormality main cause information QD1 of the output information IF2 includes at least one main cause information among the first main cause information Q101 to the ninth main cause information Q109. Fig.23In the example of , the first abnormality main cause information QD1 includes first main cause information Q101 to ninth main cause information Q109.

[0502] The second abnormality main cause information QD2 includes at least one main cause information among the first main cause information Q201 to the seventh main cause information Q207. Fig.23 In the example of , the second abnormality main cause information QD2 includes the first main cause information Q201 to the seventh main cause information Q207.

[0503] Here, the content of the first abnormality main cause information QD1 is the same as the content of the first abnormality main cause information YD1. Specifically, the content of the first main cause information Q101 to the ninth main cause information Q109 is the same as the content of the first main cause information Y101 to the ninth main cause information Y109 ( Fig.17 ) is the same as the content of the second abnormal main cause information QD2. Furthermore, the content of the second abnormal main cause information QD2 is the same as the content of the second abnormal main cause information YD2. Specifically, the content of the first main cause information Q201 to the seventh main cause information Q207 is the same as the content of the first main cause information Y201 to the seventh main cause information Y207 ( Fig.17 ) have the same content.

[0504] Therefore, by describing the first abnormality main cause information YD1, the first main cause information Y101 to the ninth main cause information Y109, the second abnormality main cause information YD2, and the first main cause information Y201 to the seventh main cause information Y207 ( Figure 8 to Figure 17 ), the substrate processing device 200 is appropriately renamed as the substrate processing device 300, and the learning object substrate W1 is renamed as the processing object substrate W2, thereby replacing the description of the first abnormal main cause information QD1, the first main cause information Q101 to the ninth main cause information Q109, the second abnormal main cause information QD2, and the first main cause information Q201 to the seventh main cause information Q207.

[0505] Moreover, similarly to the case where the first abnormality main cause information YD1 and the second abnormality main cause information YD2 are correlated with the first feature amount information XD1, the first abnormality main cause information QD1 and the second abnormality main cause information QD2 are also correlated with the first feature amount information PD1.

[0506] Next, refer to Fig.21 and Fig.24 The abnormality cause estimating method executed by the abnormality cause estimating device 600 will be described. Fig.24 FIG. 4 is a flowchart showing the method for estimating the main cause of abnormality according to the present embodiment. Fig.24As shown, the abnormality cause estimation method includes steps S41 to S49.

[0507] like Fig.21 and Fig. 22 As shown, in step S41 , the acquisition unit 811 of the abnormality cause estimation apparatus 600 acquires the time series data TD2 from the substrate processing apparatus 300 .

[0508] Next, in step S42, the acquisition unit 811 acquires at least one section data SP from the time series data TD2. In this embodiment, the acquisition unit 811 acquires a plurality of section data SP from the time series data TD2.

[0509] Next, in step S43, the feature quantity calculation unit 813 of the abnormality main cause estimation device 600 calculates the first feature quantity information PD1 based on the plurality of interval data SP. Specifically, the feature quantity calculation unit 813 calculates the first information P1, the second information P2, the third information P3, the fourth information P4, the fifth information P5 or the sixth information P6 constituting the first feature quantity information PD1 for each interval data SP based on each interval data SP.

[0510] Next, in step S44, the storage control unit 85 of the abnormality main cause estimation device 600 controls the storage unit 6B so as to store the first feature amount information PD1, the second feature amount information PD2, and the third feature amount information PD3 as the input information IF1. As a result, the storage unit 6B stores the first feature amount information PD1, the second feature amount information PD2, and the third feature amount information PD3 as the input information IF1.

[0511] Next, in step S45 , the estimating unit 83 acquires the input information IF1 from the storage unit 6B.

[0512] Next, in step S46, the estimating unit 83 inputs the input information IF1 to the learned model LM. As a result, the learned model LM estimates the main cause of the abnormality of the processing target substrate W2 processed by the processing fluid, and outputs the output information IF2.

[0513] It is preferable that the learned model LM outputs the first main cause information Q101 to the ninth main cause information Q109 and the first main cause information Q201 to the seventh main cause information Q207 constituting the output information IF2 in order of likelihood from high to low.

[0514] Next, in step S47, the estimation unit 83 acquires output information IF2 from the learned model LM. The output information IF2 indicates the main cause of the abnormality of the processing target substrate W2 processed by the processing fluid.

[0515] Next, in step S48, the storage control unit 85 controls the storage unit 6B so as to store the output information IF2. As a result, the storage unit 6B stores the output information IF2.

[0516] Next, in step S49, the display control unit 87 controls the display unit 6E so as to display the output information IF2. As a result, the display unit 6E displays the output information IF2.

