Method for substrate processing, method for generating trained model, method for acquiring nozzle speed, method for creating learning data, substrate processing apparatus, trained model, nozzle speed acquiring apparatus, and computer program

Abstract

To provide a method for substrate processing capable of acquiring target nozzle speed information to accurately achieve a target processing amount.SOLUTION: The method for substrate processing includes: a step S5 of inputting target processing amount information to a first trained model; a step S6 of acquiring predicted residence time information output by the first trained model; a step S7 of acquiring target nozzle speed information by processing the predicted residence time information; and a step S8 of processing a substrate to be processed in accordance with the target nozzle speed information, the target nozzle speed information indicating target values of moving speeds of a processing nozzle set for respective processing nozzle positions, the processing nozzle being moved while discharging a processing liquid onto the substrate to be processed and having a moving range divided into a plurality of moving sections; the target processing amount information indicating target values of processing amounts of the substrate to be processed by the processing liquid at a plurality of positions in a radial direction of the substrate to be processed; and the predicted residence time information indicating predicted values of residence times of the processing nozzle at the plurality of moving sections.SELECTED DRAWING: Figure 13

Description

[Technical Field]

[0001] The present invention relates to a substrate processing method, a trained model generating method, a nozzle speed acquiring method, a training data creating method, a substrate processing apparatus, a trained model, a nozzle speed acquiring device, and a computer program. [Background technology]

[0002] The substrate processing apparatus described in Patent Document 1 includes a nozzle, a movement mechanism, a memory unit, and a control unit. The movement mechanism moves the nozzle. The memory unit stores a trained model. The trained model is generated by learning, as learning data, learning target speed information indicating the movement speed of the nozzle and a processing amount obtained by performing processing on a training target substrate while moving the nozzle at a speed based on the learning target speed information. The control unit inputs a target amount of processing amount into the trained model, thereby outputting processing speed information from the trained model. The control unit controls the movement mechanism so that the nozzle moves at a speed based on the processing speed information when performing processing on the training target substrate. The processing speed information indicates the movement speed of the nozzle. [Prior art documents] [Patent documents]

[0003] [Patent Document 1] Patent Publication No. 2021-108367 Summary of the Invention [Problem to be solved by the invention]

[0004] However, in the substrate processing apparatus described in Patent Document 1, it is required to acquire processing speed information (target nozzle speed information) that can achieve a target processing amount (target processing amount) with higher accuracy.

[0005] The present invention has been made in consideration of the above-mentioned problems, and its purpose is to provide a substrate processing method, a trained model generation method, a nozzle speed acquisition method, a training data creation method, a substrate processing apparatus, a trained model, a nozzle speed acquisition device, and a computer program that can acquire target nozzle speed information that can accurately achieve a target processing volume. [Means for solving the problem]

[0006] According to one aspect of the present invention, a substrate processing method includes the steps of inputting target processing volume information to a first trained model constructed by machine learning first learning data, acquiring predicted dwell time information output by the first trained model, acquiring target nozzle speed information by processing the predicted dwell time information, and processing a target substrate in accordance with the target nozzle speed information. The target nozzle speed information indicates target values ​​of the movement speed of the processing nozzle set for each of multiple processing nozzle positions that divide the movement range of the processing nozzle, which moves while discharging processing liquid onto the target substrate, into multiple movement sections. The target processing volume information indicates target values ​​of the processing volume of the target substrate with the processing liquid at each of multiple positions in the radial direction of the target substrate. The predicted dwell time information indicates a predicted value of the dwell time of the processing nozzle in each of the multiple movement sections. The first learning data includes first learning information and second learning information. The first learning information is an explanatory variable indicating the processing volume of the substrate with the processing liquid at each of multiple positions in the radial direction of the substrate. The second learning information is a response variable that indicates the residence time of the nozzle in each of a plurality of movement sections that divide the movement range of the nozzle that moves while discharging the processing liquid onto the substrate.

[0007] According to another aspect of the present invention, a trained model generation method includes the steps of acquiring first training data including first training information and second training information, and generating a first trained model that inputs target processing amount information and outputs predicted residence time information by machine learning the first training data. The first training information is an explanatory variable indicating the processing amount of the substrate by the processing liquid at each of multiple positions in the radial direction of the substrate. The second training information is a target variable indicating the residence time of the nozzle in each of multiple movement sections that divide the movement range of the nozzle as it moves while discharging the processing liquid onto the substrate. The target processing amount information indicates a target value of the processing amount of the substrate to be processed by the processing liquid at each of multiple positions in the radial direction of the substrate. The predicted residence time information indicates a predicted value of the residence time of the processing nozzle in each of multiple movement sections that divide the movement range of the processing nozzle as it moves while discharging the processing liquid onto the substrate.

[0008] In one aspect of the present invention, in the trained model generation method, the first learning information is preferably corrected information based on a comparison result between first reference processing amount information indicating a processing result of a first reference substrate and second reference processing amount information indicating a processing result of a second reference substrate processed at a different timing than the first reference substrate. The first reference processing amount information preferably indicates a processing amount of the first reference substrate by the processing liquid at each of a plurality of positions in the radial direction of the first reference substrate when the first reference substrate is processed with the processing liquid under reference processing conditions. The second reference processing amount information preferably indicates a processing amount of the second reference substrate by the processing liquid at each of a plurality of positions in the radial direction of the second reference substrate when the second reference substrate is processed with the processing liquid under the reference processing conditions.

[0009] In one aspect of the present invention, in the trained model generation method, it is preferable that the second reference processing amount information is information generated on the same day as original training information, which is information before the first training information is corrected, is generated. It is preferable that the first reference processing amount information is information generated earlier than the day the second reference processing amount information is acquired.

[0010] In one aspect of the present invention, in the trained model generation method, the first training data preferably includes third training information, which preferably includes at least one of temperature information indicating a temperature of the processing liquid used to process the substrate, concentration information indicating a concentration of the processing liquid used to process the substrate, flow rate information indicating a flow rate of the processing liquid used to process the substrate, and rotation speed information indicating a rotation speed of the substrate.

[0011] According to yet another aspect of the present invention, a nozzle speed acquisition method includes a step of acquiring predicted dwell time information and a step of acquiring target nozzle speed information by processing the predicted dwell time information. The target nozzle speed information indicates target values ​​for movement speed of the processing nozzle, which are set for each of a plurality of processing nozzle positions that divide a movement range of the processing nozzle, which moves while discharging processing liquid onto a substrate, into a plurality of movement sections. The predicted dwell time information indicates a predicted value of the dwell time of the processing nozzle in each of the plurality of movement sections.

[0012] In one aspect of the present invention, in the nozzle velocity acquisition method, when the processing nozzle ejects the processing liquid while reciprocating, the dwell time preferably indicates a cumulative total of time that the processing nozzle is positioned in each of the plurality of movement sections. In the step of acquiring the target nozzle velocity information, the target nozzle velocity information is preferably calculated based on the number of reciprocating movements of the processing nozzle, the predicted value of the dwell time, and the length of the movement section in the radial direction of the substrate to be processed.

[0013] In one aspect of the present invention, in the nozzle speed acquisition method, the step of acquiring the target nozzle speed information preferably includes inputting the predicted dwell time information into a second trained model constructed by machine learning second training data, and acquiring the target nozzle speed information from the second trained model. The second training data preferably includes fourth training information and fifth training information. The fourth training information is preferably an explanatory variable indicating the dwell time of the nozzle in each of a plurality of movement sections that divide the movement range of the nozzle as it moves while ejecting the processing liquid onto the substrate. The fifth training information is preferably a target variable indicating the movement speed of the nozzle set for each of a plurality of nozzle positions that divide the movement range into the plurality of movement sections.

[0014] According to yet another aspect of the present invention, a training data creation method creates first training data that is the subject of machine learning to generate a first trained model that inputs target processing volume information and outputs predicted residence time information. The training data creation method includes the steps of: acquiring original training information; correcting the original training information based on a comparison result between first reference processing volume information indicating a processing result of a first reference substrate and second reference processing volume information indicating a processing result of a second reference substrate processed at a different timing than the first reference substrate; and outputting the corrected original training information as first training information; and acquiring second training information. The target processing volume information indicates a target value of the processing volume of the target substrate with a processing liquid at each of multiple positions in the radial direction of the target substrate. The predicted residence time information indicates a predicted value of the residence time of the processing nozzle in each of multiple movement sections that divide the movement range of the processing nozzle as it moves while discharging the processing liquid onto the target substrate. The original training information indicates the processing volume of the substrate with the processing liquid at each of multiple positions in the radial direction of the substrate. The first training information is an explanatory variable. The second learning information is a response variable indicating the dwell time of the nozzle in each of a plurality of movement sections that divide the movement range of the nozzle as it moves while ejecting the processing liquid onto the substrate. The first learning information and the second learning information constitute the first learning data. The first reference processing amount information indicates the processing amount of the first reference substrate by the processing liquid at each of a plurality of positions in the radial direction of the first reference substrate when the first reference substrate is processed with the processing liquid according to the reference processing conditions. The second reference processing amount information indicates the processing amount of the second reference substrate by the processing liquid at each of a plurality of positions in the radial direction of the second reference substrate when the second reference substrate is processed with the processing liquid according to the reference processing conditions.

[0015] According to yet another aspect of the present invention, a substrate processing apparatus includes a memory unit, a dwell time acquisition unit, a speed acquisition unit, and a substrate processing unit. The memory unit stores a first trained model constructed by machine learning first learning data. The dwell time acquisition unit inputs target processing volume information to the first trained model and acquires predicted dwell time information from the first trained model. The speed acquisition unit processes the predicted dwell time information to acquire target nozzle speed information. The substrate processing unit processes a substrate to be processed in accordance with the target nozzle speed information. The target nozzle speed information indicates target values ​​of movement speeds of the processing nozzle, which are set for multiple processing nozzle positions that divide the movement range of the processing nozzle, which moves while discharging processing liquid onto the substrate to be processed, into multiple movement sections. The target processing volume information indicates target values ​​of the processing volume of the substrate to be processed by the processing liquid at each of multiple positions in the radial direction of the substrate to be processed. The predicted dwell time information indicates predicted values ​​of the dwell time of the processing nozzle in each of the multiple movement sections. The first learning data includes first learning information and second learning information. The first learning information is an explanatory variable that indicates the amount of treatment of the substrate by the treatment liquid at each of a plurality of positions in the radial direction of the substrate, and the second learning information is a target variable that indicates the residence time of the nozzle in each of a plurality of movement sections that divide the movement range of the nozzle that moves while discharging the treatment liquid onto the substrate.

[0016] According to yet another aspect of the present invention, the trained model is constructed by machine learning first training data, and causes a computer to function to input target processing volume information and output predicted residence time information. The target processing volume information indicates a target value of the processing volume of the substrate to be processed by the processing liquid at each of a plurality of positions in the radial direction of the substrate. The predicted residence time information indicates a predicted value of the residence time of the processing nozzle in each of a plurality of movement sections that divide the movement range of the processing nozzle as it moves while discharging the processing liquid onto the substrate. The first training data includes first training information and second training information. The first training information is an explanatory variable that indicates the processing volume of the substrate to be processed by the processing liquid at each of a plurality of positions in the radial direction of the substrate. The second training information is a target variable that indicates the residence time of the nozzle in each of a plurality of movement sections that divide the movement range of the nozzle as it moves while discharging the processing liquid onto the substrate.

[0017] According to yet another aspect of the present invention, a nozzle velocity acquisition device includes a dwell time acquisition unit and a velocity acquisition unit. The dwell time acquisition unit acquires predicted dwell time information. The velocity acquisition unit processes the predicted dwell time information to acquire target nozzle velocity information. The target nozzle velocity information indicates target values ​​for the movement velocity of the processing nozzle, which are set for each of a plurality of processing nozzle positions that divide the movement range of the processing nozzle, which moves while discharging processing liquid onto a substrate to be processed, into a plurality of movement sections. The predicted dwell time information indicates a predicted value for the dwell time of the processing nozzle in each of the plurality of movement sections.

[0018] According to yet another aspect of the present invention, a computer program causes a computer to execute the above-described trained model generation method.

