Method for generating training data and method for generating trained model

Abstract

To provide a data generation method for learning and a learned model generation method, capable of reducing a load of an operator.SOLUTION: A substrate processing device 100 comprises: a storage part 103; and a control part 102. The storage part 103 stores a learned model 136. The learned model 136 is generated by learning a learning data containing a processing amount to be acquired by executing a processing to a substrate W of a learning object. The control part 102 acquires the processing amount by executing the processing by supplying a processing liquid to the substrate on the basis of an output of the learned model 136, and is used as a learning data for an additional learning in the case where the processing amount is not matched with the target processing amount within a permission range.SELECTED DRAWING: Figure 16

Description

[Technical Field]

[0001] The present invention relates to a method for generating training data and a method for generating a trained model. [Background technology]

[0002] There are known substrate processing apparatuses that process wafers having a coating formed on their surfaces and adjust the thickness of the coating and remove foreign matter by liquid processing. One type of such substrate processing apparatus is a single-wafer type substrate processing apparatus equipped with a nozzle that supplies an etching processing liquid to the wafer surface (see, for example, Patent Document 1). The substrate processing apparatus of Patent Document 1 ejects the etching processing liquid from the nozzle while moving the nozzle based on a predetermined speed profile during the etching process. [Prior art documents] [Patent documents]

[0003] [Patent Document 1] Japanese Patent Application Publication No. 2019-54104 Summary of the Invention [Problem to be solved by the invention]

[0004] However, in a substrate processing apparatus that moves a nozzle based on a predetermined velocity profile, it is necessary to redevelop the velocity profile in accordance with, for example, a change in pre-processing performed on a wafer. This is because a change in pre-processing changes the thickness of the coating film before the etching process and the variation in the thickness of the coating film before the etching process. This places a burden on the workers involved in developing the velocity profile.

[0005] One embodiment of the present invention provides a method for generating training data and a method for generating a trained model that can reduce the burden on workers. [Means for solving the problem]

[0006] One embodiment of the present invention provides a method for generating training data for additional learning using a substrate processing apparatus that supplies a processing liquid to a substrate and processes the substrate based on the output of a trained model generated by training training data. The method includes the steps of: calculating a difference between a thickness distribution of the substrate before processing and a target thickness distribution of the substrate to obtain a target processing amount that is a target amount of processing amount indicating an amount of the substrate to be processed by the processing; performing the processing on the substrate to obtain the processing amount; determining whether the processing amount and the target processing amount match within an allowable range; and, if it is determined that the processing amount and the target processing amount do not match within the allowable range, assigning a flag to the processing amount indicating use of the processing amount as training data for the additional learning.

[0007] One embodiment of the present invention provides a method for generating a trained model for use in a substrate processing apparatus that supplies a processing liquid to a substrate to be processed and performs the processing on the substrate based on an output of a trained model, the method including the steps of: supplying the processing liquid to the substrate to be processed and performing the processing on the substrate to be processed, acquiring a processing amount indicating an amount of the substrate to be processed by the processing, generating training data including the processing amount of the substrate to be processed, and a learning step of classifying the training data into classes based on predetermined classification rules and learning the training data for each class to generate the trained model.

[0008] In one embodiment, the classification rule may stipulate that the learning data is classified into each of the classes according to the number of processed substrates that are the learning target.

[0009] In one embodiment, in the step of performing the process on the learning substrate, a learning lot number, which is the number of lots since the start of use of the processing liquid used in the process, is obtained, and the classification rule may specify that the learning data is classified into each of the classes according to the learning lot number.

[0010] In one example, a substrate processing apparatus supplies a processing liquid to a substrate to process the substrate. The substrate processing apparatus includes a nozzle, a movement mechanism, a memory unit, and a control unit. The nozzle supplies the processing liquid to the substrate. The movement mechanism moves at least one of the nozzle and the substrate. The memory unit stores a trained model. The control unit controls the movement mechanism using the trained model. The trained model is generated by learning, as training data, learning target speed information indicating at least one of the movement speed of the nozzle and the movement speed of the training target substrate, or indicating the relative movement speed between the nozzle and the training target substrate, and a processing amount obtained by performing the processing on the training target substrate while moving at least one of the nozzle and the training target substrate at a speed based on the learning target speed information. The processing amount indicates the amount of the substrate processed by the processing. The control unit inputs the target amount of processing amount into the trained model, causing the trained model to output processing speed information. The control unit controls the movement mechanism so that at least one of the nozzle and the substrate to be processed moves at a speed based on the processing speed information when performing the processing on the substrate to be processed, The processing speed information indicates at least one of the movement speed of the nozzle and the movement speed of the substrate to be processed, or indicates the relative movement speed between the nozzle and the substrate to be processed.

[0011] In one example, the learning target speed information indicates at least one of a nozzle movement speed set for each nozzle position that divides a movement section in which the nozzle moves into a plurality of sections, and a movement speed of the learning target substrate set for each substrate position that divides a movement section in which the learning target substrate moves into a plurality of sections, or indicates a relative movement speed between the nozzle and the learning target substrate for each nozzle position or each substrate position. The processing time speed information indicates at least one of a nozzle movement speed set for each nozzle position and a movement speed of the processing target substrate set for each substrate position, or indicates a relative movement speed between the nozzle and the processing target substrate for each nozzle position or each substrate position.

[0012] In one example, the substrate processing apparatus further includes a substrate holding unit and a substrate rotating unit. The substrate holding unit holds the substrate horizontally. The substrate rotating unit rotates the substrate and the substrate holding unit together around a central axis extending in the vertical direction. The relative movement speed indicates the relative movement speed between the rotating surface of the substrate and the nozzle.

[0013] In one example, the control unit learns the processing speed information and the processing amount obtained by performing the processing on the substrate to be processed while moving at least one of the nozzle and the substrate to be processed at a speed based on the processing speed information as learning data for additional learning, and generates a trained model after additional learning.

[0014] In one example, the control unit determines whether the processing capacity is available for the additional learning.

[0015] In one example, the substrate processing apparatus further includes a measuring instrument. The measuring instrument measures a value of a measurement target. The storage unit stores a recipe for controlling execution of the process. The recipe indicates a set value for the measurement target. When a difference between the value of the measurement target and the set value is equal to or greater than a specified value, the control unit issues a different error notification depending on whether the processing amount is available for the additional learning.

[0016] In one example, the substrate processing apparatus further includes a processing amount detection unit for detecting the processing amount.

[0017] In one example, the memory unit stores the learning target speed information. The control unit controls the movement mechanism so that at least one of the nozzle and the learning target substrate moves at a speed based on the learning target speed information, and supplies the processing liquid from the nozzle to the learning target substrate, thereby performing the processing on the learning target substrate. The control unit generates the learning data including the processing amount acquired by performing the processing on the learning target substrate and the learning target speed information. The control unit classifies the learning data into classes based on predetermined classification rules. The control unit learns the learning data for each class to generate the trained model. When performing the processing on the processing target substrate, the control unit selects one of the classes based on predetermined selection rules. The control unit inputs the target amount into the trained model corresponding to the selected class, thereby outputting the processing speed information from the trained model.

[0018] In one example, the classification rules stipulate that the learning data be classified into each of the classes according to the number of processed substrates for the learning target. The number of processed substrates for the learning target indicates the number of processed substrates for the learning target substrate since the start of use of the processing liquid used for the processing. The selection rules stipulate that one of the classes be selected according to the number of processed substrates for the processing target. The number of processed substrates for the processing target indicates the number of processed substrates for the processing target substrate since the start of use of the processing liquid used for the processing.

[0019] In one example, the control unit acquires a learning lot number when performing the process on the substrate to be learned. The learning lot number is the number of lots since the start of use of the processing liquid used in the process. The classification rule specifies that the learning data is classified into each of the classes according to the learning lot number. The control unit acquires a processing lot number when performing the process on the substrate to be learned. The processing lot number is the number of lots since the start of use of the processing liquid used in the process. The selection rule specifies that one of the classes is selected according to the processing lot number.

[0020] In one example, the substrate processing apparatus further includes an input unit. The input unit inputs, for each class, an instruction to select reference training data from the training data included in each of the classes. The control unit selects, for each class, the training data from the training data included in each of the classes, whose correlation with the reference training data satisfies a predetermined condition. The control unit trains, for each class, the selected training data to generate the trained model.

[0021] This specification also discloses a substrate processing method. In one example, the substrate processing method is a method of supplying a processing liquid from a nozzle to a substrate to be processed to perform the processing on the substrate. The substrate processing method includes the steps of: acquiring a target processing amount indicating an amount of the substrate to be processed by the processing; inputting the target amount into a trained model to output processing speed information from the trained model; and supplying the processing liquid to the substrate to be processed while moving at least one of the nozzle and the substrate to be processed at a speed based on the processing speed information. The trained model is generated by learning, as training data, learning target speed information indicating at least one of the movement speed of the nozzle and the movement speed of the substrate to be processed, or learning target speed information indicating the relative movement speed between the nozzle and the substrate to be processed, and the processing amount obtained by performing the processing on the substrate to be processed while moving at least one of the nozzle and the substrate to be processed at a speed based on the learning target speed information. The processing speed information indicates at least one of the movement speed of the nozzle and the movement speed of the substrate to be processed, or indicates the relative movement speed between the nozzle and the substrate to be processed.

[0022] In one example, the learning target speed information indicates at least one of a nozzle movement speed set for each nozzle position that divides a movement section in which the nozzle moves into a plurality of sections, and a movement speed of the learning target substrate set for each substrate position that divides a movement section in which the learning target substrate moves into a plurality of sections, or indicates a relative movement speed between the nozzle and the learning target substrate for each nozzle position or each substrate position. The processing time speed information indicates at least one of a nozzle movement speed set for each nozzle position and a movement speed of the processing target substrate set for each substrate position, or indicates a relative movement speed between the nozzle and the processing target substrate for each nozzle position or each substrate position.

[0023] In one example, in the processing step, the substrate to be processed is rotated around a central axis extending in the vertical direction while being held horizontally. The relative movement speed between the nozzle and the learning substrate indicates the relative movement speed between the nozzle and the surface of the rotating learning substrate. The relative movement speed between the nozzle and the processing substrate indicates the relative movement speed between the nozzle and the surface of the rotating processing substrate.

[0024] In one example, the substrate processing method further includes a learning data generation step of generating learning data for additional learning based on the processing speed information output in the output step and the processing volume of the substrate to be processed processed in the processing step, and a step of learning the learning data for additional learning to generate a trained model after additional learning.

[0025] In one example, the step of generating data for learning includes the step of determining whether the amount of processing is available for the additional learning.

[0026] In one example, the substrate processing method further includes a step of determining whether a difference between a value of the measurement object and a predetermined set value for the measurement object is equal to or greater than a specified value, and if the difference between the value of the measurement object and the set value is equal to or greater than the specified value, executing an error notification that differs depending on whether the processing amount is available for the additional learning.

[0027] In one example, the substrate processing method further includes the steps of: supplying the processing liquid to the learning substrate while moving at least one of the nozzle and the learning substrate at a speed based on the learning target speed information, and performing the processing on the learning target substrate; acquiring the processing amount of the learning target substrate; generating the learning data including the processing amount of the learning target substrate and the learning target speed information; and classifying the learning data into classes based on predetermined classification rules and learning the learning data for each class to generate the trained model. In the output step, one of the classes is selected based on predetermined selection rules, and the target amount is input to the trained model corresponding to the selected class, thereby outputting the processing speed information from the trained model.