[0517] Preferably, the display control unit 87 controls the display unit 6E so as to display the first main cause information Q101 to the ninth main cause information Q109 and the first main cause information Q201 to the seventh main cause information Q207 constituting the output information IF2 in the order of high likelihood. As a result, the display unit 6E displays the first main cause information Q101 to the ninth main cause information Q109 and the first main cause information Q201 to the seventh main cause information Q207 in the order of high likelihood.

[0518] As described above, the abnormality cause estimating device 600 executes steps S41 to S49, thereby outputting one piece of output information IF2 corresponding to one piece of the process target substrate W2, and the process ends.

[0519] In the above description, the flow rate of the processing fluid is mainly described as a typical example of the physical quantity of the object. Other examples of the physical quantity of the object will be described below.

[0520] [Object: Processing fluid, Physical quantity: Temperature]

[0521] When the time series data TD1 outputted from the substrate processing apparatus 200 indicates the temperature of the processing fluid detected by the temperature sensor TS, the first feature quantity information XD1 includes at least one of the first information X1 to the fourth information X4. Fig.10 In the definitions of the first information X1 to the fourth information X4 described above, the processing fluid is changed to temperature. By using the temperature as the time series data TD1, the learned model LM is generated by machine learning, so that the learned model LM can be used to estimate the main cause of the abnormality when the etching failure occurs on the substrate W. The reason is that the temperature of the processing fluid has an influence on the etching.

[0522] Similar to the first feature quantity information XD1 in the learning phase, the first feature quantity information PD1 in the utilization phase also includes, for example, at least one of the first information P1 to the fourth information P4, where the processed fluid is replaced by temperature.

[0523] [Object: Processing fluid, Physical quantity: Concentration]

[0524] When the time series data TD1 outputted from the substrate processing apparatus 200 indicates the concentration of the processing fluid detected by the sensor SN, the first feature quantity information XD1 includes, for example, the third information X3. Fig.10 In the definition of the third information X3 described in the above, the processing fluid is changed to concentration. By using the concentration as the time series data TD1, the learned model LM is generated by machine learning, so that the learned model LM can be used to estimate the main cause of the abnormality when the etching of the substrate W is poor. The reason is that the concentration of the processing fluid has an impact on the etching.

[0525] Similar to the first feature quantity information XD1 in the learning phase, the first feature quantity information PD1 in the utilization phase also includes, for example, third information P3 in which the processing fluid is replaced by concentration.

[0526] [Object: rotating chuck 13, physical quantity: rotation speed]

[0527] When the timing data TD1 outputted from the substrate processing apparatus 200 indicates the rotation speed of the spin chuck 13 detected by the sensor SN, the first feature quantity information XD1 includes at least one of the first information X1 to the fourth information X4. Fig.10 In the definitions of the first information X1 to the fourth information X4 described above, the processing fluid is renamed as the rotary chuck 13, and the flow rate is renamed as the rotation speed. By using the rotation speed as the time series data TD1, the learned model LM is generated by machine learning, so that the learned model LM can be used to infer the main cause of the abnormality when the substrate W has a particle defect.

[0528] Similar to the first feature quantity information XD1 in the learning phase, the first feature quantity information PD1 in the utilization phase also includes at least one of the first information P1 to the fourth information P4, for example, the processing fluid is renamed the spin chuck 13, and the flow rate is renamed the rotation speed.

[0529] [Object: blocking plate 291, physical quantity: position, displacement, rotation speed]

[0530] When the timing data TD1 outputted from the substrate processing apparatus 200 indicates the position, displacement or rotation speed of the blocking plate 291 detected by the sensor SN, the first feature quantity information XD1 is represented by, for example, time that can be calculated from the position or displacement of the blocking plate 291 during lifting or lowering, or by the rotation speed of the blocking plate 291 itself during rotation. The same is true for the first feature quantity information PD1.

[0531] [Object: Arm AM, Physical quantity: position, displacement]

[0532] When the timing data TD1 outputted by the substrate processing apparatus 200 indicates the position or displacement of the arm AM detected by the sensor SN, the first feature quantity information XD1 is represented by time that can be calculated from the position or displacement of the arm AM during the scanning process.

[0533] [Object: guard plate 33, physical quantity: position, displacement]

[0534] When the timing data TD1 outputted from the substrate processing apparatus 200 indicates the position or displacement of the guard 33 detected by the sensor SN, the first feature quantity information XD1 is represented by, for example, time that can be calculated from the position or displacement of the guard 33 when the guard 33 is raised or lowered. The same is true for the first feature quantity information PD1.

[0535] [Object: valve VB, physical quantity: opening]

[0536] When the time series data TD1 outputted from the substrate processing apparatus 200 indicates the opening of the valve VB detected by the sensor SN, the first feature quantity information XD1 is represented by, for example, the time variation of the opening of the valve VB when the valve VB is opened and closed. The same is true for the first feature quantity information PD1.