[0019] According to yet another aspect of the present invention, a computer program causes a computer to execute the nozzle velocity acquisition method. [Effects of the Invention]

[0020] According to the present invention, it is possible to acquire target nozzle speed information that enables a target throughput to be achieved with high accuracy. [Brief explanation of the drawings]

[0021] [Figure 1] 1 is a schematic cross-sectional view showing a substrate processing apparatus according to an embodiment of the present invention. [Figure 2] 10 is a plan view showing a substrate scanning process by a nozzle of the substrate processing apparatus according to the embodiment. FIG. [Figure 3] 10 is a plan view showing a substrate scanning process by an optical probe of the substrate processing apparatus according to the present embodiment. FIG. [Figure 4] FIG. 2 is a block diagram showing a control device of the substrate processing apparatus according to the present embodiment. [Figure 5] 10 is a diagram showing the thickness of an object constituting a substrate in the substrate processing apparatus according to the present embodiment. FIG. [Figure 6] 5A and 5B are diagrams showing operation patterns of nozzles of the substrate processing apparatus according to the present embodiment. [Figure 7] 3 is a diagram showing divided regions of a substrate in the substrate processing apparatus according to the present embodiment; FIG. [Figure 8] FIG. 10 is a diagram showing the dwell time of a nozzle in the substrate processing apparatus according to the present embodiment. [Figure 9] FIG. 10 is a diagram showing the operation of a first trained model of the substrate processing apparatus according to the present embodiment. [Figure 10] 10A and 10B are diagrams illustrating an operation of a speed acquisition unit of the substrate processing apparatus according to the present embodiment. [Figure 11] 10 is a diagram showing the relationship between the nozzle position and the target value of the nozzle movement speed in the substrate processing apparatus according to the present embodiment. FIG. [Figure 12] 10A and 10B are diagrams for explaining a speed derivation algorithm executed by a speed acquisition unit of the substrate processing apparatus according to the present embodiment. [Figure 13] 3 is a flowchart showing a substrate processing method executed by the substrate processing apparatus according to the present embodiment. [Figure 14] 1 is a block diagram showing a learning device according to an embodiment of the present invention; [Figure 15] FIG. 2 is a diagram illustrating the operation of the learning device according to the present embodiment. [Figure 16] 10 is a flowchart showing a learning method executed by the learning device according to the present embodiment. [Figure 17] 1 is a block diagram showing a learning data creation device according to an embodiment of the present invention; [Figure 18] (a) is a diagram showing an example of original learning information according to this embodiment; (b) is a diagram showing an example of second reference processing amount information according to this embodiment; (c) is a diagram showing an example of first reference processing amount information according to this embodiment; (d) is a diagram showing an example of comparison information according to this embodiment; (e) is a diagram showing an example of correction values ​​according to this embodiment; and (f) is a diagram showing an example of first learning information according to this embodiment. [Figure 19] 1 is a flowchart showing a learning data creation method executed by the learning data creation device according to the present embodiment. [Figure 20] FIG. 20(a) is a diagram showing an example of target processing volume information before correction according to a first modified example of this embodiment. FIG. 20(b) is a diagram showing an example of second reference processing volume information according to the first modified example. FIG. 20(c) is a diagram showing an example of first reference processing volume information according to the first modified example. FIG. 20(d) is a diagram showing an example of comparison information according to the first modified example. FIG. 20(e) is a diagram showing an example of correction values ​​according to the first modified example. FIG. 20(f) is a diagram showing an example of target processing volume information after correction according to the first modified example. [Figure 21] 10 is a flowchart showing a correction process for target processing amount information according to a first modified example of the present embodiment. [Figure 22] FIG. 10 is a diagram illustrating the operation of a second trained model according to a second modified example of the present embodiment. [Figure 23] 10 is a graph showing the nozzle residence time and the substrate throughput according to an embodiment of the present invention. DETAILED DESCRIPTION OF THE INVENTION

[0022] Hereinafter, embodiments of the present invention will be described with reference to the drawings. In the drawings, the same or corresponding parts are designated by the same reference numerals and description thereof will not be repeated. In the embodiments of the present invention, the X-axis, Y-axis, and Z-axis are mutually orthogonal, the X-axis and Y-axis are parallel to the horizontal direction, and the Z-axis is parallel to the vertical direction. Note that "planar view" refers to viewing an object from vertically above.

[0023] A substrate processing apparatus 100 according to an embodiment of the present invention will be described with reference to Figures 1 to 13. Figure 1 is a diagram showing the substrate processing apparatus 100. The substrate processing apparatus 100 shown in Figure 1 processes a substrate W with a processing liquid. The substrate processing apparatus 100 typically etches the substrate W with the processing liquid. In this case, the processing liquid is an etching liquid.

[0024] For example, the processing liquid is a chemical liquid, such as dilute hydrofluoric acid (DHF), hydrofluoric acid (HF), hydrofluoric nitric acid (a mixture of hydrofluoric acid and nitric acid (HNO3)), buffered hydrofluoric acid (BHF), ammonium fluoride, HFEG (a mixture of hydrofluoric acid and ethylene glycol), phosphoric acid (H3PO4), sulfuric acid, acetic acid, nitric acid, hydrochloric acid, ammonia water, hydrogen peroxide water, organic acid (e.g., citric acid, oxalic acid), organic alkali (e.g., TMAH: tetramethylammonium hydroxide), sulfuric acid hydrogen peroxide water mixture (SPM), ammonia hydrogen peroxide water mixture (SC1), or hydrochloric acid hydrogen peroxide water mixture (SC2).

[0025] Specifically, the substrate processing apparatus 100 processes an object constituting a substrate W with a processing liquid. Hereinafter, an object to be processed with a processing liquid will be referred to as an "object TG." The object TG constituting the substrate W is, for example, a substrate body (e.g., a substrate body made of silicon) or a film formed on the surface of the substrate body. The film formed on the surface of the substrate body is, for example, a film made of the same material as the substrate body (e.g., a film made of silicon) or a film made of a different material from the substrate body (e.g., a silicon oxide film, a silicon nitride film, or a resist). Furthermore, in this specification, processing a substrate W refers to processing the object TG. Therefore, the processing amount of the substrate W refers to the processing amount of the object TG.

[0026] The substrate processing apparatus 100 is a single-wafer processing apparatus that processes a single substrate W. The substrate W has a substantially circular disk shape.

[0027] 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 a magneto-optical disk, a substrate for a photomask, a ceramic substrate, or a substrate for a solar cell. In the following description of the embodiment, the substrate W is a semiconductor wafer.

[0028] As shown in FIG. 1, the substrate processing apparatus 100 includes a substrate processing section 1 and a control device .

[0029] The control device 21 controls the substrate processing unit 1. The substrate processing unit 1 processes the substrate W by discharging a processing liquid onto the substrate W. Specifically, the substrate processing unit 1 includes a chamber 2, a spin chuck 3, a spin motor 5, a nozzle 7, a first moving unit 9, a guard 13, a thickness measuring unit 15, a second moving unit 17, a valve V1, and a supply pipe K1.

[0030] The chamber 2 has a substantially box shape and accommodates the substrate W, the spin chuck 3, the spin motor 5, the nozzle 7, the first moving part 9, the guard 13, the thickness measuring part 15, the second moving part 17, and part of the supply pipe K1.

[0031] The spin chuck 3 holds and rotates the substrate W. Specifically, the spin chuck 3 rotates the substrate W about a rotation axis AX1 while holding the substrate W horizontally in the chamber 2. In this case, the spin chuck 3 is driven to rotate by a spin motor 5.

[0032] The spin chuck 3 includes a plurality of chuck members 32 and a spin base 33. The plurality of chuck members 32 are provided on the spin base 33 along the periphery of the substrate W. The plurality of chuck members 32 hold the substrate W in a horizontal position. The spin base 33 is substantially disk-shaped and supports the plurality of chuck members 32 in a horizontal position. The spin motor 5 rotates the spin base 33 about a rotation axis AX1. Therefore, the spin base 33 rotates about the rotation axis AX1. As a result, the substrate W held by the plurality of chuck members 32 provided on the spin base 33 rotates about the rotation axis AX1. Specifically, the spin motor 5 includes a motor main body 51 and a shaft 53. The shaft 53 is coupled to the spin base 33. The motor main body 51 rotates the shaft 53, thereby rotating the spin base 33.

[0033] The nozzle 7 discharges the processing liquid toward the substrate W while the substrate W is rotating. A supply pipe K1 supplies the processing liquid to the nozzle 7. A valve V1 switches between starting and stopping the supply of the processing liquid to the nozzle 7.

[0034] The first moving unit 9 moves the nozzle 7 in the substantially vertical and horizontal directions. Specifically, the first moving unit 9 includes an arm 91, a rotating shaft 93, and a first moving mechanism 95. The arm 91 extends substantially horizontally. The nozzle 7 is disposed at the tip of the arm 91. The arm 91 is coupled to the rotating shaft 93. The rotating shaft 93 extends substantially vertically. The first moving mechanism 95 rotates the rotating shaft 93 about a rotation axis AX2 that extends substantially vertically, causing the arm 91 to rotate along a substantially horizontal plane. As a result, the nozzle 7 moves along the substantially horizontal plane. In other words, the nozzle 7 rotates horizontally around the rotation axis AX2. Furthermore, the first moving mechanism 95 raises and lowers the rotating shaft 93 in the substantially vertical direction, raising and lowering the arm 91. As a result, the nozzle 7 moves along the substantially vertical direction. The first movement mechanism 95 includes, for example, a ball screw mechanism and an electric motor that provides a driving force to the ball screw mechanism.

[0035] The thickness measurement unit 15 measures the thickness of the object TG in a non-contact manner and outputs information indicating the thickness of the object TG to the control device 21. The thickness measurement unit 15 measures the thickness of the object TG, for example, by spectral interferometry. Specifically, the thickness measurement unit 15 includes an optical probe 151, a connecting line 153, and a thickness measurement device 155. The optical probe 151 includes a lens. The connecting line 153 connects the optical probe 151 and the thickness measurement device 155. The connecting line 153 includes an optical fiber. The thickness measurement device 155 includes a light source and a light receiving element. Light emitted from the light source of the thickness measurement device 155 is emitted to the object TG via the connecting line 153 and the optical probe 151. The light reflected by the object TG is received by the light receiving element of the thickness measurement device 155 via the optical probe 151 and the connecting line 153. The thickness measuring device 155 analyzes the received light and calculates the thickness of the object TG based on the analysis result. The thickness measuring device 155 outputs information indicating the thickness of the object TG to the control device 21.

[0036] The second moving unit 17 moves the optical probe 151 in the substantially vertical direction and the substantially horizontal direction. Specifically, the second moving unit 17 includes an arm 171, a rotating shaft 173, and a second moving mechanism 175. The arm 171 extends in the substantially horizontal direction. The optical probe 151 is disposed at the tip of the arm 171. The arm 171 is coupled to the rotating shaft 173. The rotating shaft 173 extends in the substantially vertical direction. The second moving mechanism 175 rotates the rotating shaft 173 about a rotation axis AX3 that extends in the substantially vertical direction, thereby rotating the arm 171 along a substantially horizontal plane. As a result, the optical probe 151 moves along the substantially horizontal plane. In other words, the optical probe 151 rotates horizontally around the rotation axis AX3. Furthermore, the second moving mechanism 175 raises and lowers the rotating shaft 173 in the substantially vertical direction, thereby raising and lowering the arm 171. As a result, the optical probe 151 moves substantially vertically. The second movement mechanism 175 includes, for example, a ball screw mechanism and an electric motor that provides a driving force to the ball screw mechanism.

[0037] Here, an example of how to use the substrate processing apparatus 100 will be described. For example, the "processing conditions" and "processing results" in the substrate processing apparatus 100 can be used as learning data for machine learning. In this case, for example, the processing amount 77 and the residence time sn of the nozzle 7 in the substrate processing apparatus 100 are used as first learning data TND (FIG. 15).

[0038] For example, the substrate processing apparatus 100 can process the substrate W according to "processing conditions" based on a trained model obtained by machine learning the training data. In this case, the substrate W is processed according to target nozzle speed information RP1 based on the first trained model LM1 obtained by machine learning the first training data TND (FIG. 4). The substrate W in this case corresponds to an example of a "substrate to be processed" in the present invention. Furthermore, the nozzle 7 in this case corresponds to an example of a "processing nozzle" in the present invention.

[0039] Next, the scanning process of the substrate W by the nozzle 7 will be described with reference to Fig. 2. Fig. 2 is a plan view showing the scanning process of the substrate W by the nozzle 7. As shown in Fig. 2, the scanning process by the nozzle 7 is a process of discharging the processing liquid onto the substrate W while moving the nozzle 7 so that the landing position of the processing liquid on the surface of the substrate W forms an arc-shaped trajectory TJ1 in a plan view. The scanning process of the substrate W by the nozzle 7 is performed while the substrate W is rotating. The nozzle 7 moves in an area above the substrate W held by the spin chuck 3.

[0040] The trajectory TJ1 along which the nozzle 7 moves passes through the substrate center OP, which is the center of the substrate W. One end of the trajectory TJ1 is indicated by an operating end EP1 inside the peripheral edge EG of the substrate W. The other end of the trajectory TJ1 is indicated by an operating end EP2 inside the peripheral edge EG of the substrate W. The scan of the nozzle 7 moving from the operating end EP1 to the substrate center OP is indicated by arrow a1. The scan of the nozzle 7 moving from the substrate center OP to the operating end EP2 is indicated by arrow a2. The scan of the nozzle 7 moving from the operating end EP2 to the substrate center OP is indicated by arrow a3. The scan of the nozzle 7 moving from the substrate center OP to the operating end EP1 is indicated by arrow a4.