[0028] In one example, the classification rules stipulate that the learning data be classified into each of the classes according to the number of processed substrates for the learning target. The number of processed substrates for the learning target indicates the number of processed substrates for the learning target substrate since the start of use of the processing liquid used for the processing. The selection rules stipulate that one of the classes be selected according to the number of processed substrates for the processing target. The number of processed substrates for the processing target indicates the number of processed substrates for the processing target substrate since the start of use of the processing liquid used for the processing.

[0029] In one example, in the step of performing the process on the substrate to be learned, a learning lot number is obtained. The learning lot number is the number of lots since the start of use of the processing liquid used in the process. The classification rule specifies that the learning data is classified into each of the classes according to the learning lot number. In the step of performing the process on the substrate to be learned, a processing lot number is obtained. The processing lot number is the number of lots since the start of use of the processing liquid used in the process. The selection rule specifies that one of the classes is selected according to the processing lot number.

[0030] In one example, the substrate processing method further includes the steps of: selecting, for each class, reference training data from the training data included in each of the classes; and selecting, for each class, the training data from the training data included in each of the classes, the training data whose correlation with the reference training data satisfies a predetermined condition. In the training step, the selected training data is trained for each class to generate the trained model.

[0031] This specification also discloses a substrate processing system. In one example, the substrate processing system includes a substrate processing apparatus and an information processing apparatus. The substrate processing apparatus supplies a processing liquid to a substrate to process the substrate. The information processing apparatus transmits processing speed information to the substrate processing apparatus. The information processing apparatus includes a communication unit, a memory unit, and a processing unit. The communication unit of the information processing apparatus receives a target amount of processing of the substrate to be processed by the processing from the substrate processing apparatus. The communication unit of the information processing apparatus transmits the processing speed information to the substrate processing apparatus. The memory unit stores a trained model. The processing unit inputs the target amount acquired from the substrate processing apparatus to the trained model via the communication unit, thereby outputting the processing speed information from the trained model. The substrate processing apparatus includes a communication unit, a nozzle, a movement mechanism, and a control unit. The communication unit of the substrate processing apparatus transmits the target amount to the information processing apparatus. The communication unit of the substrate processing apparatus receives the processing speed information from the information processing apparatus. The nozzle supplies the processing liquid to the substrate. The movement mechanism moves at least one of the nozzle and the substrate. The control unit controls the movement mechanism based on the processing speed information. The trained model is generated by learning, as learning data, learning object speed information indicating at least one of the movement speed of the nozzle and the movement speed of the training target substrate, or the relative movement speed between the nozzle and the training target substrate, and a processing amount acquired by performing the process on the training target substrate while moving at least one of the nozzle and the training target substrate at a speed based on the learning object speed information. The processing amount indicates the amount of the training target substrate processed by the process. The control unit controls the movement mechanism so that at least one of the nozzle and the training target substrate moves at a speed based on the processing speed information when performing the process on the training target substrate. The processing speed information indicates at least one of the movement speed of the nozzle and the training target substrate, or the relative movement speed between the nozzle and the training target substrate.

[0032] This specification also discloses a learning device. In one example, the learning device includes a memory unit and a learning unit. The memory unit stores a learning program. The learning unit learns learning data based on the learning program. The learning unit learns, as the learning data, speed information indicating at least one of a nozzle movement speed and a substrate movement speed, or a relative movement speed between the nozzle and the substrate, and a processing amount obtained by supplying processing liquid from the nozzle to the substrate while moving at least one of the nozzle and the substrate at a speed based on the speed information. The processing amount indicates the amount of processing of the substrate by the supply of the processing liquid.

[0033] This specification also discloses a learning method. In one example, the learning method includes steps of acquiring learning data and learning the learning data based on a learning program. The learning data indicates speed information and a processing amount obtained by supplying processing liquid from the nozzle to the substrate while moving at least one of the nozzle and the substrate at a speed based on the speed information. The speed information indicates at least one of the movement speed of the nozzle and the movement speed of the substrate, or indicates the relative movement speed between the nozzle and the substrate. The processing amount indicates the amount of processing of the substrate by the supply of the processing liquid.

[0034] This specification also discloses a method for generating a trained model. In one example, the method for generating a trained model includes the steps of acquiring training data and learning the training data based on a training program to generate a trained model. The training data indicates speed information and a processing amount obtained by supplying a processing liquid from the nozzle to the substrate while moving at least one of the nozzle and the substrate at a speed based on the speed information. The speed information indicates at least one of the movement speed of the nozzle and the movement speed of the substrate, or indicates the relative movement speed between the nozzle and the substrate. The processing amount indicates the amount of processing of the substrate by supplying the processing liquid.

[0035] This specification also discloses a method for generating training data. In one example, the method for generating training data includes the steps of: supplying a processing solution to a substrate from a nozzle while moving at least one of a nozzle and a substrate at a speed based on speed information to perform a process on the substrate; acquiring a processing amount indicating an amount of the substrate processed by the process; and generating training data based on the speed information and the processing amount. This method may also include the steps of determining whether the training data can be used for machine learning based on a difference between the processing amount and a predetermined value; assigning a first flag to the training data if it is determined that the training data can be used for machine learning; and assigning a second flag to the training data if it is determined that the training data cannot be used for machine learning. The speed information indicates at least one of a movement speed of the nozzle and a movement speed of the substrate, or a relative movement speed between the nozzle and the substrate.

[0036] In one example, the step of processing the substrate is performed by a substrate processing apparatus. The method for generating learning data further includes the steps of determining whether an abnormal state has occurred in the substrate processing apparatus, and issuing an error notification if it is determined that the abnormal state has occurred. If the first flag is assigned to the learning data, the error notification indicates that the abnormal state has occurred in the substrate processing apparatus. If the second flag is assigned to the learning data, the error notification indicates that the processing amount is equal to or greater than the specified value and that the abnormal state has occurred in the substrate processing apparatus. [Effects of the Invention]

[0037] The learning data generation method and trained model generation method according to the present invention can reduce the burden on workers. [Brief explanation of the drawings]

[0038] [Figure 1]1 is a schematic view of a substrate processing apparatus according to a first embodiment of the present invention. [Figure 2] FIG. 1 is a schematic diagram of a processing unit according to a first embodiment of the present invention. [Figure 3] FIG. 2 is a plan view illustrating a scanning process according to the first embodiment of the present invention. [Figure 4] FIG. 4 is a diagram showing scan speed information according to the first embodiment of the present invention. [Figure 5] 4 is a graph showing an example of the moving speed of a first nozzle according to the first embodiment of the present invention. [Figure 6] FIG. 2 is a plan view showing a thickness measurement process in the first embodiment of the present invention. [Figure 7] FIG. 2 is a schematic view of an etching liquid supply unit according to the first embodiment of the present invention. [Figure 8] FIG. 2 is a block diagram of a control device according to the first embodiment of the present invention. [Figure 9] 3 is a flowchart showing a method for generating learning data according to the first embodiment of the present invention. [Figure 10] 1 is a flowchart showing an etching process according to a first embodiment of the present invention. [Figure 11] 4 is a flowchart showing a process for generating learning data according to the first embodiment of the present invention. [Figure 12] 4 is a flowchart showing an abnormality determination process according to the first embodiment of the present invention. [Figure 13] FIG. 2 is a diagram showing an example of a training dataset table according to the first embodiment of the present invention. [Figure 14] FIG. 2 is a block diagram of a control device according to the first embodiment of the present invention. [Figure 15] 1 is a flowchart showing a method for generating a trained model in embodiment 1 of the present invention. [Figure 16] FIG. 2 is a block diagram of a control device according to the first embodiment of the present invention. [Figure 17] 1 is a flowchart showing a substrate processing method according to a first embodiment of the present invention. [Figure 18]1 is a flowchart showing a substrate processing method according to a first embodiment of the present invention. [Figure 19] FIG. 2 is a diagram showing an example of a table for additional learning datasets according to the first embodiment of the present invention. [Figure 20] FIG. 10 is a diagram showing a substrate processing apparatus and a learning apparatus according to a second embodiment of the present invention. [Figure 21] 10 is a flowchart showing a method for generating a trained model in embodiment 2 of the present invention. [Figure 22] FIG. 10 is a diagram showing a substrate processing system according to a third embodiment of the present invention. [Figure 23] 4 is a flowchart showing a process executed by a control device of the substrate processing apparatus and a process executed by a processing unit of the learning apparatus. [Figure 24] FIG. 10 is a schematic diagram of a processing unit according to a fourth embodiment of the present invention. [Figure 25] FIG. 10 is a schematic view showing the configuration of a substrate processing apparatus according to a fourth embodiment of the present invention. [Figure 26] 10 is a flowchart showing a method for generating learning data according to a fourth embodiment of the present invention. [Figure 27] 10 is a flowchart showing a process for generating learning data according to a fourth embodiment of the present invention. [Figure 28] FIG. 13 is a diagram showing an example of a training dataset table according to the fourth embodiment of the present invention. [Figure 29] 10 is a flowchart showing a method for generating a trained model in embodiment 4 of the present invention. [Figure 30] 10 is a flowchart showing a part of a substrate processing method according to a fourth embodiment of the present invention. [Figure 31] 10 is a flowchart showing a process for acquiring processing speed information according to the fourth embodiment of the present invention. [Figure 32] 10 is a flowchart showing a method for generating learning data according to a fifth embodiment of the present invention. [Figure 33] 13 is a flowchart showing an example of a counting process according to the fifth embodiment of the present invention. [Figure 34]FIG. 13 is a diagram showing an example of a training dataset table according to the fifth embodiment of the present invention. [Figure 35] 10 is a flowchart showing a part of a substrate processing method according to a fifth embodiment of the present invention. DETAILED DESCRIPTION OF THE INVENTION

[0039] Hereinafter, embodiments of the present invention will be described with reference to the drawings. However, the present invention is not limited to the following embodiments, and can be implemented in various forms without departing from the spirit of the present invention. Note that where explanations are repeated, they may be omitted as appropriate. In addition, in the drawings, the same or equivalent parts are designated by the same reference numerals, and explanations will not be repeated.

[0040] [Embodiment 1] A substrate processing apparatus 100 according to this embodiment will be described with reference to Fig. 1. Fig. 1 is a schematic diagram of the substrate processing apparatus 100 according to this embodiment. More specifically, Fig. 1 is a schematic plan view of the substrate processing apparatus 100. The substrate processing apparatus 100 is a single-wafer processing apparatus that processes substrates W one by one. In this embodiment, the substrates W are semiconductor wafers. The substrates W are approximately disk-shaped.

[0041] As shown in FIG. 1, the substrate processing apparatus 100 includes a plurality of processing units 1, a fluid cabinet 100A, a plurality of fluid boxes 100B, a plurality of load ports LP, an indexer robot IR, a center robot CR, and a controller 101.

[0042] Each load port LP accommodates a stack of substrates W. The indexer robot IR transports the substrates W between the load port LP and the center robot CR. The center robot CR transports the substrates W between the indexer robot IR and the processing units 1. Each processing unit 1 supplies a processing liquid to the substrate W to process the substrate W. The fluid cabinet 100A accommodates the processing liquid.