[0537] Next, a case will be described in which the object used in the substrate processing apparatuses 200 and 300 is a processing fluid and the physical quantity of the object represents the physical quantity of the processing fluid in the discharge pipe P7 from which the processing fluid is discharged.

[0538] [Flow meter FW3 of discharge pipe P7]

[0539] The timing data TD1 output by the substrate processing apparatus 200 is inserted in Figure 5 In the case of the flow rate of the treatment fluid detected by the flow meter FW3 of the discharge pipe P7 shown in the figure, the first feature quantity information XD1 is information in which the characteristic part of the time transition of the flow rate is expressed by time. The same is true for the first feature quantity information PD1.

[0540] [Temperature sensor TS7 of discharge pipe P7]

[0541] The timing data TD1 output by the substrate processing apparatus 200 is configured in Figure 5 In the case of the temperature of the process fluid detected by the temperature sensor TS7 of the discharge pipe P7 shown in the figure, the first feature quantity information XD1 is information representing the characteristic portion of the time transition of the temperature by time. The same is true for the first feature quantity information PD1.

[0542] Here, Figure 1The physical structure of the substrate processing apparatus 200, the substrate processing apparatus 300, the learning data preparation apparatus 400, the learning apparatus 500, and the abnormality main cause estimation apparatus 600 may be indicated, or the logical structure may be indicated. Therefore, part or all of the substrate processing apparatus 200, the substrate processing apparatus 300, the learning data preparation apparatus 400, the learning apparatus 500, and the abnormality main cause estimation apparatus 600 may be configured as one apparatus.

[0543] For example, the substrate processing apparatus 200 may include the learning data creating apparatus 400. In this case, for example, the control unit 3A and the storage unit 3B of the substrate processing apparatus 200 function as the processing unit 4A and the storage unit 4B of the learning data creating apparatus 400.

[0544] For example, the substrate processing apparatus 200 may include the learning data generating apparatus 400 and the learning apparatus 500. In this case, for example, the control unit 3A and the storage unit 3B of the substrate processing apparatus 200 function as the processing unit 4A and the storage unit 4B of the learning data generating apparatus 400, and the processing unit 5A and the storage unit 5B of the learning apparatus 500.

[0545] For example, the substrate processing apparatus 200 may also include a learning data preparation device 400, a learning device 500, and an abnormality main cause estimation device 600. In this case, for example, the control unit 3A and the storage unit 3B of the substrate processing apparatus 200 function as the processing unit 4A and the storage unit 4B of the learning data preparation device 400, the processing unit 5A and the storage unit 5B of the learning device 500, and the processing unit 6A and the storage unit 6B of the abnormality main cause estimation device 600.

[0546] Similarly, the substrate processing apparatus 200 may include the learning data creating apparatus 400 , or may include the learning data creating apparatus 400 and the learning apparatus 500 , or may include the learning data creating apparatus 400 , the learning apparatus 500 , and the abnormality main cause estimating apparatus 600 .

[0547] For example, the abnormality main cause estimation device 600 may also include the learning device 500. For example, the abnormality main cause estimation device 600 may also include the learning device 500 and the learning data production device 400. In this case, the duplicated functions in the pre-processing unit 81 and the learning data production unit 61 may be omitted. For example, the learning device 500 may also include the learning data production device 400.

[0548] For example, a server or a host computer may receive the output information IF2 from the abnormality cause estimation device 600 via a network, and display the output information IF2 on a display device.

[0549] For example, two or more of the learning data creating device 400, the learning device 500, and the abnormality cause estimating device 600 may be combined, and the combined device may be constituted by a server.

[0550] For example, at least one of the learning data creating device 400, the learning device 500, and the abnormality cause estimating device 600 may be constituted by a server.

[0551] For example, the substrate processing apparatus 200 that processes the learning target substrate W1 may function as the substrate processing apparatus 300 that processes the processing target substrate W2 , or vice versa.

[0552] For example, the pre-processing unit 81 and the estimating unit 83 of the abnormality cause estimating device 600 may be configured as separate devices.

[0553] The embodiments of the present invention are described above with reference to the accompanying drawings. However, the present invention is not limited to the embodiments, and can be implemented in various ways within the scope of the present invention. Moreover, the multiple structural elements disclosed in the embodiments can be appropriately changed. For example, a structural element of all the structural elements shown in a certain embodiment can be added to the structural elements of other embodiments, or some structural elements of all the structural elements shown in a certain embodiment can be deleted from the embodiments.

[0554] Moreover, the drawings are intended to facilitate understanding of the invention, and the components are schematically shown on the main body, and the thickness, length, number, spacing, etc. of the components shown in the drawings may be different from the actual components for the convenience of making the drawings. Moreover, the structure of each component shown in the above embodiment is an example and is not particularly limited. Of course, various changes can be made within the scope that does not substantially deviate from the effect of the present invention.