[0041] In this embodiment, the start position of the scanning process by the nozzle 7 is the operating end EP1 or the operating end EP2. Note that the start position of the scanning process is not limited to the operating end EP1 or EP2, and may be, for example, above the center OP of the substrate.

[0042] Next, the scanning process of the substrate W by the optical probe 151 will be described with reference to Fig. 3. Fig. 3 is a plan view showing the scanning process of the substrate W by the optical probe 151. As shown in Fig. 3, the scanning process by the optical probe 151 is a process of measuring the thickness of the object TG while moving the optical probe 151 so that the thickness measurement position with respect to the object TG forms an arc-shaped locus TJ2 in a plan view. The scanning process of the substrate W by the optical probe 151 is performed while the substrate W is rotating. The optical probe 151 moves in an area above the substrate W held by the spin chuck 3.

[0043] Specifically, the optical probe 151 moves between the substrate center OP and the peripheral edge EG of the substrate W in a plan view, while moving to a different measurement position. In other words, the thickness measurement unit 15 measures the thickness of the object TG at each of a plurality of measurement positions on the object TG. As a result, the thickness distribution of the object TG is measured from the substrate center OP to the peripheral edge EG. In other words, the thickness distribution of the object TG in the radial direction of the substrate W is measured.

[0044] Next, the control device 21 of the substrate processing apparatus 100 will be described with reference to Fig. 4. Fig. 4 is a block diagram showing the control device 21. The control device 21 is, for example, a computer. As shown in Fig. 4, the control device 21 includes a control unit 21A, a storage unit 21B, a communication unit 21C, an input unit 21D, and a display unit 21E.

[0045] The control unit 21A includes processors such as a CPU (Central Processing Unit) and a GPU (Graphics Processing Unit). The storage unit 21B includes a storage device and stores data and computer programs. The processor of the control unit 21A executes the computer programs stored in the storage device of the storage unit 21B to control the substrate processing unit 1.

[0046] For example, the storage unit 21B 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 21B may also include a removable medium such as an optical disk. The storage unit 21B is, for example, a non-transitory computer-readable storage medium.

[0047] Specifically, the memory unit 21B stores a control program PG1, a first learned model LM1, and a recipe RP. The recipe RP indicates the processing content and processing procedure for the substrate W. The recipe RP includes target nozzle speed information RP1 and processing condition information RP2. The memory unit 21B may also store a reference recipe R1. The reference recipe R1 includes reference nozzle speed information R11 and processing condition information R12. The reference recipe R1 will be described later in a modified example.

[0048] The communication unit 21C is connected to a network and communicates with external devices. In this embodiment, the network includes, for example, the Internet, a local area network (LAN), a public telephone network, and a short-range wireless network. The communication unit 21C is a communication device, for example, a network interface controller.

[0049] The input unit 21D is an input device for inputting various types of information to the control unit 21A. For example, the input unit 21D is a keyboard and pointing device, or a touch panel.

[0050] The display unit 21E displays an image and is, for example, a liquid crystal display or an organic electroluminescence display.

[0051] The substrate processing apparatus 100 may also include a temperature sensor 21F, a concentration sensor 21G, a flow rate sensor 21H, and a rotation sensor 21K.

[0052] The temperature sensor 21F detects the temperature of the treatment liquid and outputs temperature information indicating the temperature. The temperature sensor 21F is disposed, for example, in the vicinity of the nozzle 7 on the supply pipe K1. The concentration sensor 21G directly or indirectly detects the concentration of the treatment liquid and outputs concentration information indicating the concentration. The concentration may be indicated by the specific gravity of the treatment liquid. In this case, the concentration sensor 21G detects the specific gravity of the treatment liquid. The concentration sensor 21G is disposed, for example, in a tank that stores the treatment liquid.

[0053] The flow rate sensor 21H detects the flow rate of the processing liquid and outputs flow rate information indicating the flow rate. The flow rate sensor 21H is disposed, for example, near the nozzle 7 on the supply pipe K1. The rotation sensor 21K detects the rotation speed of the substrate W and outputs rotation speed information indicating the rotation speed. The rotation speed indicates, for example, the number of rotations of the substrate W per unit time. Specifically, the rotation sensor 21K detects the rotation speed of the spin chuck 3. In this case, the rotation sensor 21K detects the rotation speed of the spin chuck 3 optically, electrically, or mechanically. The rotation speed of the spin chuck 3 indicates the rotation speed of the substrate W. The rotation sensor 21K may also detect the rotation speed of the spin motor 5.

[0054] The control unit 21A includes a measurement control unit 210, a target throughput calculation unit 211, a residence time acquisition unit 212, a speed acquisition unit 213, and a substrate processing control unit 214. The control unit 21A executes a control program PG1 to function as the measurement control unit 210, the target throughput calculation unit 211, the residence time acquisition unit 212, the speed acquisition unit 213, and the substrate processing control unit 214. The control unit 21A may include a correction unit 215. In this case, the control unit 21A executes the control program PG1 to function as the correction unit 215. The correction unit 215 will be described later in a modified example.

[0055] The measurement control unit 210 controls the thickness measurement unit 15 and the second movement unit 17 (FIG. 1) to measure the thickness of the target object TG constituting the substrate W before and after processing by the substrate processing unit 1. As a result, the thickness measurement unit 15 performs a scan process using the optical probe 151, thereby measuring the thickness of the target object TG at a plurality of measurement positions in the radial direction of the substrate W before and after processing.

[0056] The thickness measuring unit 15 outputs information indicating the thickness of the object TG at each of a plurality of measurement positions in the radial direction of the substrate W before processing with the processing liquid (hereinafter, this may be referred to as "pre-processing measured thickness information") to the control device 21. The memory unit 21B then stores the pre-processing measured thickness information. The thickness measuring unit 15 also outputs information indicating the thickness of the object TG at each of a plurality of measurement positions in the radial direction of the substrate W after processing with the processing liquid (hereinafter, this may be referred to as "post-processing measured thickness information") to the control device 21. The memory unit 21B then stores the post-processing measured thickness information.

[0057] FIG. 5 is a diagram showing the thickness of the object TG that constitutes the substrate W. The object TG is, for example, a film formed on the surface of the substrate body. Referring to FIG. 5, the horizontal axis indicates the radial position of the substrate W, and the vertical axis indicates the thickness of the object TG. The origin of the horizontal axis indicates the center of the substrate W. The thickness of the object TG that constitutes the substrate W before being processed by the substrate processing unit 1 is indicated by a solid line 18. The substrate processing unit 1 processes the substrate W with a processing liquid, thereby adjusting the thickness (e.g., film thickness) of the object TG. The thickness of the object TG that constitutes the substrate W after being processed by the substrate processing unit 1 is indicated by a dotted line 19.

[0058] The difference between the thickness of the target object TG before being processed by the substrate processing unit 1 and the thickness of the target object TG after being processed by the substrate processing unit 1 is the processing amount 19a (for example, the etching amount).

[0059] Furthermore, it is preferable that the thickness of the target object TG constituting the substrate W is uniform over the entire surface of the substrate W. Therefore, a target thickness is set for the processing performed by the substrate processing unit 1. The target thickness is indicated by a dashed dotted line 20. Information indicating the target thickness of the target object TG is stored in the memory unit 21B (FIG. 4). Hereinafter, information indicating the target thickness may be referred to as "target thickness information."

[0060] The difference between the thickness of the target object TG before being processed by the substrate processing unit 1 and the target thickness is the target value 20a of the processing amount.

[0061] 4, the target processing amount calculation unit 211 acquires target thickness information from the storage unit 21B. The target processing amount calculation unit 211 also acquires pre-processing measured thickness information from the storage unit 21B. The target processing amount calculation unit 211 then calculates target processing amount information based on the target thickness information and the pre-processing measured thickness information. The target processing amount information indicates target values ​​20a of the processing amount of the substrate W (specifically, the target object TG) by the processing liquid at each of a plurality of positions in the radial direction of the substrate W.

[0062] Specifically, the target processing amount calculation unit 211 calculates the difference between the thickness of the target object TG constituting the substrate W before processing and the target thickness of the target object TG for each radial position on the substrate W. The difference indicates the target value 20a of the processing amount.

[0063] The residence time acquisition unit 212 inputs target processing volume information to the first trained model LM1 and acquires predicted residence time information from the first trained model LM1. The predicted residence time information indicates a predicted value of the residence time of the nozzle 7 in each of a plurality of movement sections that divide the movement range of the nozzle 7 as it moves while discharging the processing liquid onto the substrate W.

[0064] Next, the dwell time of the nozzle 7 will be described with reference to FIGS. 6 to 8. FIG. 6 is a diagram showing the operation pattern of the nozzle 7. In FIG. 6, the vertical axis represents the relative position of the nozzle 7 with respect to the substrate W, and the horizontal axis represents elapsed time (seconds). In this embodiment, the scan period from the start to the end of the operation of moving the nozzle 7 with respect to the substrate W is equal to the processing time. As an example, if the processing time is set to 60 seconds, the operation pattern of the nozzle 7 shows the relative position for the period from 0 seconds to 60 seconds. The relative position of the nozzle 7 is represented by a negative value at the substrate center OP, and a positive value at the range from the substrate center OP to the operation end EP1, and a positive value at the range from the substrate center OP to the operation end EP2.

[0065] As an example, if the diameter of the substrate W is 300 mm, the distance from the substrate center OP to the operating end EP1 is set to −150 mm or less. Also, the distance from the substrate center OP to the operating end EP2 is set to +150 mm or less. Here, the distance from the substrate center OP to the operating end EP1 is set to −147 mm, and the distance from the substrate center OP to the operating end EP2 is set to +147 mm.

[0066] 6, the relative position of the nozzle 7 when the nozzle 7 is located at the substrate center OP is indicated by 0. Furthermore, the relative position of the nozzle 7 when the nozzle 7 is located at the operation end EP1 is indicated by -147 mm. Furthermore, the relative position of the nozzle 7 when the nozzle 7 is located at the operation end EP2 is indicated by 147 mm.

[0067] The operation pattern of the nozzle 7 shown in Fig. 6 is shown as five reciprocating movements between the operating end EP1 and the operating end EP2. That is, an example is shown in which the nozzle 7 performs five reciprocating movements. For the first reciprocating movement in the operation pattern of the nozzle 7, the parts of the relative positions corresponding to the movements indicated by arrows a1 to a4 in Fig. 2 are assigned the same symbols.

[0068] The operation pattern of the nozzle 7 may be calculated by simulation or may be actually measured.

[0069] FIG. 7 is a diagram showing divided regions b1 to b15 of a substrate W. As shown in FIG. 7, the upper surface of the substrate W is divided into 15 divided regions b1 to b15 by a plurality of concentric circles centered on the substrate center OP. The divided region b15 is a circle, and the divided regions b1 to b14 are annular. The divided regions b1 to b14 each have the same radial length of the substrate W. The radial length of each of the divided regions b1 to b14 is the difference between the radius of the outer periphery and the radius of the inner periphery. The radius of the divided region b15 is the same as the radial length of each of the divided regions b1 to b14. Here, the radius of the divided region b15 is 10 mm, and the radial length of each of the divided regions b1 to b14 is 10 mm. The radial length of each of the divided regions b1 to b14 and the radius of the divided region b15 are equal to or greater than the inner diameter of the nozzle 7.

[0070] The nozzle 7 rotates about a rotation axis AX2, so the center of rotation is different from the substrate center OP. The range of movement of the nozzle 7 is a trajectory described as it moves from the operating end EP1 through the substrate center OP to the operating end EP2, and is an arc. The range of movement is divided into 30 movement sections d1 to d30 by divided regions b1 to b15. The movement sections d1 to d15 are sections of the trajectory of the nozzle 7 moving between the operating end EP1 and the substrate center OP that cross each of the divided regions b1 to b15. For example, the movement section d1 is a section that crosses the divided region b1 while the nozzle 7 moves between the operating end EP1 and the substrate center OP.

[0071] Furthermore, movement sections d16 to d30 are sections of the trajectory of the nozzle 7 as it moves between the operating end EP2 and the substrate center OP that cross the divided regions b15 to b1, respectively. For example, movement section d30 is a section where the nozzle 7 crosses the divided region b1 as it moves between the operating end EP2 and the substrate center OP.

[0072] The number of divided regions b1 to b15 is not limited to 15 and can be set to any value. In this case, the number of divisions into which the movement range is divided, in other words, the number of movement sections, differs.