[0043] The multiple processing units 1 form multiple towers TW (four towers TW in FIG. 1) arranged to surround the center robot CR in a plan view. Each tower TW includes multiple processing units 1 stacked vertically (three processing units 1 in FIG. 1). Each fluid box 100B corresponds to one of the multiple towers TW. The processing liquid in the fluid cabinet 100A is supplied to all processing units 1 included in the tower TW corresponding to the fluid box 100B via one of the fluid boxes 100B.

[0044] The controller 101 controls the operation of each part of the substrate processing apparatus 100. For example, the controller 101 controls the load port LP, the indexer robot IR, and the center robot CR.

[0045] In this embodiment, the control device 101 functions as a learning device. Specifically, the control device 101 executes machine learning. The control device 101 of this embodiment further executes additional learning. The machine learning and additional learning are, for example, any of supervised learning, unsupervised learning, semi-supervised learning, reinforcement learning, and deep learning.

[0046] Generally, machine learning is a concept that includes additional learning, but here, for convenience, the present embodiment will be described using the terms "machine learning" and "additional learning" interchangeably. In this embodiment, "machine learning" refers to machine learning performed before the manufacture of semiconductor products. In other words, "machine learning" refers to machine learning performed during the manufacturing stage of the substrate processing apparatus 100. On the other hand, "additional learning" refers to machine learning performed during the manufacturing stage of semiconductor products. In other words, "additional learning" refers to machine learning performed when adjusting the substrate processing apparatus 100.

[0047] Next, the processing unit 1 of this embodiment will be described with reference to Fig. 2. Fig. 2 is a schematic diagram of the processing unit 1 of this embodiment. More specifically, Fig. 2 is a schematic cross-sectional view of the processing unit 1.

[0048] The processing unit 1 processes the object constituting the substrate W with a processing liquid. Hereinafter, the object to be processed with the processing liquid will be referred to as "object TG." The object TG is, for example, a substrate body (e.g., a substrate body made of silicon) or a substance formed on the surface of the substrate body. The substance formed on the surface of the substrate body is, for example, a substance made of the same material as the substrate body (e.g., a layer made of silicon) or a substance made of a different material from the substrate body (e.g., a silicon oxide film, a silicon nitride film, or a resist). The "substance" may form a film.

[0049] In this embodiment, the processing liquid includes an etching liquid, and the processing unit 1 performs an etching process. The target TG is processed (etched) by the etching liquid. The etching liquid is a chemical liquid. The etching liquid is, for example, hydrofluoric nitric acid (a mixture of hydrofluoric acid (HF) and nitric acid (HNO3)), hydrofluoric acid, buffered hydrofluoric acid (BHF), ammonium fluoride, HFEG (a mixture of hydrofluoric acid and ethylene glycol), or phosphoric acid (H3PO4).

[0050] 2, the processing unit 1 includes a chamber 2, a spin chuck 3, a spin motor unit 5, a nozzle movement mechanism 6, a thickness measurement unit 8, a probe movement mechanism 9, a plurality of guards 10 (two guards 10 in FIG. 2), a first nozzle 41, and a second nozzle 71. The substrate processing apparatus 100 also includes an etching liquid supply unit 4 and a rinsing liquid supply unit 7. The etching liquid supply unit 4 has a first supply pipe 42, and the rinsing liquid supply unit 7 has a second supply pipe 72.

[0051] The chamber 2 has a generally box-like shape and accommodates the substrate W, the spin chuck 3, the spin motor unit 5, the nozzle movement mechanism 6, multiple guards 10, the thickness measurement unit 8, the probe movement mechanism 9, the first nozzle 41, the second nozzle 71, part of the first supply pipe 42, and part of the second supply pipe 72.

[0052] The spin chuck 3 holds the substrate W horizontally. The spin chuck 3 is an example of a substrate holding section. Specifically, the spin chuck 3 has 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.

[0053] The spin motor unit 5 rotates the substrate W and the spin chuck 3 together around a first rotation axis AX1. The first rotation axis AX1 extends in the up-down direction. In this embodiment, the first rotation axis AX1 extends in a substantially vertical direction. The first rotation axis AX1 is an example of a central axis, and the spin motor unit 5 is an example of a substrate rotation unit. More specifically, the spin motor unit 5 rotates the spin base 33 around the first rotation axis AX1. Therefore, the spin base 33 rotates around the first rotation axis AX1. As a result, the substrate W held by the multiple chuck members 32 provided on the spin base 33 rotates around the first rotation axis AX1.

[0054] Specifically, the spin motor unit 5 has a motor body 51, a shaft 53, and an encoder 55. The shaft 53 is coupled to the spin base 33. The motor body 51 rotates the shaft 53. As a result, the spin base 33 rotates.

[0055] The encoder 55 is an example of a measuring instrument that measures the value of a measurement target, and the measurement target of the encoder 55 is the rotational speed of the substrate W. The encoder 55 generates a signal that indicates the rotational speed of the substrate W. More specifically, the encoder 55 generates a rotational speed signal that indicates the rotational speed of the motor main body 51.

[0056] The first nozzle 41 supplies the etching liquid to the substrate W. More specifically, the first nozzle 41 ejects the etching liquid toward the rotating substrate W. The etching liquid supply unit 4 supplies the etching liquid to the first nozzle 41. More specifically, the first nozzle 41 is connected to one end of a first supply pipe 42. The etching liquid is supplied to the first nozzle 41 via the first supply pipe 42. The first supply pipe 42 is a tubular member through which the etching liquid flows.

[0057] The nozzle movement mechanism 6 moves the first nozzle 41. The nozzle movement mechanism 6 is an example of a movement mechanism. In this embodiment, the nozzle movement mechanism 6 moves the first nozzle 41 in a substantially horizontal direction. More specifically, the nozzle movement mechanism 6 rotates the first nozzle 41 around a second rotation axis AX2 that is aligned in a substantially vertical direction. The first nozzle 41 ejects the etching liquid toward the substrate W while moving (while rotating). The first nozzle 41 is sometimes referred to as a scan nozzle.

[0058] Specifically, the nozzle movement mechanism 6 has a nozzle arm 61, a first rotation shaft 63, and a first drive unit 65. The nozzle arm 61 extends in a substantially horizontal direction. The first nozzle 41 is disposed at the tip of the nozzle arm 61. The nozzle arm 61 is coupled to the first rotation shaft 63. The first rotation shaft 63 extends in a substantially vertical direction. The first drive unit 65 rotates the first rotation shaft 63 about the second rotation axis AX2, causing the nozzle arm 61 to rotate about the first rotation axis 63 along a substantially horizontal plane. As a result, the first nozzle 41 moves along the substantially horizontal plane. More specifically, the first nozzle 41 revolves around the first rotation shaft 63 about the second rotation axis AX2. The first drive unit 65 includes, for example, a stepping motor.

[0059] The second nozzle 71 supplies the rinse liquid to the substrate W. More specifically, the second nozzle 71 ejects the rinse liquid toward the rotating substrate W. The rinse liquid supply unit 7 supplies the rinse liquid to the second nozzle 71. More specifically, the rinse liquid is supplied to the second nozzle 71 via a second supply pipe 72. The second supply pipe 72 is a tubular member through which the rinse liquid flows. The rinse liquid is, for example, deionized water, carbonated water, electrolytic ionized water, hydrogen water, ozone water, or hydrochloric acid water with a diluted concentration (for example, about 10 ppm to 100 ppm). The second nozzle 71 ejects the rinse liquid in a stationary state. The second nozzle 71 is sometimes referred to as a fixed nozzle. The second nozzle 71 may be a scan nozzle.

[0060] Each of the guards 10 has a substantially cylindrical shape. The guards 10 receive the etching liquid and the rinsing liquid discharged from the substrate W.

[0061] The thickness measuring unit 8 measures the thickness of the object TG in a non-contact manner and generates a thickness detection signal indicating the thickness of the object TG. The thickness detection signal is input to the control device 101.

[0062] The thickness measurement unit 8 measures the thickness of the object TG by, for example, spectroscopic interferometry. Specifically, the thickness measurement unit 8 includes an optical probe 81, a signal line 83, and a thickness measurement device 85. The optical probe 81 has a lens. The signal line 83 connects the optical probe 81 and the thickness measurement device 85. The signal line 83 includes, for example, an optical fiber. The thickness measurement device 85 has a light source and a light receiving element. Light emitted from the light source of the thickness measurement device 85 is emitted to the object TG via the signal line 83 and the optical probe 81. Light reflected by the object TG is received by the light receiving element of the thickness measurement device 85 via the optical probe 81 and the signal line 83. The thickness measurement device 85 analyzes the light received by the light receiving element to calculate the thickness of the object TG. The thickness measurement device 85 generates a thickness detection signal indicating the calculated thickness of the object TG.

[0063] The probe movement mechanism 9 moves the optical probe 81 in a substantially horizontal direction. Specifically, the probe movement mechanism 9 rotates the optical probe 81 around a third rotation axis AX3 that is aligned in a substantially vertical direction. While moving (while rotating), the optical probe 81 emits light toward the substrate W. Therefore, the thickness detection signal indicates the thickness distribution of the target TG.

[0064] Specifically, the probe movement mechanism 9 has a probe arm 91, a second rotation shaft 93, and a second drive unit 95. The probe arm 91 extends in a substantially horizontal direction. The optical probe 81 is disposed at the tip of the probe arm 91. The probe arm 91 is coupled to the second rotation shaft 93. The second rotation shaft 93 extends in a substantially vertical direction. The second drive unit 95 rotates the second rotation shaft 93 about a third rotation axis AX3, causing the probe arm 91 to rotate about the second rotation axis 93 along a substantially horizontal plane. As a result, the optical probe 81 moves along the substantially horizontal plane. More specifically, the optical probe 81 revolves around the second rotation shaft 93 about the third rotation axis AX3. The second drive unit 95 includes, for example, a stepping motor.

[0065] In this embodiment, the thickness measuring unit 8 is an example of a processing amount detecting unit, and is used to detect the processing amount. The processing amount indicates the amount of the substrate W processed by the processing unit 1 when the substrate W is processed. More specifically, the processing amount indicates the amount of the substrate W processed by the etching liquid (processing liquid) supplied to the substrate W from the first nozzle 41. In this embodiment, the processing amount indicates the etching amount. The etching amount indicates the difference between the thickness of the target object TG before the etching process and the thickness of the target object TG after the etching process.

[0066] When machine learning is performed, the control device 101 acquires the etching amount by calculating the etching amount based on the thickness detection signal input from the thickness measurement unit 8 (thickness measurement device 85). More specifically, the control device 101 acquires the distribution of the etching amount (distribution of the processing amount). The control device 101 uses the acquired etching amount to generate learning data (learning data set) for machine learning.

[0067] Furthermore, the control device 101 acquires the etching amount by calculating the etching amount based on the thickness detection signal input from the thickness measurement unit 8 (thickness measurement device 85) during the manufacture of the semiconductor product. The control device 101 determines whether or not to generate learning data for additional learning based on the acquired etching amount during the manufacture of the semiconductor product. When the control device 101 determines to generate learning data for additional learning, it generates the learning data for additional learning based on the acquired etching amount.