[0555] [Industrial Applicability]

[0556] The present invention relates to a method for generating a model after learning, a device for estimating the main cause of an abnormality, a substrate processing device, a method for estimating the main cause of an abnormality, a learning method, a learning device and a method for producing learning data, and has industrial applicability.

Claims

1. A method for generating a learned model, comprising the following steps: Access to learning data; and A learning model is generated by performing machine learning on the learning data, wherein the learning model estimates the main cause of an abnormality of the substrate to be processed after being processed by the processing fluid. The learning data includes feature quantities and abnormal main cause information, In the step of generating the learned model, the machine learning is performed by using the feature quantity as an explanatory variable equivalent to the input of the learned model and using the abnormality main cause information as a target variable equivalent to the output of the learned model, The abnormality main cause information indicates the main cause of the abnormality of the learning object substrate after being processed by the processing fluid, and includes at least one of the first abnormality main cause information and the second abnormality main cause information. The feature quantity includes first feature quantity information, the first feature quantity information is a feature representing the time passage of interval data in time series data, the time series data representing a physical quantity of an object used by a substrate processing device that processes the learning object substrate using the processing fluid, The first feature quantity information is represented by time, The first abnormality main cause information and the second abnormality main cause information are related to the first feature quantity information, The learned model inputs input information and outputs output information. The input information includes first input information showing a time-lapse feature of interval data in time series data representing a physical quantity of an object used in a substrate processing apparatus that processes the processing target substrate with the processing fluid, The output information indicates a major cause of an abnormality of the substrate to be processed after being processed by the processing fluid, and includes at least one of first output information and second output information. The first output information and the second output information are related to the first input information, The first abnormal main cause information and the first output information each include at least one main cause information of first main cause information, second main cause information, third main cause information, fourth main cause information, fifth main cause information, sixth main cause information, seventh main cause information, eighth main cause information, and ninth main cause information, The first major cause information is information indicating the degree of poor adhesion of the treatment fluid. The second main cause information is information indicating the degree of scattering of the treatment fluid. The third main cause information is information showing the degree of instability of the flow rate of the process fluid, The fourth major cause information is information showing the degree of change in the concentration of the treatment fluid. The fifth major cause information is information showing the degree of instability of the flow rate of the mixed fluid when a plurality of the treatment fluids are mixed. The sixth main cause information is information showing the extent to which the substrate is not covered by the processing fluid, The seventh major cause information is information showing the amount of the treatment fluid that falls when the treatment fluid attached to the outer surface of the nozzle falls. The eighth major cause information is information showing the degree to which the processing fluid ejected first to the substrate is mixed with the processing fluid ejected later. The ninth major cause information is information showing the degree of mixing of other treatment fluids different from the treatment fluid to be recovered, The second abnormality main cause information and the second output information each include at least one main cause information of the tenth main cause information, the eleventh main cause information, the twelfth main cause information, and the thirteenth main cause information, The tenth major cause information is information showing the degree of change in the source pressure of the treatment fluid supplied to the nozzle. The eleventh major cause information is information indicating that the flow meter is malfunctioning. The twelfth major cause information is information indicating the degree of inappropriateness of the recipe information that specifies the processing content and the processing sequence. The thirteenth major cause information is parameter information indicating the setting value of the component of the substrate processing device, and information indicating the degree of setting error. said object being said process fluid, The physical quantity represents the flow rate of the treatment fluid, the temperature of the treatment fluid, or the concentration of the treatment fluid, and When the physical quantity is the flow rate of the processing fluid, the abnormality of the processing target substrate and the abnormality of the learning target substrate represent defects caused by particles. When the physical quantity is the temperature or concentration of the processing fluid, the abnormality of the processing target substrate and the abnormality of the learning target substrate indicate etching failure.

2. The method for generating a model after learning according to claim 1, wherein: The first abnormality main cause information is information related to the processing fluid when the substrate processing device processes the learning target substrate. The second abnormality main cause information is information related to a component of the substrate processing apparatus.

3. The method for generating a model after learning according to claim 1 or 2, wherein: The first feature information includes at least one of first information, second information, third information, fourth information, fifth information and sixth information. The first information is information indicating a state of the physical quantity when the physical quantity increases toward a target value. The second information is information indicating an overshoot of the physical quantity, The third information is information indicating a change in the physical quantity. The fourth information is information indicating a state of the physical quantity when the physical quantity decreases from the target value. The fifth information is information indicating an overlap between the physical quantity of one of the two objects and the physical quantity of the other object when the substrate processing apparatus uses at least two of the two objects that are different from each other. The sixth information is information indicating a time interval between the one object and the other object on a time axis.

4. The method for generating a model after learning according to claim 1 or 2, wherein: The feature quantity also includes second feature quantity information, The second feature amount information is processing order information indicating a processing order for a predetermined number of the learning target substrates constituting one batch, or batch interval information indicating a time interval from the end of processing a batch to the start of processing a next batch.