[0073] Fig. 8 is a diagram showing the residence time of the nozzle 7. In Fig. 8, the horizontal axis indicates the position on the substrate W. The position of the substrate center OP is indicated by 0 mm, one end in the radial direction of the substrate W is indicated by -150 mm, and the other end in the radial direction of the substrate W is indicated by 150 mm. Movement sections d1 to d30 are allocated between -150 mm and +150 mm on the horizontal axis. Hereinafter, the movement sections d1 to d30 may be collectively referred to as "movement section dn", where n is an integer of 1 or greater.

[0074] The vertical axis indicates the residence time of the nozzle 7 in each of the movement sections d1 to d30. Here, the residence time in each of the movement sections d1 to d30 is shown when the nozzle 7 moves according to the operation pattern shown in FIG. 6. When the nozzle 7 ejects processing liquid while moving back and forth (when performing a scan process), the residence time is the cumulative total of the time that the nozzle 7 is positioned in each of the movement sections d1 to d30. For example, when the nozzle 7 moves according to the operation pattern of the nozzle 7 shown in FIG. 6, the nozzle 7 crosses the movement section d2 10 times. The residence time in the movement section d2 is the cumulative total of the time that the nozzle 7 crosses the movement section d2. In this example, if the time it takes the nozzle 7 to cross the movement section d2 once is "T," the residence time in the movement section d2 is "T x 10."

[0075] That is, the residence time of the nozzle 7 in a certain movement section dn is the value (T×K) obtained by multiplying the time T for the nozzle 7 to cross that movement section dn once by the number of times K the nozzle 7 crosses that movement section dn. If the number of times the nozzle 7 moves back and forth is "M", then K = M×2.

[0076] Returning to FIG. 4, the first trained model LM1 is constructed by machine learning the first training data TND (FIG. 14). The first training data TND includes first training information 74 and second training information 75 (FIG. 14). The first training information 74 is an explanatory variable indicating the amount of treatment of the substrate W by the treatment liquid at each of multiple radial positions of the substrate W. The second training information 75 is a target variable indicating the residence time of the nozzle 7 in each of multiple movement sections dn that divide the movement range of the nozzle 7 as it moves while ejecting the treatment liquid onto the substrate W.

[0077] The first trained model LM1 causes the control device 21 (computer) to function so as to input target throughput information and output predicted stay time information.

[0078] 9 is a diagram showing the operation of the first trained model LM1. As shown in Fig. 4 and Fig. 9, the residence time acquisition unit 212 inputs target processing amount information 41 to the first trained model LM1. The target processing amount information 41 includes information indicating a radial position 42 of the substrate W and information on a target value 20a of the processing amount of the substrate W.

[0079] The target value 20a of the processing amount may be referred to as the "target processing amount 20a."

[0080] In the target processing amount information 41, each radial position 42 is indicated by the radial distance (mm) from the substrate center OP to each position 42. In the target processing amount information 41, a target processing amount 20a is associated with each of the multiple radial positions 42.

[0081] When target processing amount information 41 is input, the first trained model LM1 outputs predicted stay time information 43.

[0082] The predicted residence time information 43 includes information on the movement section dn of the nozzle 7 and information on the predicted value tn of the residence time of the nozzle 7. The predicted value tn of the residence time may be referred to as "predicted residence time tn."

[0083] In the predicted stay time information 43, a predicted stay time tn is associated with each of a plurality of movement sections dn.

[0084] 4, the speed acquisition unit 213 acquires target nozzle speed information RP1 by processing the predicted residence time information 43. The target nozzle speed information RP1 indicates target values ​​for the movement speed of the nozzle 7 set for each of a plurality of nozzle positions that divide the movement range of the nozzle 7, which moves while discharging the processing liquid onto the substrate W, into a plurality of movement sections dn. The control device 21 corresponds to an example of the "nozzle speed acquisition device" of the present invention.

[0085] Fig. 10 is a diagram showing the operation of the velocity acquisition unit 213. As shown in Fig. 4 and Fig. 10, the velocity acquisition unit 213 receives predicted residence time information 43. Then, the velocity acquisition unit 213 processes the predicted residence time information 43 in accordance with a velocity derivation algorithm to generate target nozzle velocity information RP1.

[0086] The target nozzle speed information RP1 includes information indicating the nozzle position zn and information indicating the target value vn of the movement speed of the nozzle 7, where n is an integer greater than or equal to 1. The target value vn of the movement speed may be referred to as the "target movement speed vn."

[0087] In the target nozzle speed information RP1, each nozzle position zn is indicated by the radial distance (mm) from the substrate center OP to the nozzle position zn. In the target nozzle speed information RP1, a target movement speed vn is associated with each of the multiple nozzle positions zn. Furthermore, the multiple nozzle positions zn each correspond to a multiple movement sections dn (FIG. 7) that divide the movement range of the nozzle 7.

[0088] FIG. 11 is a diagram showing the relationship between nozzle position zn and target movement speed vn of the nozzle 7. FIG. 11 shows the substrate W in a plan view and a graph GP1. As shown in FIG. 11, nozzle positions z1 to z12 are determined corresponding to movement sections d1 to d12 of the nozzle 7. In addition, in graph GP1, the horizontal axis represents nozzle position zn (mm), and the vertical axis represents target movement speed vn (mm / sec) of the nozzle 7. In other words, graph GP1 represents target nozzle speed information RP1 (FIG. 10). As shown in graph GP1, the speed acquisition unit 213 (FIG. 4) calculates the target movement speed vn of the nozzle 7 for each nozzle position zn, thereby obtaining target nozzle speed information RP1.

[0089] Next, an example of a speed derivation algorithm for deriving target nozzle speed information RP1 will be described with reference to Figures 4 and 12. Figure 12 is a diagram for explaining the speed derivation algorithm. Figure 12 shows a graph GP2. The graph GP2 represents predicted residence time information 43 (Figure 9). That is, in graph GP2, the horizontal axis shows the movement section dn of the nozzle 7, and the vertical axis shows the predicted residence time tn (seconds) of the nozzle 7. Furthermore, the value in parentheses on the horizontal axis indicates the nozzle position zn.

[0090] In the speed derivation algorithm executed by the speed acquisition unit 213, first, the speed acquisition unit 213 determines the operating end portions EP1 and EP2 (FIG. 2) of the nozzle 7 during the scanning process of the substrate W. Specifically, the speed acquisition unit 213 selects two predicted residence times t1 and t12 that are equal to or greater than the threshold value TH from among the multiple predicted residence times tn in the predicted residence time information 43. Then, the speed acquisition unit 213 determines two nozzle positions z1 and z12 in two movement sections d1 and d12 corresponding to the two predicted residence times t1 and t12, respectively, as the operating end portions EP1 and EP2. Next, the speed acquisition unit 213 determines the number M of reciprocating movements of the nozzle 7 during the scanning process. M is an integer greater than or equal to 1.

[0091] Next, the speed acquisition unit 213 calculates, for each movement section dn, the time Tn that the nozzle 7 takes to cross the movement section dn once. If the predicted residence time in the movement section dn is "tn", then Tn = tn / 2M, where n is an integer equal to or greater than 1.

[0092] However, the speed acquisition unit 213 excludes the delay time in the horizontal rotation of the nozzle 7 caused by the electric motor of the first movement mechanism 95 (FIG. 1) for time T1 at the operating end EP1 (movement section d1) and time T12 at the operating end EP2 (movement section d12). In other words, a delay occurs in the horizontal rotation when the electric motor starts to rotate. For this reason, the speed acquisition unit 213 calculates the target nozzle speed information RP1 without including the delay time.

[0093] Specifically, at the operating end EP1, the nozzle 7 crosses 2M times, but the electric motor starts rotating M times. Therefore, if the delay time at the start of one rotation of the electric motor is "Td," the total delay time td at the operating end EP1 is td = Td × M. Therefore, the speed acquisition unit 213 sets the new predicted stay time t1# at the operating end EP1 as the time (t1 - td) obtained by subtracting the delay time td from the predicted stay time t1. As a result, the time T1 for the nozzle 7 to cross the movement section d1 corresponding to the operating end EP1 once is T1 = t1# / 2M. Similarly, the time T12 for the nozzle 7 to cross the movement section d12 corresponding to the operating end EP12 once is calculated taking the delay time Td into account.

[0094] Next, the speed acquisition unit 213 calculates the target movement speed vn of the nozzle 7 at the nozzle position zn in the movement section dn by dividing the radial length Ln of the movement section dn by the time Tn it takes for the nozzle 7 to cross the movement section dn once (vn=Ln / Tn). In the example of Fig. 12, the lengths L1 to L12 are equal. In other words, the movement sections d1 to d12 are equally spaced.

[0095] However, the speed acquisition unit 213 sets the target movement speed v1 of the nozzle 7 at the operating end EP1 and the target movement speed v12 of the nozzle 7 at the operating end EP12 to zero (v1 = v12 = 0). This is because the nozzle 7 turns back at the operating end parts EP1 and EP2, and the movement speed of the nozzle 7 becomes zero. However, in this case, the predicted residence times t1 and t12 in the movement sections d1 and d12 corresponding to the operating end parts EP1 and EP12 will not be reflected in the target nozzle speed information RP1.

[0096] Therefore, in order to reflect the predicted dwell times t1 and t12 in the target nozzle speed information RP1, the speed acquisition unit 213, as an example, reflects the predicted dwell times t1 and t12 in the movement sections d1 and d12 corresponding to the operating end portions EP1 and EP12 to the target movement speeds v2 and v11 at nozzle positions z2 and z11 (movement sections d2 and d11) adjacent to nozzle positions z1 and z12 (movement sections d1 and d12) where the target movement speeds v1 and v12 are set to zero.

[0097] That is, the speed acquisition unit 213 calculates the virtual speed v1# at the nozzle position z1 based on the predicted stay time t1 in the movement section d1 corresponding to the operating end portion EP1 and the length L1 of the movement section d1, as shown in the following equation. v1#=L1 / T1

[0098] Furthermore, the speed acquisition unit 213 calculates a virtual speed v2# at the nozzle position z2 based on the predicted stay time t2 in the movement section d2 adjacent to the movement section d1 and the length L2 of the movement section d2, as shown in the following equation. v2#=L2 / T2

[0099] Furthermore, the speed acquisition unit 213 calculates the average value of the virtual speeds v1# and v2# as shown in the following equation, and sets the average value as the target movement speed v2 at the nozzle position z2. v2=(v1#+v2#) / 2

[0100] Similarly, the speed acquisition unit 213 calculates the target movement speed v11 at the nozzle position z11 adjacent to the nozzle position z12 based on the following equation.

[0101] v12#=L12 / T12 v11#=L11 / T11 v11=(v11#+v12#) / 2

[0102] Next, a substrate processing method according to this embodiment will be described with reference to Figures 4 and 13. Figure 13 is a flowchart showing the substrate processing method executed by the substrate processing apparatus 100. Specifically, the substrate processing method is executed by the control unit 21A executing a control program PG1. Therefore, the control program PG1 corresponds to an example of a "computer program" in the present invention.

[0103] 13, the substrate processing method includes steps S1 to S10. The substrate processing method preferably includes step S100 between step S4 and step S5. Step S100 will be described later in a modified example.

[0104] As shown in FIGS. 4 and 13, first, in step S1, the target processing amount calculation unit 211 acquires target thickness information of the target object TG constituting the substrate W.

[0105] Next, in step S2, the control unit 21A controls the transfer robot (not shown) to load the substrate W into the chamber 2. As a result, the transfer robot loads the substrate W into the chamber 2. Then, the spin chuck 3 starts rotating the substrate W.

[0106] Next, in step S3, the measurement control unit 210 controls the thickness measurement unit 15 and the second moving unit 17 so that the thickness measurement unit 15 measures the thickness of the target object TG constituting the substrate W. As a result, the thickness measurement unit 15 measures the thickness of the target object TG before processing, and outputs pre-processing measured thickness information indicating the thickness of the target object TG before processing to the control device 21.

[0107] Next, in step S4, the target processing amount calculation unit 211 calculates target processing amount information 41 based on the target thickness information and the pre-processing measured thickness information. Specifically, the target processing amount calculation unit 211 calculates, for each radial position 42 on the substrate W, a target processing amount 20a, which is the difference between the target thickness of the target object TG constituting the substrate W indicated by the target thickness information and the thickness of the object TG indicated by the pre-processing measured thickness information. Then, the memory unit 21B stores the target processing amount information 41 indicating the target processing amount 20a.

[0108] Next, in step S5, the stay time acquisition unit 212 inputs the target processing amount information 41 to the first trained model LM1 constructed by machine learning the first training data TND. As a result, the first trained model LM1 outputs predicted stay time information 43.

[0109] Next, in step S6, the stay time acquisition unit 212 acquires predicted stay time information 43 from the first trained model LM1.