[0068] The controller 101 also receives a rotation speed signal from the encoder 55. The rotation speed of the substrate W affects the etching amount. The controller 101 determines whether an abnormality has occurred in the substrate processing apparatus 100 based on the rotation speed signal. Specifically, as will be described with reference to FIG. 8 , the controller 101 stores a recipe 131 for controlling each part of the substrate processing apparatus 100, and the recipe 131 indicates a set value for the rotation speed of the motor main body 51. The set value for the rotation speed of the motor main body 51 is an example of a set value for a measurement target. The controller 101 controls the processing performed by the processing unit 1 by referring to the recipe 131. The controller 101 determines that the rotation speed of the substrate W indicates an abnormal value when the difference between the rotation speed indicated by the rotation speed signal and the set value for the rotation speed of the motor main body 51 included in the recipe 131 is equal to or greater than a predetermined value. When the controller 101 determines that the rotation speed of the substrate W indicates an abnormal value, the controller 101 determines that an abnormality has occurred in the substrate processing apparatus 100.

[0069] Next, the scanning process of the substrate W by the first nozzle 41 will be described with reference to FIG. 3. FIG. 3 is a plan view showing the scanning process of this embodiment. As shown in FIG. 3, the scanning process is a process in which the first nozzle 41 ejects the processing liquid onto the target object TG while moving so that the landing position of the processing liquid on the surface of the target object TG forms an arc-shaped trajectory TJ1 in a plan view. The trajectory TJ1 passes through the center CT of the substrate W. The center CT indicates the portion of the substrate W through which the first rotation axis AX1 passes. The scanning process is performed while the substrate W is rotating.

[0070] In this embodiment, the first nozzle 41 ejects the etching liquid toward the rotating substrate W while moving from the first position X1 to the ninth position X9. The positions X1 to X9 included in the first position X1 to the ninth position X9 are included in the locus TJ1. The section from the first position X1 to the ninth position X9 indicates the movement section along which the first nozzle 41 moves.

[0071] Of the first position X1 to the ninth position X9, the first position X1 indicates a discharge start position of the processing liquid (etchant), and the ninth position X9 indicates a discharge stop position of the processing liquid (etchant). The movement speed of the first nozzle 41 at the first position X1 is 0 mm / s, and the movement speed of the first nozzle 41 at the ninth position X9 is also 0 mm / s. Therefore, the first position X1 is the start position of the scanning process, and the ninth position X9 is the end position of the scanning process. Furthermore, the first position X1 is the movement start position of the first nozzle 41, and the ninth position X9 is the movement end position of the first nozzle 41. In the following description, the movement speed of the first nozzle 41 during the scanning process may be referred to as the "scanning speed."

[0072] During the scanning process, the first nozzle 41 passes through each of the intermediate positions (positions X2 to X8 from the second position X2 to the eighth position X8) between the first position X1 and the ninth position X9. Each of the intermediate positions is an example of a nozzle position, and divides the movement section of the first nozzle 41 into a plurality of sections.

[0073] Next, the scan speed information will be described with reference to Fig. 4. The scan speed information indicates the setting value of the movement speed of the first nozzle 41 during the scan process (scan speed setting value). Fig. 4 is a diagram showing the scan speed information of this embodiment. The scan speed information is an example of speed information, learning target speed information, and processing time speed information. In detail, Fig. 4 shows the relationship between each of the positions X1 to X9 included in the movement section of the first nozzle 41 described with reference to Fig. 3 and the scan speed setting value.

[0074] 4, the upper column shows positions X1 to X9 included in the movement section of the first nozzle 41, and the lower column shows the set value of the scan speed. Each of the positions X1 to X9 included in the movement section of the first nozzle 41 is defined by a radial position on the substrate W. In detail, the upper column shows the start position of the movement section of the first nozzle 41 (the movement start position of the first nozzle 41), the end position of the movement section of the first nozzle 41 (the movement end position of the first nozzle 41), and multiple intermediate positions (multiple positions through which the first nozzle 41 passes) between the start position and the end position of the movement section of the first nozzle 41.

[0075] 4, the scan speed information indicates the setting value of the scan speed for each of positions X1 to X9 included in the movement section of the first nozzle 41. Hereinafter, each of positions X1 to X9 included in the movement section of the first nozzle 41 may be referred to as a "speed setting position." In this embodiment, the scan speed information indicates nine speed setting positions.

[0076] Specifically, the speed setting positions correspond to the first position X1 to the ninth position X9 described with reference to Fig. 3. As described with reference to Fig. 3, the scan speed set at the start position (first position X1) of the movement section of the first nozzle 41 is 0 [mm / s], and the scan speed set at the end position (ninth position X9) of the movement section of the first nozzle 41 is 0 [mm / s].

[0077] 1 and 2 controls the nozzle movement mechanism 6 (first drive unit 65) with reference to Fig. 2 based on the scan speed information. As a result, the first nozzle 41 moves along the trajectory TJ1 with reference to Fig. 3 so that the scan speed at each speed setting position becomes the scan speed specified by the scan speed information.

[0078] Next, the movement speed (scanning speed) of the first nozzle 41 during the scanning process will be described with reference to Fig. 5. Fig. 5 is a graph showing an example of the movement speed of the first nozzle 41 in this embodiment.

[0079] 5, the vertical axis represents the scan speed [mm / s], and the horizontal axis represents the radial position [mm] of the substrate W. As shown in Fig. 5, the scan speed at the start of the scan process is 0 [mm / s]. Also, the scan speed at the end of the scan process is 0 [mm / s].

[0080] As described with reference to FIGS. 3 and 4, a scan speed is set for each of positions X1 to X9 (each speed setting position) from the first position X1 to the ninth position X9. As a result, as shown in FIG. 5, the scan speed continuously changes between adjacent speed setting positions from the scan speed set for one speed setting position to the scan speed set for the other speed setting position. For example, as shown in FIG. 4, a scan speed Y3 is set for the third position X3, and a scan speed Y4 is set for the fourth position X4. Therefore, the scan speed of the first nozzle 41 continuously changes from the scan speed Y3 to the scan speed Y4 while the first nozzle 41 moves from the third position X3 to the fourth position X4.

[0081] Next, the thickness measurement process by the thickness measurement unit 8 will be described with reference to FIG. 6. FIG. 6 is a plan view showing the thickness measurement process of this embodiment. As shown in FIG. 6, the thickness measurement process is a process of measuring the thickness of the object TG while the optical probe 81 moves so that the thickness measurement position relative to the object TG forms an arc-shaped trajectory TJ2 in a plan view. The trajectory TJ2 passes through the edge portion EG of the substrate W and the center portion CT of the substrate W. The edge portion EG indicates the peripheral portion of the substrate W. The thickness measurement process is performed while the substrate W is rotating.

[0082] Specifically, the optical probe 81 emits light toward the target TG while moving between the center CT and the edge EG of the substrate W in a plan view. As a result, the thickness of the target TG is measured at each measurement position included in the trajectory TJ2. Each measurement position corresponds to a radial position on the substrate W. Therefore, the thickness measurement process measures the thickness distribution of the target TG in the radial direction RD of the substrate W.

[0083] Next, the etching liquid supply unit 4 of this embodiment will be described with reference to Fig. 7. Fig. 7 is a schematic diagram of the etching liquid supply unit 4 of this embodiment. As shown in Fig. 7, the etching liquid supply unit 4 further includes a temperature sensor 421, a concentration sensor 422, a valve 423, a mixing valve 424, a flow meter 425, and a heater 426, in addition to the first supply pipe 42 described with reference to Fig. 2.

[0084] The temperature sensor 421 is an example of a measuring instrument that measures the value of a measurement target, and the measurement target of the temperature sensor 421 is the temperature of the etching liquid flowing through the first supply pipe 42. Hereinafter, the temperature of the etching liquid flowing through the first supply pipe 42 will be referred to as the "temperature of the etching liquid." The temperature sensor 421 generates a temperature signal that indicates the temperature of the etching liquid. The temperature of the etching liquid affects the amount of etching.

[0085] The temperature signal is input to the control device 101 described with reference to FIGS. 1 and 2. The control device 101 determines whether an abnormality has occurred in the substrate processing apparatus 100 based on the temperature signal. In detail, a recipe 131 described with reference to FIG. 8 indicates a set value for the temperature of the etching liquid. The set value for the temperature of the etching liquid is an example of a set value for a measurement object. The control device 101 determines that the temperature of the etching liquid indicates an abnormal value when the difference between the temperature indicated by the temperature signal and the set value for the temperature of the etching liquid included in the recipe 131 is equal to or greater than a predetermined specified value. When the control device 101 determines that the temperature of the etching liquid indicates an abnormal value, it determines that an abnormality has occurred in the substrate processing apparatus 100.

[0086] The concentration sensor 422 is an example of a measuring instrument that measures the value of a measurement target, and the measurement target of the concentration sensor 422 is the concentration of an etching component contained in the etching liquid flowing through the first supply pipe 42. Hereinafter, the concentration of the etching component contained in the etching liquid flowing through the first supply pipe 42 will be referred to as the "etching liquid concentration." The concentration sensor 422 generates a concentration signal that indicates the concentration of the etching liquid. The concentration of the etching liquid affects the amount of etching.

[0087] The concentration signal is input to the control device 101 described with reference to FIGS. 1 and 2. The control device 101 determines whether an abnormality has occurred in the substrate processing apparatus 100 based on the concentration signal. In detail, a recipe 131 described with reference to FIG. 8 indicates a set value for the concentration of the etching liquid. The set value for the concentration of the etching liquid is an example of a set value for a measurement target. The control device 101 determines that the concentration of the etching liquid indicates an abnormal value when the difference between the concentration indicated by the concentration signal and the set value for the concentration of the etching liquid included in the recipe 131 is equal to or greater than a predetermined specified value. When the control device 101 determines that the concentration of the etching liquid indicates an abnormal value, it determines that an abnormality has occurred in the substrate processing apparatus 100.

[0088] The valve 423 is disposed on the first supply pipe 42. The valve 423 switches between supplying and stopping the supply of the etching liquid to the first nozzle 41. Specifically, when the valve 423 is opened, the etching liquid is discharged from the first nozzle 41 toward the substrate W. On the other hand, when the valve 423 is closed, the discharge of the etching liquid stops. The valve 423 also controls the flow rate of the etching liquid flowing downstream of the valve 423 in the first supply pipe 42. Specifically, the flow rate of the etching liquid flowing downstream of the valve 423 is adjusted depending on the opening degree of the valve 423. Therefore, the discharge flow rate of the etching liquid is adjusted depending on the opening degree of the valve 423. The valve 423 is, for example, a motor valve.

[0089] The mixing valve 424 is disposed on the first supply pipe 42. When the mixing valve 424 is opened, pure water flows into the first supply pipe 42, and the concentration of the etching liquid is diluted.

[0090] The flow meter 425 is an example of a measuring instrument that measures the value of a measurement target, and the measurement target of the flow meter 425 is the discharge flow rate of the etching liquid. Specifically, the measurement target of the flow meter 425 is the flow rate of the etching liquid flowing through the first supply pipe 42. The flow meter 425 generates a discharge flow rate signal that indicates the discharge flow rate of the etching liquid. The discharge flow rate of the etching liquid affects the etching amount.