5. An abnormality main cause estimation device, which estimates the main cause of abnormality of a substrate to be processed after being processed by a processing fluid, and comprises: A storage unit for storing a learning model constructed by performing machine learning on learning data; as well as an estimating unit that inputs input information to the learned model and obtains output information from the learned model, The learning data includes feature quantities and abnormal main cause information, The learned model is constructed by executing the machine learning using the feature quantity as an explanatory variable equivalent to an input of the learned model and the abnormality main cause information as a target variable equivalent to an output of the learned model, The abnormality main cause information indicates the main cause of the abnormality of the learning object substrate after being processed by the processing fluid, and includes at least one of the first abnormality main cause information and the second abnormality main cause information. The feature quantity includes first feature quantity information, the first feature quantity information is a feature representing the time passage of interval data in time series data, the time series data represents a physical quantity of an object used by a substrate processing device that processes the learning object substrate using the processing fluid, the first feature quantity information is represented by time, The first abnormality main cause information and the second abnormality main cause information are related to the first feature quantity information, The learned model inputs input information and outputs output information. The input information includes first input information, the first input information represents a time-lapse feature of interval data in time series data, the time series data represents a physical quantity of an object used by a substrate processing device that processes the processing target substrate using the processing fluid, the first input information being represented by time, The output information indicates a main cause of an abnormality of the substrate to be processed after being processed by the processing fluid, and includes at least one of first output information and second output information. The first output information and the second output information are related to the first input information, The first abnormal main cause information and the first output information each include at least one main cause information of first main cause information, second main cause information, third main cause information, fourth main cause information, fifth main cause information, sixth main cause information, seventh main cause information, eighth main cause information, and ninth main cause information, The first major cause information is information indicating the degree of poor adhesion of the treatment fluid. The second main cause information is information indicating the degree of scattering of the treatment fluid. The third main cause information is information showing the degree of instability of the flow rate of the process fluid, The fourth major cause information is information showing the degree of change in the concentration of the treatment fluid. The fifth major cause information is information showing the degree of instability of the flow rate of the mixed fluid when a plurality of the treatment fluids are mixed. The sixth main cause information is information showing the extent to which the substrate is not covered by the processing fluid, The seventh major cause information is information showing the amount of the treatment fluid that falls when the treatment fluid attached to the outer surface of the nozzle falls. The eighth major cause information is information showing the degree to which the processing fluid ejected first to the substrate is mixed with the processing fluid ejected later. The ninth major cause information is information showing the degree of mixing of other treatment fluids different from the treatment fluid to be recovered, The second abnormality main cause information and the second output information each include at least one main cause information of the tenth main cause information, the eleventh main cause information, the twelfth main cause information, and the thirteenth main cause information, The tenth major cause information is information showing the degree of change in the source pressure of the treatment fluid supplied to the nozzle. The eleventh major cause information is information indicating that the flow meter is malfunctioning. The twelfth major cause information is information indicating the degree of inappropriateness of the recipe information that specifies the processing content and the processing sequence. The thirteenth major cause information is parameter information indicating the setting value of the component of the substrate processing device, and information indicating the degree of setting error. said object being said process fluid, The physical quantity represents the flow rate of the treatment fluid, the temperature of the treatment fluid, or the concentration of the treatment fluid, and When the physical quantity is the flow rate of the processing fluid, the abnormality of the processing target substrate and the abnormality of the learning target substrate represent defects caused by particles. When the physical quantity is the temperature or concentration of the processing fluid, the abnormality of the processing target substrate and the abnormality of the learning target substrate indicate etching failure.

6. The abnormality main cause estimation device according to claim 5, wherein: The first abnormality main cause information is information related to the processing fluid when the substrate processing device processes the learning target substrate. The second abnormality main cause information is information related to a part of the substrate processing device that processes the learning target substrate. The first output information is information related to the processing fluid when the substrate processing apparatus processes the processing target substrate. The second output information is information related to a component of the substrate processing apparatus that processes the processing target substrate.

7. The abnormality main cause estimation device according to claim 5 or 6, wherein: The first feature information and the first input information each include at least one of first information, second information, third information, fourth information, fifth information, and sixth information, The first information is information indicating a state of the physical quantity when the physical quantity increases toward a target value. The second information is information indicating an overshoot of the physical quantity, The third information is information indicating a change in the physical quantity. The fourth information is information indicating a state of the physical quantity when the physical quantity decreases from the target value. The fifth information is information indicating an overlap between the physical quantity of one of the two objects and the physical quantity of the other object when the substrate processing apparatus uses at least two of the two objects that are different from each other. The sixth information is information indicating a time interval between the one object and the other object on a time axis.