[0110] Next, in step S7, the speed acquisition unit 213 processes the predicted stay time information 43 indicating the predicted stay time tn of the nozzle 7 in each of the multiple movement sections dn to acquire the target nozzle speed information RP1 (FIG. 10).

[0111] Next, in step S8, the substrate processing control unit 214 controls the substrate processing unit 1 to process the substrate W in accordance with target nozzle speed information RP1 indicating the target movement speed vn of the nozzle 7 at each of the plurality of nozzle positions zn. As a result, the substrate processing unit 1 processes the substrate W in accordance with the target nozzle speed information RP1. Specifically, the nozzle 7 of the substrate processing unit 1 performs a scan process at the target movement speed vn of the nozzle 7 at each of the plurality of nozzle positions zn to process the substrate W.

[0112] Specifically, the substrate process controller 214 controls the substrate processing unit 1 to process the substrate W in accordance with the recipe RP (FIG. 4). That is, the substrate process controller 214 controls the substrate processing unit 1 to process the substrate W in accordance with the target nozzle speed information RP1 and the processing condition information RP2. The processing condition information RP2 is information that defines the process conditions for processing the substrate W. The processing condition information RP2 includes, for example, at least one of temperature information indicating the temperature of the processing liquid, concentration information indicating the concentration of the processing liquid, flow rate information indicating the flow rate of the processing liquid, and rotation speed information indicating the rotation speed of the substrate W.

[0113] Next, in step S9, the measurement control unit 210 controls the thickness measurement unit 15 and the second moving unit 17 so that the thickness measurement unit 15 measures the thickness of the target object TG constituting the substrate W. As a result, the thickness measurement unit 15 measures the thickness of the target object TG after processing, and outputs post-processing measured thickness information indicating the thickness of the target object TG after processing to the control device 21. Then, the spin chuck 3 stops the rotation of the substrate W.

[0114] Next, in step S10, the control unit 21A controls the transfer robot to unload the substrate W from the chamber 2. As a result, the transfer robot unloads the substrate W from the chamber 2. Then, the substrate processing method is completed.

[0115] 9, 10, and 13, according to this embodiment, instead of directly determining the target nozzle speed information RP1 from the target processing amount information 41, the first trained model LM1 is used to first determine the predicted residence time information 43 from the target processing amount information 41 (FIG. 9, steps S5 and S6), and then the target nozzle speed information RP1 is determined from the predicted residence time information 43 (FIG. 10, step S7). Therefore, compared to the case where the target nozzle speed information RP1 is determined directly from the target processing amount information 41, it is possible to obtain target nozzle speed information RP1 that can accurately achieve the target processing amount 20a (FIG. 9) for the substrate W.

[0116] The reason why the target throughput 20a can be achieved with high accuracy is as follows.

[0117] That is, a strong correlation is observed between the amount of processing of the substrate W by the processing liquid and the residence time of the nozzle 7 (as demonstrated by the inventors of the present application in the examples described below). Therefore, a first trained model LM1 is used, which is constructed by machine learning the first learning information 74 indicating the amount of processing of the substrate W and the second learning information 75 indicating the residence time of the nozzle 7. This is because the first trained model LM1 can accurately determine the predicted residence time tn of the nozzle 7 that can achieve the target processing amount 20a. This is because the first trained model LM1 is a trained model generated by machine learning by finding a certain rule between the processing amount of the substrate W and the residence time of the nozzle 7, which are strongly correlated.

[0118] If the predicted residence time tn of the nozzle 7 that can achieve the target processing volume 20a can be accurately determined, the target movement speed vn of the nozzle 7 that can accurately achieve the target processing volume 20a can be accurately determined from the predicted residence time tn. As a result, by controlling the nozzle 7 at the determined target movement speed vn, the target processing volume 20a of the substrate W can be accurately achieved.

[0119] Furthermore, according to this embodiment, target nozzle speed information RP1 that can accurately achieve the target processing volume 20a of the substrate W can be obtained from the first learned model LM1, and therefore the cost required to set the target nozzle speed information RP1 can be reduced compared to when the target nozzle speed information RP1 is set based on experience, such as manual work.

[0120] Furthermore, according to this embodiment, by processing the substrate W in accordance with the target nozzle speed information RP1, the target processing amount 20a for the substrate W can be achieved with high precision, thereby reducing the workload in post-processes (mechanical polishing, etc.).

[0121] Furthermore, according to this embodiment, by performing steps S6 and S7, it is possible to acquire the target movement speed vn of the nozzle 7 that can achieve the target throughput 20a from the predicted residence time tn of the nozzle 7 that can achieve the target throughput 20a. Steps S6 and S7 correspond to an example of the "nozzle speed acquisition method" of the present invention.

[0122] 12 and 13, according to this embodiment, in step S6, target nozzle speed information RP1 is calculated based on the number of reciprocating movements M of the nozzle 7, the predicted residence time tn of the nozzle 7, and the length Ln of the movement section dn in the radial direction of the substrate W. Therefore, when the nozzle 7 performs a scan process, the target nozzle speed information RP1 can be calculated by a simple calculation.

[0123] Next, learning device 200 will be described with reference to Figs. 14 to 16. Learning device 200 is, for example, a computer. Fig. 14 is a block diagram showing learning device 200. As shown in Fig. 14, learning device 200 includes processing unit 5A, storage unit 5B, communication unit 5C, input unit 5D, and display unit 5E.

[0124] The processing unit 5A includes a processor such as a CPU and a GPU. The storage unit 5B includes a storage device and stores data and computer programs. The processor of the processing unit 5A executes the computer programs stored in the storage device of the storage unit 5B to perform various processes. The hardware configuration of the storage unit 5B is similar to the hardware configuration of the storage unit 21B in FIG. 4.

[0125] The memory unit 5B stores a processing program PG2. The processing unit 5A executes the processing program PG2 and functions as an acquisition unit 71, a learning unit 72, and a memory control unit 73. In other words, the processing unit 5A includes the acquisition unit 71, the learning unit 72, and the memory control unit 73.

[0126] The communication unit 5C is connected to a network and communicates with external devices. The hardware configuration of the communication unit 5C is similar to the hardware configuration of the communication unit 21C in FIG. 4. The input unit 5D is an input device for inputting various information to the processing unit 5A. The hardware configuration of the input unit 5D is similar to the hardware configuration of the input unit 21D in FIG. 4. The display unit 5E displays images. The hardware configuration of the display unit 5E is similar to the hardware configuration of the display unit 21E in FIG. 4.

[0127] The processing unit 5A will be described with continued reference to Fig. 14. The acquisition unit 71 of the processing unit 5A acquires a plurality of first learning data TNDs from an external source. For example, the acquisition unit 71 acquires the plurality of first learning data TNDs from the learning data creation device 300 (Fig. 17) via a network and a communication unit 5C.

[0128] The storage control unit 73 controls the storage unit 5B to store the first learning data TND, so that the storage unit 5B stores the first learning data TND.

[0129] The first learning data TND includes first learning information 74 and second learning information 75. That is, the first learning information 74 and the second learning information 75 constitute the first learning data TND. The first learning information 74 is an explanatory variable that indicates the amount of processing of the substrate W by the processing liquid at each of a plurality of positions in the radial direction of the substrate W. The second learning information 75 is a target variable that indicates the residence time of the nozzle 7 in each of a plurality of movement sections dn that divide the movement range of the nozzle 7 that moves while discharging the processing liquid onto the substrate W.

[0130] The memory unit 5B stores a learning program PG3. The learning program PG3 is a program for executing a machine learning algorithm to find certain rules from a plurality of first learning data TND and generate a first trained model LM1 that expresses the found rules. In other words, the learning program PG3 is a program for executing a machine learning algorithm to find certain rules between the target processing volume 20a and the predicted stay time tn by machine learning the plurality of first learning data TND and generate a trained model LM that expresses the found rules.

[0131] The machine learning algorithm is not particularly limited 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 first trained model LM1 includes a decision tree, a nearest neighbor method, a naive Bayes classifier, a support vector machine, or a neural network. Backpropagation may be used in the machine learning to generate the first trained model LM1. The machine learning algorithm is typically a "supervised learning" algorithm.

[0132] For example, a neural network includes an input layer, one or more hidden 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), and performs deep learning. For example, a deep neural network includes an input layer, multiple hidden layers, and an output layer.

[0133] As an example, the machine learning algorithm is LightGBM (Light Gradient Boosting Machine). LightGBM is a decision tree algorithm that uses a gradient boosting algorithm. LightGBM is also a "supervised learning" algorithm.

[0134] The learning unit 72 performs machine learning on a plurality of first learning data TND based on the learning program PG3. As a result, certain rules are found from the plurality of first learning data TND, and a first trained model LM1 is generated. In other words, the first trained model LM1 is constructed by machine learning the first learning data TND. The memory unit 5B stores the first trained model LM1.

[0135] Specifically, the learning unit 72 executes the learning program PG3 to find certain rules between the explanatory variables and the target variables included in the first learning data TND, and generates the first trained model LM1.

[0136] More specifically, the learning unit 72 performs machine learning on a plurality of first learning data TND based on the learning program PG3 to calculate a plurality of learned parameters and generate a first learned model LM1 including one or more functions to which the plurality of learned parameters are applied. The learned parameters are parameters (coefficients) obtained based on the results of machine learning using the plurality of first learning data TND.

[0137] The first trained model LM1 is a computer program that causes a computer to function so as to input target processing amount information 41 and output predicted staying time information 43. In other words, the first trained model LM1 inputs target processing amount information 41 and outputs predicted staying time information 43.

[0138] Fig. 15 is a diagram showing the operation of the learning device 200. As shown in Fig. 15, the learning unit 72 executes the learning program PG3 to find certain rules between the first learning information 74 (explanatory variables) and the second learning information 75 (target variables), and generates the first trained model LM1.

[0139] The first learning information 74 includes information indicating the radial positions 42 of the substrate W and information on the processing amount 77 of the substrate W. In the first learning information 74, each radial position 42 is indicated by the radial distance (mm) from the substrate center OP to each position 42. In the first learning information 74, a processing amount 77 is associated with each of the multiple radial positions 42. The first learning information 74 may further include, for example, at least one of information indicating the number of positions 42 and information on the outermost radial position 42 ("129 mm" in the example of FIG. 15).

[0140] In this case, the target processing volume information 41 in FIG. 9 may further include, for example, at least one of information indicating the number of positions 42 and information on the outermost radial position 42 ("129 mm" in the example of FIG. 15).

[0141] The second learning information 75 includes information on the movement section dn of the nozzle 7 and information on the residence time sn of the nozzle 7, where n is an integer greater than or equal to 1. In the second learning information 75, a plurality of residence times sn are associated with each of the plurality of movement sections dn. The second learning information 75 may further include, for example, at least one of information indicating the number of movement sections dn, information indicating the length of the movement sections dn in the radial direction of the substrate W, and information on the outermost movement sections dn in the radial direction ("movement sections d1, d12" in the example of FIG. 15).

[0142] In this case, the predicted residence time information 43 in Figure 9 may further include at least one of information indicating the number of movement sections dn, information indicating the length of the movement sections dn in the radial direction of the substrate W, and information on the outermost movement sections dn in the radial direction (in the example of Figure 15, "movement sections d1, d12").

[0143] In this case, the target nozzle speed information RP1 in Figure 10 may further include, for example, at least one of information indicating the number of nozzle positions zn and information on the outermost nozzle positions zn in the radial direction (in the example of Figure 10, "-120 mm" and "120 mm").

[0144] Next, the learning method will be described with reference to Figs. 15 and 16. The learning method corresponds to an example of a "trained model generation method" of the present invention. Fig. 16 is a flowchart showing the learning method. As shown in Fig. 16, the learning method includes steps S21 to S24. The learning method is executed by the learning device 200. Specifically, the learning method is executed by the processing unit 5A executing the processing program PG2. Therefore, the processing program PG2 corresponds to an example of a "computer program" of the present invention.

[0145] As shown in FIGS. 15 and 16, first, in step S21, the learning unit 72 of the learning device 200 acquires a plurality of first learning data TND from the learning data creation device 300 (FIG. 17).

[0146] Next, in step S22, the learning unit 72 performs machine learning of a plurality of first learning data TND based on the learning program PG3.

[0147] Next, in step S23, the learning unit 72 determines whether a learning termination condition is met. The learning termination condition is a predetermined condition for terminating machine learning. For example, the learning termination condition is when the number of iterations reaches a specified number.

[0148] If the determination in step S23 is negative, the process proceeds to step S21, whereupon the machine learning is repeated.

[0149] On the other hand, if the determination in step S23 is affirmative, the process proceeds to step S24.

[0150] In step S24, the learning unit 72 outputs a model (one or more functions) to which the latest parameters (coefficients), that is, the learned parameters (coefficients), are applied as the first trained model LM1. Then, the storage unit 5B stores the first trained model LM1.