[0091] The discharge flow rate signal is input to the control device 101 described with reference to FIGS. 1 and 2. The control device 101 determines whether an abnormality has occurred in the substrate processing apparatus 100 based on the discharge flow rate signal. In detail, a recipe 131 described with reference to FIG. 8 indicates a set value for the discharge flow rate of the etching liquid. The set value for the discharge flow rate of the etching liquid is an example of a set value for a measurement target. The control device 101 determines that the discharge flow rate of the etching liquid indicates an abnormal value when the difference between the discharge flow rate indicated by the discharge flow rate signal and the set value for the discharge flow rate of the etching liquid included in the recipe 131 is equal to or greater than a predetermined specified value. When the control device 101 determines that the discharge flow rate of the etching liquid indicates an abnormal value, it determines that an abnormality has occurred in the substrate processing apparatus 100.

[0092] The heater 426 heats the etching liquid flowing through the first supply pipe .

[0093] Next, the control device 101 will be described with reference to Fig. 8. Fig. 8 is a block diagram of the control device 101 of this embodiment. In detail, Fig. 8 shows the control device 101 before machine learning is performed. As shown in Fig. 8, the control device 101 has a control unit 102, a storage unit 103, an input unit 104, and a display unit 105.

[0094] The control unit 102 includes a processor. The control unit 102 includes, for example, a central processing unit (CPU) or a micro processing unit (MPU). Alternatively, the control unit 102 includes a general-purpose computing machine. The control unit 102 may further include a neural network processing unit (NPU).

[0095] The memory unit 103 stores data and computer programs. The memory unit 103 has a main memory device. The main memory device is, for example, a semiconductor memory. The memory unit 103 may further have an auxiliary memory device. The auxiliary memory device is, for example, a semiconductor memory and / or a hard disk drive. The memory unit 103 may have removable media. The control unit 102 controls the operation of each unit of the substrate processing apparatus 100 based on the data and computer programs stored in the memory unit 103. Furthermore, the control unit 102 performs machine learning and additional learning based on the data and computer programs stored in the memory unit 103.

[0096] Specifically, the storage unit 103 stores a recipe 131, a control program 132, an inference program 133, and a learning program 134. The recipe 131 defines the processing content and processing procedure for the substrate W. The recipe 131 also indicates various setting values. For example, the recipe 131 indicates the scan speed information described with reference to FIG. 4.

[0097] The control unit 102 controls the operation of each unit of the substrate processing apparatus 100 based on a recipe 131 and a control program 132. The inference program 133 is used to generate a trained model, which will be described with reference to Fig. 15. The training program 134 is a program for finding certain rules from a training dataset and executing an algorithm for generating a model (trained model) that expresses the rules.

[0098] The storage unit 103 stores learning target speed information #A1 to #An when machine learning is performed. Here, "n" represents a positive integer. Each of the learning target speed information #A1 to #An represents scan speed information similar to the scan speed information described with reference to FIG. 4. The learning target speed information #A1 to #An differ from each other in the content of the set of scan speeds set for each speed setting position described with reference to FIG. 4.

[0099] The input unit 104 accepts input from an operator and outputs information indicating the input result to the control unit 102. For example, the input unit 104 accepts input of learning target speed information #A1 to #An. The input unit 104 includes, for example, a touch panel and a pointing device. The touch panel is, for example, disposed on the display surface of the display unit 105. The input unit 104 and the display unit 105 constitute, for example, a graphical user interface.

[0100] The display unit 105 displays various types of information. In this embodiment, the display unit 105 displays, for example, various error screens and various setting screens (input screens). The display unit 105 has, for example, a liquid crystal display or an organic EL (electroluminescence) display.

[0101] The control device 101 may have an interface for acquiring the learning target speed information #A1 to #An from a removable memory. The removable memory may include, for example, a USB (Universal Serial Bus) memory. Alternatively, the control device 101 may have a drive for acquiring the learning target speed information #A1 to #An from removable media. The removable media may include, for example, an optical disc such as a CD (Compact Disc).

[0102] Next, a method for generating learning data executed by the substrate processing apparatus 100 will be described with reference to Figures 1, 2, 8, and 9. Figure 9 is a flowchart showing the method for generating learning data in this embodiment. In detail, Figure 9 shows the process executed by the control unit 102 when generating learning data. The method for generating learning data in this embodiment includes the processes of steps S1 to S9.

[0103] 9 is started by an operator operating the input unit 104. When generating learning data (learning data set), a plurality of substrates W to be learned are accommodated in at least one of the plurality of load ports LP. The substrates W to be learned are of the same type as the substrates W used in manufacturing semiconductor products.

[0104] When generating learning data (learning data set), the control unit 102 first selects one of the plurality of processing units 1 as the processing unit 1 to be learned. The processing described with reference to Fig. 9 is sequentially performed on the plurality of processing units 1 included in the substrate processing apparatus 100.

[0105] When the control unit 102 determines the processing unit 1 to be learned, it selects one of the learning target speed information #A1 to #An and acquires the one of the learning target speed information #A1 to #An (step S1). Hereinafter, one of the learning target speed information #A1 to #An will be referred to as "learning target speed information #A." For example, the control unit 102 selects the learning target speed information #A in ascending order of the identification numbers assigned to the learning target speed information #A1 to #An, respectively.

[0106] After selecting one of the learning target speed information #A1 to #An, the control unit 102 controls the indexer robot IR and the center robot CR so that the learning target substrate W is loaded into the chamber 2 of the processing unit 1 to be learned (step S2). The control unit 102 causes the learning target substrate W loaded into the chamber 2 to be held by the spin chuck 3.

[0107] When the substrate W to be studied is held by the spin chuck 3, the control unit 102 causes the thickness measurement unit 8 to measure the thickness distribution of the target object TG included in the substrate W to be studied (step S3). The thickness distribution of the target object TG measured here indicates the thickness distribution of the target object TG before the etching process. Hereinafter, the thickness distribution of the target object TG before the etching process may be referred to as the "thickness distribution before the process."

[0108] After measuring the pre-processing thickness distribution, the control unit 102 controls the operation of each unit of the substrate processing apparatus 100 so that an etching process is performed on the learning target substrate W (step S4). Specifically, as described with reference to Figures 3 to 5, the control unit 102 controls the operation of each unit of the substrate processing apparatus 100 so that an etching liquid is supplied from the first nozzle 41 to the learning target substrate W while the first nozzle 41 moves (turns) at a speed based on the learning target speed information #A, thereby performing an etching process on the learning target substrate W.

[0109] After the etching process is performed, the control unit 102 causes the thickness measurement unit 8 to measure the thickness distribution of the target object TG included in the substrate W to be learned (step S5). The thickness distribution of the target object TG measured here indicates the thickness distribution of the target object TG after the etching process. Hereinafter, the thickness distribution of the target object TG after the etching process may be referred to as the "thickness distribution after the process." Note that, as will be described with reference to FIG. 10, the thickness distribution after the process indicates the thickness distribution after the drying process.

[0110] After measuring the thickness distribution after processing, the control unit 102 releases the spin chuck 3 from holding the learning target substrate W, and causes the indexer robot IR to unload the learning target substrate W from the chamber 2 (step S6). Thereafter, the control unit 102 controls the indexer robot IR and the center robot CR so that the learning target substrate W is transported to one of the multiple load ports LP.

[0111] Next, the control unit 102 acquires the processing amount (etching amount) (step S7). Specifically, the control unit 102 calculates the difference between the thickness distribution before the processing and the thickness distribution after the processing to acquire the etching amount.

[0112] When the control unit 102 acquires the processing amount (etching amount), it generates learning data based on the learning target speed information #A and the processing amount (etching amount) (step S8). The learning data indicates the learning target speed information #A and the processing amount (etching amount). The learning data is stored in the storage unit 103.

[0113] After generating the learning data, the control unit 102 determines whether or not an abnormality has occurred in the substrate processing apparatus 100 (step S9), and ends the process shown in Fig. 9. For example, as described with reference to Figs. 2 and 7, the control unit 102 determines whether or not at least one of the rotation speed of the substrate W, the temperature of the etching liquid, the concentration of the etching liquid, and the discharge flow rate of the etching liquid exhibits an abnormal value.

[0114] The control unit 102 repeatedly executes the process shown in Fig. 9 until all of the learning target speed information #A1 to #An are selected. As a result, a learning data set is generated. The learning data set includes a plurality of pieces of learning data. The learning data set is stored in the storage unit 103. Note that at least one of the processes of steps S7 to S9 may be executed between the process of step S5 and the process of step S6.

[0115] Next, the etching process (step S4) will be described with reference to Figures 2, 8 and 10. Figure 10 is a flowchart showing the etching process of this embodiment.

[0116] 10, after the spin chuck 3 holds the substrate W to be learned, the substrate W to be learned is processed (step S41). Specifically, the control unit 102 causes the spin motor unit 5 to rotate the substrate W held by the spin chuck 3. Thereafter, as described with reference to FIG. 9, the control unit 102 controls the nozzle moving mechanism 6 and the etching liquid supply unit 4 so that the etching liquid is supplied from the first nozzle 41 toward the substrate W to be learned. As a result, the substrate W to be learned is etched.

[0117] When etching of the learning target substrate W is completed, the control unit 102 controls the rinse liquid supply unit 7 to supply a rinse liquid to the learning target substrate W, thereby removing the etching liquid from the learning target substrate W (step S42). Specifically, the etching liquid is swept outward from the learning target substrate W by the rinse liquid and discharged around the learning target substrate W. As a result, the liquid film of the rinse liquid on the learning target substrate W is replaced with a liquid film of the rinse liquid.

[0118] After replacing the etching liquid with the rinse liquid, the control unit 102 controls the spin motor unit 5 to dry the learning target substrate W (step S43). As a result, the process shown in FIG. 10 ends. Specifically, the control unit 102 increases the rotation speed of the learning target substrate W to be higher than the rotation speed during the etching process and the rinse process. As a result, a large centrifugal force is applied to the rinse liquid on the learning target substrate W, and the rinse liquid adhering to the learning target substrate W is thrown off around the learning target substrate W. In this way, the rinse liquid is removed from the learning target substrate W, and the learning target substrate W is dried. Note that the control unit 102 stops the rotation of the substrate W by the spin motor unit 5, for example, after a predetermined time has elapsed since the learning target substrate W started to be rotated at high speed.

[0119] Next, the process of generating learning data (step S8) will be described with reference to Figures 2, 8 and 11. Figure 11 is a flowchart showing the process of generating learning data in this embodiment.

[0120] 11, the control unit 102 generates learning data using the learning target speed information #A and the processing amount (etching amount) (step S81). The control unit 102 determines whether the generated learning data (processing amount) can be used for machine learning (step S82).

[0121] Specifically, the control unit 102 determines whether the processing amount (etching amount) is equal to or greater than a specified value. If the processing amount (etching amount) is not equal to or greater than the specified value, the control unit 102 determines that the learning data (processing amount) is usable for machine learning. On the other hand, if the processing amount (etching amount) is equal to or greater than the specified value, the control unit 102 determines that the learning data (processing amount) is not usable for machine learning. For example, if at least one point in the distribution of the processing amount (etching amount) is equal to or greater than the specified value, the control unit 102 determines that the learning data (processing amount) is not usable for machine learning.