8. The abnormality main cause estimation device according to claim 5 or 6, wherein: The feature quantity also includes second feature quantity information, The second feature amount information is processing order information indicating the processing order of a predetermined number of the learning target substrates constituting one batch, or batch interval information indicating the time interval from the end of processing of a batch to the start of processing of the next batch, The input information further includes second input information related to the processing target substrate. The second input information is processing order information indicating the processing order of a predetermined number of the processing target substrates constituting one batch, or batch interval information indicating a time interval from the end of processing of a batch to the start of processing of the next batch.

9. A substrate processing device, comprising: The abnormality main cause estimation device according to any one of claims 5 to 8; as well as The processing device processes the substrate.

10. A method for estimating a main cause of an abnormality, the method comprising estimating a main cause of an abnormality of a substrate to be processed after being processed by a processing fluid, and comprising the following steps: Obtaining input information; and inputting the input information into a learned model constructed by machine learning of the learning data, and obtaining output information from the learned model, The learning data includes feature quantities and abnormal main cause information, The learned model is constructed by executing the machine learning using the feature quantity as an explanatory variable equivalent to an input of the learned model and the abnormality main cause information as a target variable equivalent to an output of the learned model, The abnormality main cause information indicates the main cause of the abnormality of the learning object substrate after being processed by the processing fluid, and includes at least one of the first abnormality main cause information and the second abnormality main cause information. The feature quantity includes feature quantity information, the feature quantity information represents a time-lapse feature of interval data in time series data, the time series data represents a physical quantity of an object used by a substrate processing device that processes the learning object substrate using the processing fluid, and the feature quantity information is represented by time. The first abnormality main cause information and the second abnormality main cause information are related to the feature quantity information. The input information includes information representing a time-lapse feature of interval data in time series data, the time series data representing a physical quantity of an object used by a substrate processing device that processes the processing target substrate using the processing fluid, the information being represented by time, The output information indicates a main cause of an abnormality of the substrate to be processed after being processed by the processing fluid, and includes at least one of first output information and second output information. The first output information and the second output information are related to the information showing the characteristics of the time lapse included in the input information, The first abnormal main cause information and the first output information each include at least one main cause information of first main cause information, second main cause information, third main cause information, fourth main cause information, fifth main cause information, sixth main cause information, seventh main cause information, eighth main cause information, and ninth main cause information, The first major cause information is information indicating the degree of poor adhesion of the treatment fluid. The second main cause information is information indicating the degree of scattering of the treatment fluid. The third main cause information is information showing the degree of instability of the flow rate of the process fluid, The fourth major cause information is information showing the degree of change in the concentration of the treatment fluid. The fifth major cause information is information showing the degree of instability of the flow rate of the mixed fluid when a plurality of the treatment fluids are mixed. The sixth main cause information is information showing the extent to which the substrate is not covered by the processing fluid, The seventh major cause information is information showing the amount of the treatment fluid that falls when the treatment fluid attached to the outer surface of the nozzle falls. The eighth major cause information is information showing the degree to which the processing fluid ejected first to the substrate is mixed with the processing fluid ejected later. The ninth major cause information is information showing the degree of mixing of other treatment fluids different from the treatment fluid to be recovered, The second abnormality main cause information and the second output information each include at least one main cause information of the tenth main cause information, the eleventh main cause information, the twelfth main cause information, and the thirteenth main cause information, The tenth major cause information is information showing the degree of change in the source pressure of the treatment fluid supplied to the nozzle. The eleventh major cause information is information indicating that the flow meter is malfunctioning. The twelfth major cause information is information indicating the degree of inappropriateness of the recipe information that specifies the processing content and the processing sequence. The thirteenth major cause information is parameter information indicating the setting value of the component of the substrate processing device, and information indicating the degree of setting error. said object being said process fluid, The physical quantity represents the flow rate of the treatment fluid, the temperature of the treatment fluid, or the concentration of the treatment fluid, and When the physical quantity is the flow rate of the processing fluid, the abnormality of the processing target substrate and the abnormality of the learning target substrate represent defects caused by particles. When the physical quantity is the temperature or concentration of the processing fluid, the abnormality of the processing target substrate and the abnormality of the learning target substrate indicate etching failure.