[0151] As described above, the learning unit 72 executes steps S21 to S24, thereby generating the first trained model LM1.

[0152] That is, the learning device 200 performs machine learning, and is therefore able to find patterns in the very complex first learning data TND, which has a huge amount of analysis targets, and create a highly accurate first learned model LM1.

[0153] 14 and 15, the learning method (trained model generation method) according to this embodiment performs machine learning on the highly correlated processing amount 77 of the substrate W (explanatory variable) and the residence time sn of the nozzle 7 (objective variable), thereby finding a certain rule between the processing amount 77 of the substrate W and the residence time sn of the nozzle 7, and generating a first trained model LM1. As a result, it is possible to generate a first trained model LM1 that can accurately determine the predicted residence time tn of the nozzle 7 that can achieve the target processing amount 20a.

[0154] Furthermore, according to this embodiment, by inputting target processing volume information 41 into the first trained model LM1, it is possible to output predicted residence time information 43 indicating the predicted residence time tn of the nozzle 7 that can achieve the target processing volume 20a. This is because the first trained model LM1 is a trained model generated by machine learning by finding a certain rule between the processing volume 77 of the substrate W and the residence time sn of the nozzle 7, which are strongly correlated.

[0155] Next, the training data creation device 300 will be described with reference to Figures 17 to 19. The training data creation device 300 creates first training data TND, which is the target of machine learning for generating the first trained model LM1.

[0156] Fig. 17 is a block diagram showing a training data creation device 300. As shown in Fig. 17, the training data creation device 300 includes a processing unit 4A, a storage unit 4B, a communication unit 4C, an input unit 4D, and a display unit 4E.

[0157] The processing unit 4A includes a processor such as a CPU and a GPU. The storage unit 4B includes a storage device and stores data and computer programs. The processor of the processing unit 4A executes the computer programs stored in the storage device of the storage unit 4B to perform various processes. The hardware configuration of the storage unit 4B is similar to the hardware configuration of the storage unit 21B in FIG. 4.

[0158] Specifically, the memory unit 4B stores a learning data creation program PG4. The processing unit 4A executes the learning data creation program PG4 to function as a learning data creation unit 61 and a memory control unit 62. That is, the processing unit 4A includes the learning data creation unit 61 and the memory control unit 62. The learning data creation unit 61 includes an acquisition unit 611 and a correction unit 613.

[0159] The communication unit 4C is connected to a network and communicates with external devices. The communication unit 4C is a communication device, such as a network interface controller.

[0160] 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 and pointing device, or a touch panel.

[0161] The display unit 4E displays an image and is, for example, a liquid crystal display or an organic electroluminescence display.

[0162] Here, the possibility of variations in the throughput of substrates W in the substrate processing apparatus 100 will be described. Even when the substrate processing apparatus 100 processes substrates W according to the same recipe RP, differences in the throughput may occur depending on the days on which the substrates W are processed. For example, even when the same recipe RP is used, the throughput of substrates W on a given day may be greater than the throughput of substrates W in the past, or may be less than the throughput of substrates W in the past. In this case, differences in the processing environment from day to day, such as weather, the state of each component constituting the substrate processing apparatus 100, the state of the substrates W, the state of the processing liquid, the internal environment of the chamber 2, and a combination of various factors, may cause variations in the throughput from day to day. As such, differences in the throughput may occur when the timing of processing substrates W, such as when the days or times on which substrates W are processed are different. Hereinafter, differences in the throughput when the timing of processing substrates W is different may be referred to as a "timing-attributable processing throughput difference."

[0163] When first learning information 74 indicating the processing amount 77 at each of multiple positions 42 of the substrate W is acquired from the substrate processing apparatus 100, if there is a timing-attributable processing amount difference, the first trained model LM1 generated by machine learning the multiple first learning information 74 may also be affected by the timing-attributable processing amount difference. This is because the multiple first learning information 74 may represent processing amounts when the substrate W is processed at different times, such as on different days or at different times.

[0164] Therefore, in this embodiment, as a preferred example, in order to reduce the influence of the timing-induced processing amount difference, the first learning information 74 acquired from the substrate processing apparatus 100 is corrected. Hereinafter, the first learning information 74 acquired from the substrate processing apparatus 100 before correction will be referred to as "original learning information 65."

[0165] Also, "timing" may refer to, for example, different days or different times.

[0166] The processing unit 4A will be described with continued reference to Fig. 17. In the learning data creation unit 61 of the processing unit 4A, the acquisition unit 611 acquires original learning information 65 from the substrate processing apparatus 100. The original learning information 65 indicates the amount of processing of the substrate W by the processing liquid at each of a plurality of positions in the radial direction of the substrate W.

[0167] The correction unit 613 corrects the original learning information 65 based on the comparison result between the first reference processing amount information 63 and the second reference processing amount information 64, and outputs the corrected original learning information 65 as the first learning information 74. In other words, the first learning information 74 is corrected information based on the comparison result between the first reference processing amount information 63 and the second reference processing amount information 64.

[0168] The first reference processing amount information 63 indicates the processing result of the first reference substrate. Specifically, the first reference processing amount information 63 indicates the processing amount of the first reference substrate by the processing liquid at each of a plurality of positions in the radial direction of the first reference substrate when the first reference substrate is processed by the processing liquid in accordance with the reference nozzle speed information R11.

[0169] The second reference processing amount information 64 indicates the processing result of a second reference substrate that is processed at a different timing from that of the first reference substrate. Specifically, the second reference processing amount information 64 indicates the processing amount of the second reference substrate by the processing liquid at each of a plurality of positions in the radial direction of the second reference substrate when the second reference substrate is processed with the processing liquid in accordance with the reference nozzle speed information R11.

[0170] The reference nozzle speed information R11 indicates the movement speed of the nozzle 7, which is set for each of a plurality of nozzle positions that divides the movement range of the nozzle 7 that ejects the processing liquid into a plurality of sections when processing the first reference substrate and the second reference substrate with the processing liquid. The reference nozzle speed information R11 corresponds to an example of the "reference processing condition" of the present invention.

[0171] According to this embodiment, the first reference processing amount information 63 and the second reference processing amount information 64 indicate the processing amounts for the first reference substrate and the second reference substrate processed at different timings under the same reference processing conditions. Then, correction is performed based on the comparison result between the first reference processing amount information 63 and the second reference processing amount information 64. Therefore, in this embodiment, the original learning information 65 can be corrected based on the comparison result between the first reference processing amount information 63 and the second reference processing amount information 64 so as to reduce the influence of the timing-related processing amount difference.

[0172] Specifically, the second reference processing amount information 64 is information generated on the same day as the day on which the original learning information 65, which is information before the first learning information 74 is corrected, was generated. In other words, the day on which the second reference substrate was processed is the same as the day on which the substrate W in the original learning information 65 was processed. The first reference processing amount information 63 is information generated earlier than the day on which the second reference processing amount information 64 was generated. In other words, the day on which the first reference substrate was processed is earlier than the day on which the second reference substrate was processed.

[0173] Therefore, according to this embodiment, the processing environment of the substrate W in the original learning information 65 can be made common to the processing environment of the second reference substrate W in the second reference processing amount information 64. As a result, by correcting the original learning information 65 based on the comparison result between the second reference processing amount information 64 and the first reference processing amount information 63, it is possible to reduce the influence of timing-attributable processing amount differences on the first learning information 74, using the first reference processing amount information 63 as a reference.

[0174] Next, a learning data creation method including a correction process will be described with reference to Figs. 17 to 19. Fig. 18(a) is a diagram showing an example of original learning information 65. Fig. 18(b) is a diagram showing an example of second reference processing amount information 64. Fig. 18(c) is a diagram showing an example of first reference processing amount information 63. Fig. 18(d) is a diagram showing an example of comparison information 66. Fig. 18(e) is a diagram showing an example of correction values ​​67. Fig. 18(f) is a diagram showing an example of first learning information 74.

[0175] 19 is a flowchart showing a training data creation method. As shown in FIG. 19, the training data creation method includes steps S31 to S34. The training data creation method is executed by the training data creation device 300. Specifically, the training data creation method is executed by the processing unit 4A by executing the training data creation program PG4. The training data creation program PG4 corresponds to an example of the "computer program" of the present invention.

[0176] 17 to 19, first, in step S31, the acquisition unit 611 acquires original learning information 65 from the substrate processing apparatus 100. The memory control unit 62 controls the memory unit 4B to store the original learning information 65. The memory unit 4B stores the original learning information 65.

[0177] 18(a), the original learning information 65 includes information indicating the radial positions 42 of the substrate W and information on the processing amount 81 of the substrate W. Each radial position 42 is indicated by the radial distance (mm) from the substrate center OP to each position 42. In the original learning information 65, the processing amount 81 is associated with each of the multiple radial positions 42.

[0178] Next, in step S31, the learning data creation unit 61 corrects the original learning information 65 and outputs the corrected original learning information 65 as the first learning information 74.

[0179] Specifically, step S31 includes steps S321 to S324.

[0180] In step S321, the acquisition unit 611 acquires the second reference processing amount information 64 from the substrate processing apparatus 100. The memory control unit 62 controls the memory unit 4B to store the second reference processing amount information 64. The memory unit 4B stores the second reference processing amount information 64.

[0181] 18(b), the second reference processing amount information 64 includes information indicating the radial positions 42 of the second reference substrate and information on the processing amounts 82 of the second reference substrate. In the second reference processing amount information 64, the processing amounts 82 are associated with each of the multiple radial positions 42.

[0182] The date on which the second reference substrate was processed in the second reference processing amount information 64 is the same as the date on which the substrate W was processed in the original learning information 65 in Fig. 18(a). In other words, the second reference processing amount information 64 and the original learning information 65 were generated on the same day.

[0183] Next, in step S322, the acquisition unit 611 acquires first reference processing amount information 63 from the substrate processing apparatus 100. The memory control unit 62 controls the memory unit 4B to store the first reference processing amount information 63. The memory unit 4B stores the first reference processing amount information 63.

[0184] 18(c), the first reference processing amount information 63 includes information indicating the radial positions 42 of the first reference substrate and information on the processing amount 83 of the first reference substrate. In the first reference processing amount information 63, the processing amount 83 is associated with each of the multiple radial positions 42.

[0185] The date on which the first reference substrate was processed in the first reference processing amount information 63 is earlier than the date on which the second reference substrate was processed in the second reference processing amount information 64. In other words, the first reference processing amount information 63 was generated earlier than the second reference processing amount information 64. Therefore, the storage unit 4B may store the first reference processing amount information 63 in advance.

[0186] Next, in step S323, the correction unit 613 compares the second reference processing amount information 64 with the first reference processing amount information 63, and outputs comparison information 66 indicating the comparison result.

[0187] As an example, the correction unit 613 calculates the difference between the processing amount 82 of the second reference processing amount information 64 and the processing amount 83 of the first reference processing amount information 63, and outputs the comparison information 66 indicating the difference amount 84.

[0188] 18(d), the comparison information 66 includes information indicating the radial positions 42 of the substrates (first reference substrate and second reference substrate) and information on the difference amount 84. The correction unit 613 calculates the difference amount 84 for each position 42. Therefore, in the comparison information 66, the difference amount 84 is associated with each of the multiple radial positions 42. In the example of FIG. 18(d), the difference amount 84 is a value obtained by subtracting the processing amount 83 of the first reference processing amount information 63 from the processing amount 82 of the second reference processing amount information 64.

[0189] 18(d), it can be inferred that the processing amount of substrates W by the substrate processing apparatus 100 on the day when the second reference processing amount information 64 and the original learning information 65 were generated is larger than the processing amount of substrates W by the substrate processing apparatus 100 on the day when the first reference processing amount information 63 was generated. Therefore, the original learning information 65 is corrected to reduce the processing amount.

[0190] Specifically, the correction unit 613 calculates the correction value 67 based on the comparison information 66. For example, the correction unit 613 calculates the average value of the multiple difference amounts 84 included in the comparison information 66, and sets the average value with the opposite sign as the correction value 67. An example of the correction value 67 in this case is shown in FIG. 18(e).

[0191] Next, in step S324, the correction unit 613 corrects the original learning information 65 based on the comparison result in step S323. Specifically, the correction unit 613 corrects the original learning information 65 using a correction value 67 calculated based on the comparison information 66. When using the correction value 67 of FIG. 18(e), the correction unit 613 corrects the original learning information 65 by adding the correction value 67 of FIG. 18(e) to the processing amount 81 of each position 42 of the original learning information 65 shown in FIG. 18(a). As a result, the first learning information 74 shown in FIG. 18(f) is generated.

[0192] The storage control unit 62 controls the storage unit 4B to store the first learning information 74. The storage unit 4B stores the first learning information 74.

[0193] Next, in step S33, the acquiring unit 611 acquires the second learning information 75 (FIG. 15) from the substrate processing apparatus 100.