[0122] In addition, the control unit 102 may determine that the learning data (processing amount) cannot be used for machine learning if multiple points in the distribution of the processing amount (etching amount) are equal to or greater than a specified value, or may determine that the learning data (processing amount) cannot be used for machine learning if the range in which the processing amount (etching amount) is equal to or greater than a specified value is equal to or greater than a predetermined range.

[0123] When the control unit 102 determines that the learning data (processing amount) can be used for machine learning (Yes in step S82), it assigns a first flag to the learning data (step S83) and ends the processing shown in Fig. 11. Specifically, the control unit 102 associates the learning data with the first flag and stores them in the storage unit 103. The first flag indicates that the learning data will be used.

[0124] On the other hand, when the control unit 102 determines that the learning data (processing amount) cannot be used for machine learning (No in step S82), it assigns a second flag to the learning data (step S84) and ends the processing shown in FIG. 11. Specifically, the control unit 102 associates the learning data with the second flag and stores them in the storage unit 103. The second flag indicates that the learning data will not be used. Therefore, the learning data to which the second flag has been assigned is excluded from the learning data set for machine learning.

[0125] 2, 8 and 12, the abnormality determination process (step S9) for determining whether or not an abnormality has occurred in the substrate processing apparatus 100 will be described below. Fig. 12 is a flowchart showing the abnormality determination process in this embodiment.

[0126] 12, the control unit 102 determines whether or not an abnormality has occurred in the substrate processing apparatus 100 (step S91). If the control unit 102 determines that an abnormality has not occurred in the substrate processing apparatus 100 (No in step S91), the control unit 102 ends the process shown in FIG.

[0127] On the other hand, when the control unit 102 determines that an abnormality has occurred in the substrate processing apparatus 100 (Yes in step S91), the control unit 102 executes different error notification depending on whether the learning data (processing amount) is available for machine learning. In the present embodiment, the control unit 102 determines whether the first flag or the second flag is associated with the learning data (step S92).

[0128] If the first flag is associated with the learning data (A in step S92), the control unit 102 causes the display unit 105 to display a first error screen (step S93), and ends the process shown in Fig. 12. The first error screen indicates that an abnormality has occurred in the substrate processing apparatus 100. In detail, the first error screen indicates the measurement target (monitoring target) in which the abnormality has occurred.

[0129] If the second flag is associated with the learning data (B in step S92), the control unit 102 causes the display unit 105 to display a second error screen (step S94), and ends the processing shown in Fig. 12. The second error screen indicates that an abnormality has occurred in the substrate processing apparatus 100 and that the processing amount (etching amount) is equal to or greater than a specified value. Therefore, the second error screen can notify the operator that the reason the processing amount (etching amount) is equal to or greater than the specified value may lie not in the substrate W to be learned but in the substrate processing apparatus 100.

[0130] Next, a training dataset table TB10 for managing training datasets will be described with reference to Fig. 13. Fig. 13 is a diagram showing an example of the training dataset table TB10 of this embodiment.

[0131] In this embodiment, the control unit 102 stores the learning data in table format in the storage unit 103. As shown in Fig. 13, the learning data set table TB10 has a speed information field TB11, a processing amount field TB12, and a flag field TB13.

[0132] The speed information field TB11 stores the learning target speed information #A. The processing amount field TB12 stores the processing amount (etching amount). The flag field TB13 stores a flag. As shown in FIG. 13, the learning target speed information #A, the processing amount (etching amount), and the flag are stored in association with each other in the learning dataset table TB10.

[0133] In the flag field TB13, "OK" indicates the first flag, and "NG" indicates the second flag. In other words, "OK" indicates that the corresponding training data is available for machine learning, and "NG" indicates that the corresponding training data is not available for machine learning.

[0134] Next, the control device 101 after the training data set 135 has been generated will be described with reference to Fig. 14. Fig. 14 is a block diagram of the control device 101 of this embodiment. In detail, Fig. 14 shows the control device 101 after the training data set 135 has been generated.

[0135] As shown in FIG. 14 , the storage unit 103 stores a training data set 135. The training data set 135 includes a plurality of pieces of training data. Specifically, as described with reference to FIG. 13 , the control unit 102 stores the plurality of pieces of training data in a training data set table TB10. After generating the training data set 135, the control unit 102 may or may not delete the learning target speed information #A1 to #An from the storage unit 103.

[0136] The control unit 102 generates trained parameters by training the training data set 135 (plurality of training data). The control unit 102 generates a trained model based on the trained parameters.

[0137] Next, a trained model generation method executed by the substrate processing apparatus 100 will be described with reference to Figures 1, 2, 13, 14, and 15. Figure 15 is a flowchart showing the trained model generation method in this embodiment. In detail, Figure 15 shows the processing executed by the control unit 102 when generating a trained model. The trained model generation method in this embodiment includes the processing of steps S11 to S14.

[0138] First, the control unit 102 acquires the learning data set 135 from the storage unit 103 (step S11). At this time, the control unit 102 acquires the learning data to which the first flag is assigned from among the plurality of learning data.

[0139] When the control unit 102 acquires the training data set 135, it trains the training data set 135 (plurality of training data) based on the training program 134 (step S12). Here, training refers to discovering a certain rule from the training data set 135. In this embodiment, the training data indicates the relationship between scan speed information (scan speed at each speed setting position) and the processing amount (etching amount). The control unit 102 trains the training data set 135 to discover a certain rule between the scan speed information and the processing amount (etching amount).

[0140] The control unit 102 generates trained parameters by training the training data set 135 (step S13). More specifically, the trained parameters are output from the training program 134. The trained parameters are parameters (coefficients) acquired based on the results of training using the training data set 135.

[0141] Next, the control unit 102 generates a trained model based on the acquired trained parameters (step S14). As a result, the process shown in FIG. 15 ends. Specifically, the control unit 102 generates a trained model by incorporating the trained parameters into the inference program 133. The inference program 133 is a program that outputs a certain result for an input by applying the incorporated trained parameters, and the trained model is a model that includes the trained parameters. In this embodiment, the "input" is the target amount of processing amount (etching amount), and the "output" is scan speed information. In other words, the target amount of processing amount (etching amount) is an explanatory variable, and the scan speed information output from the trained model is a target variable. Hereinafter, the target amount of processing amount (etching amount) may be referred to as a "target processing amount" or a "target etching amount." Furthermore, the scan speed information output from the trained model may be referred to as "processing speed information."

[0142] Next, the control device 101 during the manufacturing of semiconductor products will be described with reference to Fig. 2 and Fig. 16. Fig. 16 is a block diagram of the control device 101 of this embodiment. In detail, Fig. 16 shows the control device 101 during the manufacturing of semiconductor products.

[0143] 16 , when a semiconductor product is manufactured, the memory unit 103 stores the trained model 136. After the trained model 136 is generated, the control unit 102 may or may not delete the training dataset 135 from the memory unit 103.

[0144] The control unit 102 controls the nozzle movement mechanism 6 based on the trained model 136 during the manufacture of semiconductor products. Specifically, the control unit 102 inputs a target processing amount (target etching amount), which is an explanatory variable, into the trained model 136, thereby causing the trained model 136 to output processing speed information, which is an objective variable. The control unit 102 controls the nozzle movement mechanism 6 so that the first nozzle 41 moves at a speed based on the processing speed information during the manufacture of semiconductor products. Specifically, the first nozzle 41 moves (turns) at a speed based on the processing speed information when processing (etching) the substrate W. In the following description, the substrate W to be processed during the manufacture of semiconductor products may be referred to as the "substrate W to be processed."

[0145] Next, a substrate processing method executed by the substrate processing apparatus 100 will be described with reference to Figures 1, 2, 16, 17, and 18. Figures 17 and 18 are flowcharts showing the substrate processing method in this embodiment. In detail, Figures 17 and 18 show the processing executed by the control unit 102 when etching the substrate W to be processed. The substrate processing method in this embodiment includes the processing of steps S21 to S32.

[0146] 17 and 18 is started when an operator operates the input unit 104. A plurality of substrates W to be processed are accommodated in at least one of the plurality of load ports LP.

[0147] When etching the substrate W to be processed, first, the control unit 102 acquires the target thickness of the target object TG (step S21). More specifically, the control unit 102 acquires the distribution of the target thickness of the target object TG. The target thickness is input to the control unit 102 by the operator operating the input unit 104.

[0148] When the control unit 102 acquires the target thickness, it controls the indexer robot IR and the center robot CR so that the substrate W to be processed is loaded into the chamber 2 of one of the processing units 1 (step S22). The control unit 102 causes the spin chuck 3 to hold the substrate W to be processed that has been loaded into the chamber 2.

[0149] When the substrate W to be processed is held by the spin chuck 3, the control unit 102 causes the thickness measurement unit 8 to measure the thickness distribution of the targets TG included in the substrate W to be processed (step S23). In other words, the thickness distribution before processing is measured.

[0150] Next, the control unit 102 acquires a target processing amount (target etching amount), which is an explanatory variable, based on the distribution of the thickness before processing and the distribution of the target thickness (step S24). Specifically, the control unit 102 calculates the difference between the distribution of the thickness before processing and the distribution of the target thickness, and acquires the target processing amount (target etching amount).

[0151] Next, the control unit 102 inputs the target processing amount (target etching amount) into the trained model 136 and causes the trained model 136 to output processing speed information, thereby acquiring the processing speed information, which is the objective variable (step S25).

[0152] Upon acquiring the processing speed information, the control unit 102 controls the operation of each unit of the substrate processing apparatus 100 so that the etching process is performed on the substrate W to be processed (step S26). Specifically, as described with reference to FIGS. 3 to 5, the control unit 102 controls the operation of each unit of the substrate processing apparatus 100 so that the etching liquid is supplied from the first nozzle 41 to the substrate W to be processed while the first nozzle 41 moves (turns) at a speed based on the processing speed information, thereby performing the etching process on the substrate W to be processed. More specifically, the substrate W to be processed is etched by a process similar to the process described with reference to FIG.

[0153] After the etching process is performed, the control unit 102 causes the thickness measurement unit 8 to measure the thickness distribution of the target objects TG included in the substrate W to be processed (step S27). In other words, the thickness distribution after the process is measured.

[0154] After measuring the thickness distribution after processing, the control unit 102 releases the substrate W to be processed from the spin chuck 3 and causes the indexer robot IR to unload the substrate W to be processed from the chamber 2 (step S28). Thereafter, the control unit 102 controls the indexer robot IR and the center robot CR so that the substrate W to be processed is transported to one of the multiple load ports LP.

[0155] When the substrate W to be processed is transported, the control unit 102 acquires the processing amount (etching amount) (step S29) as shown in Fig. 18. In detail, the control unit 102 calculates the difference between the thickness distribution before processing and the thickness distribution after processing to acquire the etching amount.

[0156] When the control unit 102 acquires the processing amount (etching amount), it determines whether the processing amount (etching amount) matches the target processing amount (target etching amount) (step S30).

[0157] If the control unit 102 determines that the processing amount (etching amount) does not match the target processing amount (target etching amount) (No in step S30), the process proceeds to step S31. If the control unit 102 determines that the processing amount (etching amount) matches the target processing amount (target etching amount) (Yes in step S30), the process proceeds to step S32. Note that the control unit 102 may determine whether the processing amount (etching amount) matches the target processing amount (target etching amount) within an allowable range.