11. A learning method comprising the following steps: Access to learning data; and By performing machine learning on the learning data, a learning model is generated, and the learning model estimates the main cause of abnormality of the processing target substrate after being processed by the processing fluid. The learning data includes feature quantities and abnormal main cause information, In the step of generating the learned model, the machine learning is performed by using the feature quantity as an explanatory variable equivalent to the input of the learned model and using the abnormality main cause information as a target variable equivalent to the output of the learned model, The abnormality main cause information indicates the main cause of the abnormality of the learning object substrate after being processed by the processing fluid, and includes at least one of the first abnormality main cause information and the second abnormality main cause information. The feature quantity includes feature quantity information indicating a time-lapse feature of interval data in time series data, wherein the time series data indicates a physical quantity of an object used by a substrate processing device that processes the learning object substrate using the processing fluid, The feature information is represented by time, The first abnormality main cause information and the second abnormality main cause information are related to the feature quantity information. The learned model inputs input information and outputs output information. The input information includes first input information showing a time-lapse feature of interval data in time series data representing a physical quantity of an object used in a substrate processing apparatus that processes the processing target substrate with the processing fluid, The output information indicates a major cause of an abnormality of the substrate to be processed after being processed by the processing fluid, and includes at least one of first output information and second output information. The first output information and the second output information are related to the first input information, The first abnormal main cause information and the first output information each include at least one main cause information of first main cause information, second main cause information, third main cause information, fourth main cause information, fifth main cause information, sixth main cause information, seventh main cause information, eighth main cause information, and ninth main cause information, The first major cause information is information indicating the degree of poor adhesion of the treatment fluid. The second main cause information is information indicating the degree of scattering of the treatment fluid. The third main cause information is information showing the degree of instability of the flow rate of the process fluid, The fourth major cause information is information showing the degree of change in the concentration of the treatment fluid. The fifth major cause information is information showing the degree of instability of the flow rate of the mixed fluid when a plurality of the treatment fluids are mixed. The sixth main cause information is information showing the extent to which the substrate is not covered by the processing fluid, The seventh major cause information is information showing the amount of the treatment fluid that falls when the treatment fluid attached to the outer surface of the nozzle falls. The eighth major cause information is information showing the degree to which the processing fluid ejected first to the substrate is mixed with the processing fluid ejected later. The ninth major cause information is information showing the degree of mixing of other treatment fluids different from the treatment fluid to be recovered, The second abnormality main cause information and the second output information each include at least one main cause information of the tenth main cause information, the eleventh main cause information, the twelfth main cause information, and the thirteenth main cause information, The tenth major cause information is information showing the degree of change in the source pressure of the treatment fluid supplied to the nozzle. The eleventh major cause information is information indicating that the flow meter is malfunctioning. The twelfth major cause information is information indicating the degree of inappropriateness of the recipe information that specifies the processing content and the processing sequence. The thirteenth major cause information is parameter information indicating the setting value of the component of the substrate processing device, and information indicating the degree of setting error. said object being said process fluid, The physical quantity represents the flow rate of the treatment fluid, the temperature of the treatment fluid, or the concentration of the treatment fluid, and When the physical quantity is the flow rate of the processing fluid, the abnormality of the processing target substrate and the abnormality of the learning target substrate represent defects caused by particles. When the physical quantity is the temperature or concentration of the processing fluid, the abnormality of the processing target substrate and the abnormality of the learning target substrate indicate etching failure.

12. A learning device comprising: A storage unit for storing learning data; as well as A learning unit generates a learned model by performing machine learning on the learning data, wherein the learned model estimates a major cause of an abnormality of a substrate to be processed after being processed by a processing fluid, The learning data includes feature quantities and abnormal main cause information, The learning unit performs the machine learning by using the feature quantity as an explanatory variable corresponding to an input of the learned model and using the abnormality main cause information as a target variable corresponding to an output of the learned model. The abnormality main cause information indicates the main cause of the abnormality of the learning object substrate after being processed by the processing fluid, and includes at least one of the first abnormality main cause information and the second abnormality main cause information. The feature quantity includes feature quantity information indicating a time-lapse feature of interval data in time series data, wherein the time series data indicates a physical quantity of an object used by a substrate processing device that processes the learning object substrate using the processing fluid, The feature quantity information is represented by time, The first abnormality main cause information and the second abnormality main cause information are related to the feature quantity information. The learned model inputs input information and outputs output information. The input information includes first input information showing a time-lapse feature of interval data in time series data representing a physical quantity of an object used in a substrate processing apparatus that processes the processing target substrate with the processing fluid, The output information indicates a major cause of an abnormality of the substrate to be processed after being processed by the processing fluid, and includes at least one of first output information and second output information. The first output information and the second output information are related to the first input information, The first abnormal main cause information and the first output information each include at least one main cause information of first main cause information, second main cause information, third main cause information, fourth main cause information, fifth main cause information, sixth main cause information, seventh main cause information, eighth main cause information, and ninth main cause information, The first major cause information is information indicating the degree of poor adhesion of the treatment fluid. The second main cause information is information indicating the degree of scattering of the treatment fluid. The third main cause information is information showing the degree of instability of the flow rate of the process fluid, The fourth major cause information is information showing the degree of change in the concentration of the treatment fluid. The fifth major cause information is information showing the degree of instability of the flow rate of the mixed fluid when a plurality of the treatment fluids are mixed. The sixth main cause information is information showing the extent to which the substrate is not covered by the processing fluid, The seventh major cause information is information showing the amount of the treatment fluid that falls when the treatment fluid attached to the outer surface of the nozzle falls. The eighth major cause information is information showing the degree to which the processing fluid ejected first to the substrate is mixed with the processing fluid ejected later. The ninth major cause information is information showing the degree of mixing of other treatment fluids different from the treatment fluid to be recovered, The second abnormality main cause information and the second output information each include at least one main cause information of the tenth main cause information, the eleventh main cause information, the twelfth main cause information, and the thirteenth main cause information, The tenth major cause information is information showing the degree of change in the source pressure of the treatment fluid supplied to the nozzle. The eleventh major cause information is information indicating that the flow meter is malfunctioning. The twelfth major cause information is information indicating the degree of inappropriateness of the recipe information that specifies the processing content and the processing sequence. The thirteenth major cause information is parameter information indicating the setting value of the component of the substrate processing device, and information indicating the degree of setting error. said object being said process fluid, The physical quantity represents the flow rate of the treatment fluid, the temperature of the treatment fluid, or the concentration of the treatment fluid, and When the physical quantity is the flow rate of the processing fluid, the abnormality of the processing target substrate and the abnormality of the learning target substrate represent defects caused by particles. When the physical quantity is the temperature or concentration of the processing fluid, the abnormality of the processing target substrate and the abnormality of the learning target substrate indicate etching failure.