[0194] Next, in step S34, the storage control unit 62 stores the first learning information 74 and the second learning information 75 as first learning data TND (FIG. 15) in the storage unit 4B, and the learning data creation method is then completed.

[0195] As described above with reference to Figures 17 to 19, according to this embodiment, by correcting the original learning information 65 based on the first reference processing amount information 63 and the second reference processing amount information 64, it is possible to generate first learning information 74 that reduces the influence of timing-related processing amount differences.

[0196] In particular, correction is not necessary for the first learning information 74 generated on the same day as the generation of the first reference processing amount information 63. The correction is performed to reduce the difference in processing environment between the first learning information 74 generated on the same day as the generation of the first reference processing amount information 63 and the original learning information 65 generated on a day after the generation of the first reference processing amount information 63, because the processing environment when the first learning information 74 generated on the same day as the generation of the first reference processing amount information 63 is generated serves as the reference. In other words, the original learning information 65 generated on a day after the generation of the first reference processing amount information 63 is corrected based on the first learning information 74 generated on the same day as the generation of the first reference processing amount information 63, thereby obtaining the corrected first learning information 74.

[0197] Although the difference amount 84 has been described as an example of the comparison information 66, the comparison information 66 is not limited to this. For example, the comparison information 66 may be the ratio RA (=A2 / A1) between the average value A2 of the processing amounts 82 of the second reference processing amount information 64 and the average value A1 of the processing amounts 83 of the first reference processing amount information 63. In this case, the correction value 67 is, for example, the reciprocal of the ratio RA (=A1 / A2). Then, in this case, the correction value 67 is multiplied by each processing amount 81 of the original learning information 65 to obtain the first learning information 74.

[0198] Next, acquisition of the first reference processing amount information 63 and the second reference processing amount information 64 will be described with reference to FIG. 4. When acquiring the first reference processing amount information 63, the substrate processing control unit 214 controls the substrate processing unit 1 (FIG. 1) to process a first reference substrate according to the reference recipe R1. The substrate processing unit 1 processes the first reference substrate according to the reference recipe R1. As a result, a processing amount 83 of the first reference substrate constituting the first reference processing amount information 63 is obtained. Similarly, the substrate processing unit 1 processes a second reference substrate according to the reference recipe R1, thereby obtaining a processing amount 82 of the second reference substrate constituting the second reference processing amount information 64.

[0199] The processing condition information R12 included in the reference recipe R1 is information that defines the process conditions for processing the first reference substrate and the second reference substrate. The processing condition information R12 includes, for example, at least one of temperature information indicating the temperature of the processing liquid, concentration information indicating the concentration of the processing liquid, flow rate information indicating the flow rate of the processing liquid, and rotation speed information indicating the rotation speed of the substrates (first reference substrate and second reference substrate).

[0200] (First Modification) A first modified example of this embodiment will be described with reference to Figures 4, 13, 20, and 21. The first modified example differs from the embodiment described above with reference to Figures 1 to 13 mainly in that the target processing amount information 41 is corrected. Below, the differences between the first modified example and the embodiment will be mainly described.

[0201] 4 and 13, in step S100, the correction unit 215 of the control unit 21A corrects the target processing amount information 41 calculated in step S4. Then, in step S5, the residence time acquisition unit 212 inputs the corrected target processing amount information 41 (hereinafter referred to as "target processing amount information 41a") to the first trained model LM1.

[0202] Next, the correction process will be described in detail with reference to Figs. 4, 20, and 21. Fig. 20(a) is a diagram showing an example of target processing amount information 41 before correction. Fig. 20(b) is a diagram showing an example of second reference processing amount information 64. Fig. 20(c) is a diagram showing an example of first reference processing amount information 63. Fig. 20(d) is a diagram showing an example of comparison information 66. Fig. 20(e) is a diagram showing an example of correction value 67. Fig. 20(f) is a diagram showing an example of target processing amount information 41a after correction. Fig. 21 is a flowchart showing the details of step S100 in Fig. 13. As shown in Fig. 21, step S100 includes steps S41 to S45.

[0203] As shown in FIGS. 20 and 21, first, in step S41, the correction unit 215 acquires the target processing amount information 41 before correction.

[0204] As shown in FIG. 20(a), the target processing amount information 41 includes information indicating a radial position 42 of the substrate W and information on the target processing amount 20a of the substrate W.

[0205] Next, in step S42, the correction unit 215 acquires second reference processing amount information 64a from the substrate processing apparatus 100. The definition of the second reference processing amount information 64a is the same as the definition of the second reference processing amount information 64.

[0206] As shown in FIG. 20(b), the second reference processing amount information 64a includes information indicating the radial position 42 of the second reference substrate and information on the processing amount 82 of the second reference substrate.

[0207] The date on which the second reference substrate was processed in the second reference processing amount information 64a is the same as the date on which the substrate W was processed in the target processing amount information 41 in Fig. 20(a). In other words, the second reference processing amount information 64a and the target processing amount information 41 were generated on the same day.

[0208] Next, in step S43, the correction unit 215 acquires the first reference processing amount information 63 from the substrate processing apparatus 100.

[0209] As shown in FIG. 20(c), the first reference processing amount information 63 includes information indicating the radial position 42 of the first reference substrate and information on the processing amount 83 of the first reference substrate.

[0210] The date on which the first reference substrate was processed in the first reference processing amount information 63 is an earlier date than the date on which the second reference substrate was processed in the second reference processing amount information 64a, as in Figures 18(b) and 18(c).

[0211] Next, in step S44, the correction unit 215 compares the second reference processing amount information 64a with the first reference processing amount information 63, and outputs comparison information 66a indicating the comparison result. As an example, as shown in FIG. 20(d), the comparison information 66a includes information indicating the radial position 42 of the substrate (first reference substrate and second reference substrate) and information on the difference amount 84. In this respect, it is the same as FIG. 18(d).

[0212] Specifically, the correction unit 215 calculates a correction value 67a based on the comparison information 66a. For example, the correction unit 215 calculates the average value of the multiple difference amounts 84 included in the comparison information 66a, and sets the average value with the opposite sign as the correction value 67a. An example of the correction value 67a in this case is shown in FIG. 20(e). In this respect, it is similar to FIG. 18(e).

[0213] Next, in step S45, the correction unit 215 corrects the target processing amount information 41 based on the comparison result in step S44. Specifically, the correction unit 215 corrects the target processing amount information 41 using a correction value 67a calculated based on the comparison information 66a. When using the correction value 67a of FIG. 20(e), the correction unit 215 corrects the target processing amount information 41 by adding the correction value 67a of FIG. 20(e) to the target processing amount 20a at each position 42 of the target processing amount information 41 shown in FIG. 20(a). As a result, corrected target processing amount information 41a shown in FIG. 20(f) is generated. The target processing amount information 41a includes information indicating the radial position 42 of the substrate W and information on the corrected target processing amount 20b. When step S45 is completed, the process proceeds to step S5 of FIG. 13.

[0214] 20 and 21, according to the first modification, it is possible to generate target processing amount information 41a in which the influence of the timing-attributable processing amount difference is reduced by correcting the target processing amount information 41 based on the first reference processing amount information 63 and the second reference processing amount information 64a. Therefore, by inputting the corrected target processing amount information 41a into the first trained model LM1, it is possible to obtain more accurate predicted stay time information 43.

[0215] Although the difference amount 84 has been described as an example of the comparison information 66a, the comparison information 66a is not limited to this, as in the embodiment described above with reference to FIG.

[0216] (Second Modification) A second modified example of this embodiment will be described with reference to Figures 4, 13, and 22. The second modified example differs mainly from the embodiment described above with reference to Figures 1 to 13 in that the staying time acquisition unit 212 uses the second trained model LM2. Below, the differences between the second modified example and the embodiment described above will be mainly described.

[0217] In the first modified example, the storage unit 21B of the substrate processing apparatus 100 shown in FIG. 4 stores a second trained model LM2.

[0218] Fig. 22 is a diagram showing the operation of the second trained model LM2 according to the second modified example. As shown in Fig. 13 and Fig. 22, in step S7, the dwell time acquisition unit 212 (Fig. 4) inputs predicted dwell time information 43 to the second trained model LM2 constructed by machine learning the second learning data, and acquires target nozzle speed information RP1 from the second trained model LM2.

[0219] The machine learning algorithm for generating the second trained model LM2 is the same as the machine learning algorithm for generating the first trained model LM1. Furthermore, the learning device 200 in Fig. 14 generates the second trained model LM2 by executing a learning method similar to the learning method shown in Fig. 16.

[0220] In this case, the second learning data includes fourth learning information and fifth learning information. The fourth learning information is an explanatory variable indicating the residence time of the nozzle 7 in each of a plurality of movement sections dn that divide the movement range of the nozzle 7 as it moves while discharging the processing liquid onto the substrate W. The fifth learning information is a response variable indicating the movement speed of the nozzle 7 set for each of a plurality of nozzle positions zn that divide the movement range of the nozzle 7 into a plurality of movement sections dn. The second learning data is acquired from the substrate processing apparatus 100.

[0221] The learning device 200 executes the learning program PG3 to find certain rules between the fourth learning information (explanatory variables) and the fifth learning information (target variables), and generates the second trained model LM2.

[0222] As described above, according to the second modified example, by inputting predicted dwell time information 43 into the second trained model LM2, it is possible to obtain target nozzle speed information RP1 that indicates the target movement speed vn that can achieve the target processing volume 20a. This is because, while there is a strong correlation between the processing volume of the substrate W and the dwell time of the nozzle 7 (FIG. 23), the second trained model LM2 is a trained model that was generated by finding a certain rule between the dwell time and movement speed of the nozzle 7.

[0223] Alternatively, a symbolic regression may be performed using the second learning data to obtain a mathematical formula that inputs the predicted dwell time information 43 and outputs the target nozzle speed information RP1.

[0224] (Third Modification) A third modified example of this embodiment will be described with reference to Figures 14 and 17. The third modified example differs from the embodiment described above with reference to Figures 1 to 19 mainly in that the first learning data TND includes third learning information 79. Below, the differences between the third modified example and the embodiment will be mainly described.

[0225] As shown in FIG. 17, in the third modification, the first learning data TND includes third learning information 79.

[0226] Specifically, in the learning data creating device 300, the learning data creating unit 61 acquires the third learning information 79 from the substrate processing apparatus 100. Then, the learning data creating unit 61 creates the first learning data TND using the first learning information 74, the second learning information 75, and the third learning information 79.

[0227] The third learning information 79 includes at least one of temperature information indicating the temperature of the processing liquid used to process the substrate W, concentration information indicating the concentration of the processing liquid used to process the substrate W, flow rate information indicating the flow rate of the processing liquid used to process the substrate W, and rotation speed information indicating the rotation speed of the substrate W. The third learning information 79 is an explanatory variable.

[0228] Then, as shown in Figure 14, in the learning device 200, the learning unit 72 generates a first learned model LM2 by machine learning the first learning data TND including the first learning information 74, the second learning information 75, and the third learning information 79.

[0229] The temperature information, concentration information, flow rate information, and rotation speed information are correlated with the processing amount of the substrate W. Therefore, according to the third modified example, by including at least one of the temperature information, concentration information, flow rate information, and rotation speed information in the first learning data TND, it is possible to create a first trained model LM2 that can more accurately predict the dwell time of the nozzle 7 that can achieve the target processing amount 20a.

[0230] In the third modified example, the target processing volume information 41 (FIG. 9) input to the first trained model LM2 includes accompanying information. The accompanying information includes at least one of temperature information indicating the temperature of the processing liquid for processing the substrate W, concentration information indicating the concentration of the processing liquid for processing the substrate W, flow rate information indicating the flow rate of the processing liquid for processing the substrate W, and rotation speed information indicating the rotation speed of the substrate W.

[0231] Here, for example, the temperature information, concentration information, flow rate information, and rotation speed information can use the set values ​​for the substrate processing unit 1. In other words, the temperature information, concentration information, flow rate information, and rotation speed information included in the processing condition information RP2 (FIG. 4) can be used.

[0232] Alternatively, for example, the temperature information may be temperature information output by temperature sensor 21F (FIG. 4). For example, the concentration information may be concentration information output by concentration sensor 21G. For example, the flow rate information may be flow rate information output by flow rate sensor 21H. For example, the rotation speed information may be rotation speed information output by rotation sensor 21K.

[0233] The third learning information 79 may further include at least one of information on a pattern collapse rate and a particle removal rate. The pattern collapse rate indicates the collapse rate of multiple structures constituting a pattern on the surface of the substrate W. The particle removal rate indicates the removal rate of particles when particles are attached to the substrate W and substrate cleaning is performed.

[0234] In this case, the accompanying information included in the target processing amount information 41 further includes at least one of information on a pattern collapse rate and a particle removal rate.