[0158] In step S31, the control unit 102 generates learning data for additional learning based on the processing speed information and the processing amount (etching amount). Specifically, the learning data for additional learning is generated by a process similar to the process described with reference to Fig. 11. Once the learning data for additional learning is generated, the process proceeds to step S32.

[0159] In step S32, the control unit 102 determines whether or not an abnormality has occurred in the substrate processing apparatus 100 (step S32), and ends the processing shown in Figures 17 and 18. In detail, whether or not an abnormality has occurred in the substrate processing apparatus 100 is determined by the same processing as that described with reference to Figure 12.

[0160] During the manufacture of semiconductor products, the control unit 102 repeatedly executes the processes of FIGS. 17 and 18. As a result, a training dataset for additional learning is generated. The training dataset for additional learning is stored in the storage unit 103. Note that at least one of the processes of steps S29 to S32 may be executed between steps S26 and S27. Hereinafter, the training dataset for additional learning may be referred to as an "additional learning dataset." Furthermore, the training data for additional learning may be referred to as "additional learning data."

[0161] Next, an additional training dataset table TB20 for managing additional training datasets will be described with reference to Fig. 19. Fig. 19 is a diagram showing an example of the additional training dataset table TB20 of this embodiment.

[0162] In this embodiment, the control unit 102 stores additional training data in table format in the storage unit 103. As shown in Fig. 19, the additional training data set table TB20 has a speed information field TB21, a processing amount field TB22, and a flag field TB23.

[0163] The speed information field TB21 stores processing speed information #B. The processing amount field TB22 stores the processing amount (etching amount). The flag field TB23 stores a flag, similar to the flag field TB13 described with reference to Fig. 13. As shown in Fig. 19, the processing speed information #B, the processing amount (etching amount), and the flag are associated with each other and stored in the additional learning dataset table TB20.

[0164] When the additional learning dataset is generated, the control unit 102 generates a trained model, similar to when machine learning is performed. Hereinafter, a method for generating a trained model after additional learning will be described with reference to FIG. 15 .

[0165] First, the control unit 102 acquires an additional learning data set from the storage unit 103 (step S11). At this time, the control unit 102 acquires additional learning data to which a first flag has been assigned from among the multiple additional learning data.

[0166] When the control unit 102 acquires the additional training data set, the control unit 102 learns the additional training data set (plurality of additional training data) based on the training program 134 (step S12).

[0167] The control unit 102 generates new trained parameters by training the additional training data set (step S13). More specifically, the training program 134 outputs new trained parameters.

[0168] Next, the control unit 102 generates a trained model after additional training based on the new trained parameters (step S14). As a result, the process shown in Fig. 15 ends. In detail, the control unit 102 generates a trained model after additional training by incorporating the new trained parameters into the inference program 133.

[0169] 1 to 19, the first embodiment of the present invention has been described above. According to this embodiment, it is possible to reduce the burden on the worker involved in the development of scan speed information. Specifically, by generating the trained model 136 using various types of training target substrates W with different pre-processing, even if the pre-processing is changed during the manufacturing stage of semiconductor products, the worker does not need to newly develop scan speed information.

[0170] Furthermore, according to this embodiment, even if there is variation in the thickness of the object TG before processing, the trained model 136 outputs scan speed information for adjusting the thickness of the object TG after processing to the target thickness. Therefore, there is no need to develop a large number of scan speed information items to match the variation in the thickness of the object TG before processing. Furthermore, even if there is variation in the thickness of the object TG before processing, the thickness distribution after processing can be stably adjusted to the target thickness distribution. For example, the surface of the substrate W can be made approximately parallel to the horizontal plane. Alternatively, the thickness of a specific region of the object TG can be made to be the target thickness.

[0171] Furthermore, according to this embodiment, even if an error occurs during the manufacture of semiconductor products, causing the thickness of the processed target object TG to differ from the target thickness, the substrate processing apparatus 100 performs additional learning, so that after the error occurs, the learned parameters are adjusted during the manufacture of the semiconductor products, and the thickness after processing can be made to match the target thickness. Also, because the substrate processing apparatus 100 performs additional learning, there is no need for the operator to develop new scan speed information in response to the occurrence of an error. Furthermore, even if the pre-processing is changed, there is no need for the operator to develop new scan speed information, because the learned parameters are adjusted by additional learning.

[0172] In addition to the scan speed, factors that affect the processing amount (etching amount) include, for example, the rotation speed of the substrate W, the temperature of the processing liquid, the concentration of the processing liquid, and the discharge flow rate of the processing liquid. However, the control unit 102 does not actively perform processing (control) to change the values ​​of these other factors when generating learning data for machine learning or when processing the target substrate W. In other words, the control unit 102 adjusts the scan speed to set the processing amount to the target amount, but does not adjust (change) the values ​​of these other factors. Therefore, when generating learning data and when processing the target substrate W, the values ​​of these other factors show approximately the same values ​​except when an abnormality occurs. In this embodiment, the processing speed information is selected as the objective variable from among the multiple factors that affect the processing amount because, when homogenizing the distribution of the processing amount is used as the evaluation index, the correlation between the scan speed and the evaluation index is stronger than the correlation between the evaluation index and factors other than the scan speed. According to this embodiment, in order to generate learning data, it is only necessary to create a data set of learning target speed information #A (scan speed information), thereby reducing the burden on workers involved in the development of the learned model or scan speed information.

[0173] [Embodiment 2] Next, a second embodiment of the present invention will be described with reference to Figures 20 and 21. However, differences from the first embodiment will be described, and a description of the same aspects as in the first embodiment will be omitted. Unlike the first embodiment, the second embodiment has a learning device 200 that generates a trained model 136.

[0174] 20 is a diagram showing a substrate processing apparatus 100 and a learning apparatus 200 according to this embodiment. In this embodiment, the control apparatus 101 of the substrate processing apparatus 100 generates a learning dataset 135, as described with reference to FIGS. 1 to 13. The learning dataset 135 is input to the learning apparatus 200.

[0175] The learning device 200 generates a trained model 136 using a training dataset 135. The trained model 136 is input to the control device 101 of the substrate processing apparatus 100. Specifically, the learning device 200 includes a processing unit 201 and a memory unit 202. The processing unit 201 is an example of a learning unit.

[0176] The processing unit 201 includes a processor. The processing unit 201 includes, for example, a CPU or an MPU. Alternatively, the processing unit 201 includes a general-purpose computing machine. The processing unit 201 may further include an NPU. The storage unit 202 stores data and computer programs. The storage unit 202 includes a main storage device. The main storage device is, for example, a semiconductor memory. The storage unit 202 may further include an auxiliary storage device. The auxiliary storage device is, for example, a semiconductor memory and / or a hard disk drive. The storage unit 202 may include removable media. The processing unit 201 performs machine learning based on the data and computer programs stored in the storage unit 202.

[0177] Specifically, the storage unit 202 stores the inference program 133 and the learning program 134 described in embodiment 1. Similar to the control unit 102 described in embodiment 1, the processing unit 201 learns a learning dataset 135 (plurality of learning data) to generate learned parameters, and generates a learned model 136 based on the learned parameters.

[0178] The training dataset 135 may be input to the training device 200 via a removable memory, or may be input to the training device 200 via a removable medium. Similarly, the trained model 136 may be input to the substrate processing apparatus 100 via a removable memory, or may be input to the substrate processing apparatus 100 via a removable medium. Specifically, the training device 200 may have an interface that acquires the training dataset 135 from the removable memory and stores the trained model 136 in the removable memory, or may have a drive that acquires the training dataset 135 from the removable medium and stores the trained model 136 in the removable medium.

[0179] Alternatively, the substrate processing apparatus 100 and the learning apparatus 200 may be communicatively connected to each other. In this case, the learning dataset 135 is transmitted from the substrate processing apparatus 100 to the learning apparatus 200, and the trained model 136 is transmitted from the learning apparatus 200 to the substrate processing apparatus 100. The communication method may be a wired method or a wireless method. The substrate processing apparatus 100 and the learning apparatus 200 may be communicatively connected to each other via a public communication network such as the Internet.

[0180] Next, a method for generating the trained model 136 executed by the learning device 200 will be described with reference to Fig. 20 and Fig. 21. Fig. 21 is a flowchart showing the method for generating the trained model 136 of this embodiment. In detail, Fig. 21 shows the processing executed by the processing unit 201 when generating the trained model 136. The method for generating the trained model 136 of this embodiment includes the processing of steps S200 to S203.

[0181] First, the processing unit 201 acquires the training data set 135 from the storage unit 202 (step S200). At this time, the processing unit 201 acquires training data to which a first flag has been assigned from among the plurality of training data.

[0182] When the processing unit 201 acquires the training data set 135, the processing unit 201 trains the training data set 135 (plurality of training data) based on the training program 134 (step S201).

[0183] The processing unit 201 generates trained parameters by training the training data set 135 (step S202). More specifically, the training program 134 outputs the trained parameters.

[0184] Next, the processing unit 201 generates the trained model 136 based on the acquired trained parameters (step S203). As a result, the processing shown in Fig. 21 ends. In detail, the processing unit 201 generates the trained model 136 by incorporating the trained parameters into the inference program 133.

[0185] The second embodiment of the present invention has been described above with reference to Fig. 20 and Fig. 21. According to this embodiment, similar to the first embodiment, it is possible to reduce the burden on the workers involved in the development of scan speed information. Furthermore, according to this embodiment, it is not necessary to perform machine learning in the substrate processing apparatus 100. Therefore, an existing substrate processing apparatus can be used as the substrate processing apparatus 100.

[0186] In this embodiment, the learning data set 135 is input from the substrate processing apparatus 100 to the learning device 200, but learning data may be input from the substrate processing apparatus 100 to the learning device 200.

[0187] Furthermore, the learning device 200 may further perform the additional learning described in embodiment 1. When the learning device 200 performs additional learning, an additional learning data set or additional learning data is input to the learning device 200 from the substrate processing apparatus 100.

[0188] Furthermore, in this embodiment, the learning data set 135 is input from the substrate processing apparatus 100 to the learning device 200, but learned parameters may also be input from the substrate processing apparatus 100 to the learning device 200. When learned parameters are input from the substrate processing apparatus 100 to the learning device 200, the learning device 200 (processing unit 201) executes the process of step S203 described with reference to FIG.

[0189] [Embodiment 3] Next, a third embodiment of the present invention will be described with reference to Figures 17, 18, 22, and 23. However, differences from the first and second embodiments will be described, and a description of the same aspects as the first and second embodiments will be omitted. Unlike the first and second embodiments, the third embodiment causes the learning device 200 to output processing speed information, which is a response variable, from the trained model 136.

[0190] 22 is a diagram showing a substrate processing system 300 according to this embodiment. As shown in FIG. 22, the substrate processing system 300 includes a substrate processing apparatus 100 and a learning apparatus 200. The learning apparatus 200 is an example of an information processing apparatus. In this embodiment, a processing unit 201 of the learning apparatus 200 generates a trained model 136 as described in the second embodiment. The processing unit 201 stores the trained model 136 in a storage unit 202.

[0191] 22, the substrate processing apparatus 100 further includes a communication unit 106. The learning device 200 further includes a communication unit 203. The communication unit 106 and the communication unit 203 are, for example, a LAN board or a wireless LAN board. The communication unit 106 and the communication unit 203 are connected to each other so as to be able to communicate with each other via a public communication network such as the Internet.