13. A method for preparing learning data, comprising the following steps: acquiring at least one interval data from time series data, the time series data representing a physical quantity of an object used by a substrate processing apparatus that processes a learning object substrate using a processing fluid; calculating feature quantity information indicating a feature of the time transition of the section data; and controlling the storage unit so as to store the abnormality main cause information in association with the feature amount information, The abnormality main cause information indicates the main cause of the abnormality of the learning object substrate after being processed by the processing fluid, and includes at least one of first abnormality main cause information and second abnormality main cause information. The first abnormality main cause information and the second abnormality main cause information are related to the feature amount information, The feature quantity information is represented by time, The feature quantity information and the abnormality main cause information constitute learning data as a subject of machine learning, wherein the machine learning is used to generate a learning model, wherein the learning model estimates the main cause of the abnormality of the processing target substrate after being processed by the processing fluid, The learned model is generated by executing the machine learning using the feature quantity information as an explanatory variable equivalent to the input of the learned model and the abnormality main cause information as a target variable equivalent to the output of the learned model, The learned model inputs input information and outputs output information. The input information includes first input information showing a time-lapse feature of interval data in time series data representing a physical quantity of an object used in a substrate processing apparatus that processes the processing target substrate with the processing fluid, The output information shows a main cause of an abnormality of the substrate to be processed after being processed by the processing fluid, and includes at least one of first output information and second output information. The first output information and the second output information are related to the first input information, The first abnormal main cause information and the first output information each include at least one main cause information of first main cause information, second main cause information, third main cause information, fourth main cause information, fifth main cause information, sixth main cause information, seventh main cause information, eighth main cause information, and ninth main cause information, The first major cause information is information indicating the degree of poor adhesion of the treatment fluid. The second main cause information is information indicating the degree of scattering of the treatment fluid. The third main cause information is information showing the degree of instability of the flow rate of the process fluid, The fourth major cause information is information showing the degree of change in the concentration of the treatment fluid. The fifth major cause information is information showing the degree of instability of the flow rate of the mixed fluid when a plurality of the treatment fluids are mixed. The sixth main cause information is information showing the extent to which the substrate is not covered by the processing fluid, The seventh major cause information is information showing the amount of the treatment fluid that falls when the treatment fluid attached to the outer surface of the nozzle falls. The eighth major cause information is information showing the degree to which the processing fluid ejected first to the substrate is mixed with the processing fluid ejected later. The ninth major cause information is information showing the degree of mixing of other treatment fluids different from the treatment fluid to be recovered, The second abnormality main cause information and the second output information each include at least one main cause information of the tenth main cause information, the eleventh main cause information, the twelfth main cause information, and the thirteenth main cause information, The tenth major cause information is information showing the degree of change in the source pressure of the treatment fluid supplied to the nozzle. The eleventh major cause information is information indicating that the flow meter is malfunctioning. The twelfth major cause information is information indicating the degree of inappropriateness of the recipe information that specifies the processing content and the processing sequence. The thirteenth major cause information is parameter information indicating the setting value of the component of the substrate processing device, and information indicating the degree of setting error. said object being said process fluid, The physical quantity represents the flow rate of the treatment fluid, the temperature of the treatment fluid, or the concentration of the treatment fluid, and When the physical quantity is the flow rate of the processing fluid, the abnormality of the processing target substrate and the abnormality of the learning target substrate represent defects caused by particles. When the physical quantity is the temperature or concentration of the processing fluid, the abnormality of the processing target substrate and the abnormality of the learning target substrate indicate etching failure.