[0235] Next, the present invention will be described in detail based on examples, but the present invention is not limited to the following examples. [Example]

[0236] An example of the present invention will be described with reference to Fig. 23. In the example, a strong correlation between the throughput of the substrate W and the residence time of the nozzle 7 was demonstrated.

[0237] Fig. 23 is a diagram showing the residence time 205 of the nozzle 7 and the throughput 220 of the substrate W according to an embodiment of the present invention. In Fig. 23, the horizontal axis represents the position on the substrate W. The position of the substrate center OP is indicated by 0 mm, and one end in the radial direction of the substrate W is indicated by 150 mm. The vertical axis on the left side of the figure represents the throughput 220 of the target TG constituting the substrate W. The vertical axis on the right side of the figure represents the residence time 205 of the nozzle 7. As shown in Fig. 23, there was a strong correlation between the throughput 220 (curve) and the residence time 205 (vertical bar).

[0238] In the example, a wafer with a diameter of 300 mm was used. The processing liquid was hydrofluoric nitric acid (a mixture of hydrofluoric acid (HF) and nitric acid (HNO3)). The processing liquid temperature was 30°C, the processing liquid concentration was 1 (HF):9 (HNO3), the processing liquid flow rate was 41 ml / min, and the rotation speed of the substrate W was 350 rpm. The number of reciprocating movements in the scanning process was 16. The nozzle movement speed was as shown in the table below.

[0239] [Table 1]

[0240] [Table 2]

[0241] The embodiments of the present invention have been described above with reference to the drawings. However, the present invention is not limited to the above embodiments and can be implemented in various forms without departing from the spirit of the present invention. Furthermore, the components disclosed in the above embodiments can be modified as appropriate. For example, some of the components shown in one embodiment may be added to the components of another embodiment, or some of the components shown in one embodiment may be deleted from the embodiment.

[0242] Furthermore, the drawings mainly show each component in a schematic manner to facilitate understanding of the invention, and the thickness, length, number, spacing, etc. of each component shown in the drawings may differ from the actual ones due to the convenience of creating the drawings. Furthermore, the configuration of each component shown in the above embodiment is merely an example and is not particularly limited, and it goes without saying that various modifications are possible within a range that does not substantially deviate from the effects of the present invention.

[0243] In the embodiment (including the modified example) described with reference to FIGS. 1 to 22, the entirety of the multiple movement sections dn is treated as one scan section. However, multiple scan sections may be provided, and a movement section dn may be assigned to each scan section. For example, when three scan sections are provided, movement sections d1 to d10 are assigned to the first scan section, movement sections d11 to d20 are assigned to the second scan section, and movement sections d21 to d30 are assigned to the third scan section. Then, first training data TND is prepared for each scan section, and a first trained model LM1 is created for each scan section. Then, target processing volume information 41 is also prepared for each scan section. Therefore, the target processing volume information 41 for each scan section is input to the first trained model LM1 for each scan section, and predicted dwell time information 43 is output for each scan section. As a result, predicted dwell time information 43 for each scan section is also input to the speed acquisition unit 213 or the second trained model LM2. Therefore, target nozzle speed information RP1 is output for each scan section. In this case, the optimum target nozzle speed information RP1 can be acquired for each scan section. [Industrial Applicability]

[0244] The present invention relates to a substrate processing method, a trained model generation method, a nozzle speed acquisition method, a training data creation method, a substrate processing apparatus, a trained model, a nozzle speed acquisition device, and a computer program, and has industrial applicability. [Explanation of symbols]

[0245] 1. Substrate processing section 7 Nozzle (processing nozzle) 21B Storage section 212 Stay Time Acquisition Unit 213 Speed ​​acquisition part LM1 First trained model (trained model) LM2 Second trained model W substrate (substrate to be processed)

Claims

1. inputting target processing amount information into a first trained model constructed by machine learning the first training data; A step of acquiring predicted stay time information output by the first trained model; processing the predicted dwell time information to obtain target nozzle velocity information; processing the substrate to be processed in accordance with the target nozzle speed information; Including, the target nozzle speed information indicates a target value of the movement speed of the processing nozzle set for each of a plurality of processing nozzle positions that divides a movement range of the processing nozzle that moves while discharging the processing liquid onto the processing target substrate into a plurality of movement sections; the target processing amount information indicates a target value of the processing amount of the substrate to be processed by the processing liquid at each of a plurality of positions in a radial direction of the substrate to be processed, the predicted residence time information indicates a predicted value of the residence time of the processing nozzle in each of the plurality of movement sections, the first learning data includes first learning information and second learning information; the first learning information is an explanatory variable indicating a treatment amount of the substrate with the treatment liquid at each of a plurality of positions in a radial direction of the substrate, A substrate processing method, wherein the second learning information is a response variable indicating the dwell time of the nozzle in each of a plurality of movement sections that divide the movement range of the nozzle as it moves while ejecting the processing liquid onto the substrate.

2. acquiring first learning data including first learning information and second learning information; generating a first trained model that inputs target processing amount information and outputs predicted stay time information by machine learning the first training data; Including, the first learning information is an explanatory variable indicating a treatment amount of the substrate by the treatment liquid at each of a plurality of positions in a radial direction of the substrate, the second learning information is a response variable indicating a residence time of the nozzle in each of a plurality of movement sections that divide a movement range of the nozzle that moves while discharging the processing liquid onto the substrate, the target processing amount information indicates a target value of the processing amount of the substrate to be processed by the processing liquid at each of a plurality of positions in a radial direction of the substrate to be processed, A trained model generation method, in which the predicted residence time information indicates a predicted value of the residence time of the processing nozzle in each of a plurality of movement sections that divide the movement range of the processing nozzle as it moves while ejecting the processing liquid onto the substrate to be processed.

3. the first learning information is corrected information based on a comparison result between first reference processing amount information indicating a processing result of a first reference substrate and second reference processing amount information indicating a processing result of a second reference substrate processed at a timing different from that of the first reference substrate, the first reference processing amount information indicates a processing amount of the first reference substrate by the processing liquid at each of a plurality of positions in a radial direction of the first reference substrate when the first reference substrate is processed by the processing liquid under reference processing conditions; 3. The trained model generation method according to claim 2, wherein the second reference processing amount information indicates a processing amount of the second reference substrate by the processing liquid at each of a plurality of radial positions of the second reference substrate when the second reference substrate is processed by the processing liquid according to the reference processing conditions.

4. the second reference processing amount information is information generated on the same day as the day on which original learning information, which is information before correcting the first learning information, was generated; The trained model generation method according to claim 3 , wherein the first reference processing amount information is information generated earlier than the date on which the second reference processing amount information is acquired.

5. the first learning data includes third learning information, 4. The trained model generating method according to claim 2, wherein the third learning information includes at least one of temperature information indicating a temperature of the processing liquid used to process the substrate, concentration information indicating a concentration of the processing liquid used to process the substrate, flow rate information indicating a flow rate of the processing liquid used to process the substrate, and rotation speed information indicating a rotation speed of the substrate.

6. acquiring predicted stay time information; processing the predicted dwell time information to obtain target nozzle velocity information; Including, the target nozzle speed information indicates a target value of the movement speed of the processing nozzle set for each of a plurality of processing nozzle positions that divides a movement range of the processing nozzle that moves while discharging the processing liquid onto the substrate to be processed into a plurality of movement sections; A nozzle speed acquisition method, wherein the predicted residence time information indicates a predicted value of the residence time of the processing nozzle in each of the plurality of movement sections.

7. When the processing nozzle ejects the processing liquid while moving back and forth, the residence time indicates a cumulative total of time that the processing nozzle is positioned in each of the plurality of movement sections, 7. The nozzle speed acquisition method of claim 6, wherein in the process of acquiring the target nozzle speed information, the target nozzle speed information is calculated based on the number of reciprocating movements of the processing nozzle, the predicted value of the dwell time, and the length of the movement section in the radial direction of the substrate to be processed.

8. In the step of acquiring the target nozzle speed information, the predicted dwell time information is input to a second trained model constructed by machine learning second training data, and the target nozzle speed information is acquired from the second trained model; the second learning data includes fourth learning information and fifth learning information, the fourth learning information is an explanatory variable indicating a residence time of the nozzle in each of a plurality of movement sections that divide a movement range of the nozzle that moves while discharging the processing liquid onto the substrate, The nozzle speed acquisition method according to claim 6 , wherein the fifth learning information is a response variable indicating the nozzle movement speed set for each of a plurality of nozzle positions that divide the movement range into the plurality of movement sections.

9. A learning data creation method for creating first learning data that is a target of machine learning for generating a first trained model that inputs target processing amount information and outputs predicted stay time information, comprising: obtaining original learning information; a step of correcting the original learning information based on a comparison result between first reference processing amount information indicating the processing result of a first reference substrate and second reference processing amount information indicating the processing result of a second reference substrate processed at a timing different from that of the first reference substrate, and outputting the corrected original learning information as first learning information; acquiring second learning information; Including, the target processing amount information indicates a target value of the processing amount of the substrate to be processed by the processing liquid at each of a plurality of positions in a radial direction of the substrate to be processed; the predicted residence time information indicates a predicted value of a residence time of the processing nozzle in each of a plurality of movement sections that divide a movement range of the processing nozzle that moves while discharging the processing liquid onto the processing target substrate, the original learning information indicates the amount of treatment of the substrate by the treatment liquid at each of a plurality of positions in the radial direction of the substrate; the first learning information is an explanatory variable, the second learning information is a response variable indicating a residence time of the nozzle in each of a plurality of movement sections that divide a movement range of the nozzle that moves while discharging the processing liquid onto the substrate, the first learning information and the second learning information constitute the first learning data, the first reference processing amount information indicates a processing amount of the first reference substrate by the processing liquid at each of a plurality of positions in a radial direction of the first reference substrate when the first reference substrate is processed by the processing liquid under reference processing conditions; A learning data creation method, wherein the second reference processing amount information indicates the processing amount of the second reference substrate by the processing liquid at each of multiple radial positions of the second reference substrate when the second reference substrate is processed by the processing liquid according to the reference processing conditions.

10. a storage unit that stores a first trained model constructed by machine learning the first training data; a stay time acquisition unit that inputs target processing amount information to the first trained model and acquires predicted stay time information from the first trained model; a speed acquisition unit that processes the predicted residence time information to acquire target nozzle speed information; a substrate processing unit that processes a substrate to be processed in accordance with the target nozzle speed information; Equipped with the target nozzle speed information indicates a target value of the movement speed of the processing nozzle set for each of a plurality of processing nozzle positions that divides a movement range of the processing nozzle that moves while discharging the processing liquid onto the processing target substrate into a plurality of movement sections; the target processing amount information indicates a target value of the processing amount of the substrate to be processed by the processing liquid at each of a plurality of positions in a radial direction of the substrate to be processed, the predicted residence time information indicates a predicted value of the residence time of the processing nozzle in each of the plurality of movement sections, the first learning data includes first learning information and second learning information; the first learning information is an explanatory variable indicating a treatment amount of the substrate with the treatment liquid at each of a plurality of positions in a radial direction of the substrate, The second learning information is a target variable indicating the residence time of the nozzle in each of a plurality of movement sections that divide the movement range of the nozzle as it moves while ejecting the processing liquid onto the substrate.

11. A trained model constructed by machine learning first training data, which causes a computer to input target throughput information and output predicted stay time information, the target processing amount information indicates a target value of the processing amount of the substrate to be processed by the processing liquid at each of a plurality of positions in a radial direction of the substrate to be processed; the predicted residence time information indicates a predicted value of a residence time of the processing nozzle in each of a plurality of movement sections that divide a movement range of the processing nozzle that moves while discharging the processing liquid onto the processing target substrate, the first learning data includes first learning information and second learning information; the first learning information is an explanatory variable indicating a treatment amount of the substrate with the treatment liquid at each of a plurality of positions in a radial direction of the substrate, The second learning information is a trained model, which is a target variable indicating the dwell time of the nozzle in each of a plurality of movement sections that divide the movement range of the nozzle as it moves while ejecting the processing liquid onto the substrate.

12. a stay time acquisition unit that acquires predicted stay time information; a speed acquisition unit that acquires target nozzle speed information by processing the predicted residence time information; Equipped with the target nozzle speed information indicates a target value of the movement speed of the processing nozzle set for each of a plurality of processing nozzle positions that divides a movement range of the processing nozzle that moves while discharging the processing liquid onto the substrate to be processed into a plurality of movement sections; The predicted residence time information indicates a predicted value of the residence time of the processing nozzle in each of the plurality of movement sections.

13. A computer program for causing a computer to execute the trained model generation method according to claim 2 or 3.

14. A computer program for causing a computer to execute the nozzle velocity acquisition method according to claim 6 or 7.

Images