[0192] The communication unit 106 of the substrate processing apparatus 100 is controlled by the control device 101 (control unit 102) to transmit a target processing amount (target etching amount) to the communication unit 203 of the learning device 200. The communication unit 106 of the substrate processing apparatus 100 is also controlled by the control device 101 (control unit 102) to receive processing speed information from the communication unit 203 of the learning device 200. As a result, the control device 101 (control unit 102) acquires the processing speed information. The control unit 102 stores the acquired processing speed information in the memory unit 103.

[0193] The communication unit 203 of the learning device 200 is controlled by the processing unit 201 to receive the target processing amount (target etching amount) from the communication unit 106 of the substrate processing apparatus 100. As a result, the processing unit 201 acquires the target processing amount (target etching amount). In addition, the communication unit 203 of the learning device 200 is controlled by the processing unit 201 to transmit processing speed information to the communication unit 106 of the substrate processing apparatus 100.

[0194] 22 and 23, a description will be given of the processing performed by the control device 101 (controller 102) and the processing performed by the processing unit 201. Fig. 23 is a flowchart showing the processing performed by the control device 101 (controller 102) of the substrate processing apparatus 100 and the processing performed by the processing unit 201 of the learning device 200.

[0195] When the control device 101 (control unit 102) of the substrate processing apparatus 100 acquires the target processing amount (target etching amount) as described in embodiment 1 (steps S21 to S24 in Figure 17), it controls the communication unit 106 to transmit the target processing amount (target etching amount) to the communication unit 203 of the learning device 200 (step S210).

[0196] The processing unit 201 of the learning device 200 controls the communication unit 203 to receive the target processing amount (target etching amount) (step S211). As a result, the processing unit 201 acquires the target processing amount (target etching amount).

[0197] The processing unit 201 inputs the target processing amount (target etching amount), which is an explanatory variable, into the trained model 136, thereby causing the trained model 136 to output processing speed information, which is an objective variable. As a result, the processing unit 201 of the learning device 200 acquires the processing speed information (step S212).

[0198] The processing unit 201 of the learning device 200 controls the communication unit 203 to transmit the processing speed information to the communication unit 106 of the substrate processing apparatus 100 (step S213). The control device 101 (control unit 102) of the substrate processing apparatus 100 controls the communication unit 106 to cause the communication unit 106 to receive the processing speed information (step S214). As a result, the control device 101 (control unit 102) acquires the processing speed information. Thereafter, the control device 101 (control unit 102) executes the processes from step S26 onwards, which have been described with reference to FIGS. 17 and 18.

[0199] The third embodiment of the present invention has been described above with reference to Figures 17, 18, 22, and 23. According to this embodiment, as in the first and second embodiments, it is possible to reduce the burden on the workers involved in developing the scan speed information. Furthermore, according to this embodiment, it is not necessary to store the trained model 136 in the memory unit 103 of the substrate processing apparatus 100.

[0200] [Embodiment 4] Next, a fourth embodiment of the present invention will be described with reference to Figures 1, 8, 14, 16, and 24 to 31. However, only differences from the first to third embodiments will be described, and descriptions of the same aspects as the first to third embodiments will be omitted. Unlike the first to third embodiments, the fourth embodiment classifies the trained model 136 into multiple classes.

[0201] FIG. 24 is a schematic diagram of the processing unit 1 of this embodiment. More specifically, FIG. 24 is a schematic cross-sectional view of the processing unit 1. As shown in FIG. 24, the substrate processing apparatus 100 of this embodiment further includes an etching liquid recovery unit 12. The etching liquid recovery unit 12 has a recovery pipe 121. The chamber 2 further accommodates a portion of the recovery pipe 121. Note that another portion of the recovery pipe 121 is accommodated in a corresponding one of the fluid boxes 100B described with reference to FIG. 1.

[0202] As shown in FIG. 24, the guard 10 includes a cup 11. The cup 11 has a guard 11a and a liquid receiving portion 11b. The guard 11a has a generally cylindrical shape and receives the etching liquid (processing liquid) splashed from the rotating substrate W. The liquid receiving portion 11b is connected to the lower end of the guard 11a. The liquid receiving portion 11b is annular and forms an annular groove. The etching liquid (processing liquid) received by the guard 11a flows down to the liquid receiving portion 11b due to its own weight. As a result, the etching liquid (processing liquid) is collected in the liquid receiving portion 11b. One end of a recovery pipe 121 is connected to the liquid receiving portion 11b. The etching liquid (processing liquid) collected in the liquid receiving portion 11b flows into the recovery pipe 121 due to its own weight.

[0203] The etching liquid (processing liquid) that flows into the recovery pipe 121 is returned to a container (processing liquid container 45) that stores the etching liquid (processing liquid), and is reused for processing (etching) the substrate W, as will be described with reference to FIG.

[0204] Next, the substrate processing apparatus 100 will be further described with reference to Fig. 25. Fig. 25 is a schematic diagram showing the configuration of the substrate processing apparatus 100 of this embodiment. In detail, Fig. 25 shows the internal configuration of the fluid box 100B described with reference to Fig. 1. Note that in Fig. 25, for simplification of the drawing, the temperature sensor 421, concentration sensor 422, mixing valve 424, flow meter 425, and heater 426 described with reference to Fig. 7 are omitted.

[0205] First, the etching liquid supply unit 4 will be further described with reference to FIG. 25. As shown in FIG. 25, the etching liquid supply unit 4 further includes a circulation pipe 43, a circulation pump 44, and a processing liquid container 45. The processing liquid container 45 stores an etching liquid (processing liquid). The processing liquid container 45 is replaceably disposed in the fluid box 100B (FIG. 1). When the processing liquid container 45 is replaced, a new processing liquid container 45 storing the etching liquid is disposed in the fluid box 100B. As described with reference to FIG. 24, the etching liquid (processing liquid) that has flowed into the recovery pipe 121 is returned to the processing liquid container 45 and reused for processing (etching) the substrate W. If the etching liquid is continuously reused, the etching amount decreases, so it is necessary to replace the processing liquid container 45 in the fluid box 100B as appropriate. The processing liquid container 45 is, for example, a tank.

[0206] The circulation pipe 43 and the circulation pump 44 are housed in a fluid box 100B (FIG. 1). Both ends of the circulation pipe 43 are connected to a processing liquid container 45. The circulation pump 44 is attached to the circulation pipe 43. The circulation pump 44 is controlled by a control device 101 (control unit 102).

[0207] When the circulation pump 44 is driven, the etching liquid stored in the processing liquid container 45 flows into the circulation pipe 43 from one end thereof and circulates through the circulation pipe 43. The etching liquid guided to the other end of the circulation pipe 43 returns to the processing liquid container 45 from the other end thereof.

[0208] The other end of the first supply pipe 42 is connected to the circulation pipe 43. The etching liquid circulating through the circulation pipe 43 flows into the first supply pipe 42. When the valve 423 is in an open state, the etching liquid that has flowed from the circulation pipe 43 into the first supply pipe 42 is guided to the first nozzle 41 (FIG. 24). When the valve 423 is in a closed state, the flow of the etching liquid that has flowed from the circulation pipe 43 into the first supply pipe 42 is stopped by the valve 423.

[0209] Next, the etching liquid recovery unit 12 will be further described with reference to Figure 25. As shown in Figure 25, the etching liquid recovery unit 12 further includes a recovery pump 122. The recovery pump 122 is housed in the fluid box 100B (Figure 1).

[0210] The recovery pump 122 is attached to the recovery pipe 121. The recovery pump 122 is controlled by the control device 101 (controller 102). The recovery pump 122 drives the etching liquid so that the etching liquid that has flowed into the recovery pipe 121 from the processing unit 1 (liquid receiver 11b) flows through the recovery pipe 121. The other end of the recovery pipe 121 is connected to the processing liquid container 45, and by driving the recovery pump 122, the etching liquid is guided to the other end of the recovery pipe 121 and recovered from the other end of the recovery pipe 121 into the processing liquid container 45. Therefore, the etching liquid after being used for processing (etching) the substrate W can be reused.

[0211] Next, a method for generating learning data executed by the substrate processing apparatus 100 of this embodiment will be described with reference to Fig. 8 and Fig. 26. Fig. 26 is a flowchart showing the method for generating learning data in this embodiment. In detail, Fig. 26 shows the processing executed by the control unit 102 when generating learning data. The method for generating learning data in this embodiment further includes processing in steps S101 and S102 in addition to steps S1 to S9 described with reference to Fig. 9.

[0212] In this embodiment, after the substrate W to be learned is carried in (step S2), the control unit 102 increments the processed number counter value (step S101). That is, the control unit 102 increments the processed number counter value by "1." The processed number counter value at the time of generating the learning data indicates the number of processed substrates W to be learned. The control unit 102 stores the incremented processed number counter value in the memory unit 103.

[0213] Specifically, the control unit 102 counts the number of processed substrates W of the learning object from the start of use of the processing liquid (etching liquid) used to process (etch) the learning object substrate W. More specifically, the control unit 102 starts counting from the start of processing (etching) the learning object substrate W using the etching liquid in the processing liquid container 45 (FIG. 25). The control unit 102 resets the processed number counter value when the processing liquid container 45 (FIG. 25) is replaced with a new processing liquid container 45. Here, the processed number counter value is reset to "0".

[0214] When the control unit 102 acquires the processing amount (etching amount) (step S7), it associates the acquired processing amount with the processed number counter value stored in the memory unit 103 and stores it in the memory unit 103 (step S102).

[0215] The timing at which the processes of steps S101 and S102 are performed is not limited to the timing shown in Fig. 26. The timing at which the processes of steps S101 and S102 are performed is not particularly limited as long as it is performed before the next substrate W is loaded.

[0216] Next, the process of generating learning data (step S8) will be described with reference to Fig. 8 and Fig. 27. Fig. 27 is a flowchart showing the process of generating learning data in this embodiment.

[0217] In this embodiment, the process of generating learning data further includes the process of step S110 in addition to steps S81 to S84 described with reference to Fig. 11. After generating learning data (step S81), the control unit 102 classifies the learning data into classes based on predetermined classification rules. The classification rules are stored in the storage unit 103.

[0218] In this embodiment, the classification rules prescribe that the learning data be classified into classes according to the number of processed substrates W (processed substrate counter value) of the learning target substrates W. More specifically, the classification rules prescribe that the processed substrate counter value be divided into a plurality of ranges (a plurality of classes), and the control unit 102 classifies the learning data by class by referring to the processed substrate counter value associated with the learning data.

[0219] The timing at which the process of step S110 is executed is not limited to the timing shown in Fig. 27. The timing at which the process of step S110 is executed is not particularly limited as long as it is executed after the generation of the learning data.

[0220] Next, a training dataset table TB30 for managing the training dataset 135 will be described with reference to Fig. 28. Fig. 28 is a diagram showing an example of the training dataset table TB30 of this embodiment.

[0221] 28, the training dataset table TB30 has a class information field TB31 and a training dataset information field TB32. A range of processed sheet counter values ​​is predefined for each field of the class information field TB31. In this embodiment, the control unit 102 creates the training dataset table TB10 described with reference to FIG. 13 for each class. The control unit 102 associates each field of the training dataset information field TB32 with the training dataset table TB10 corresponding to each field of the class information field TB31 (each range of processed sheet counter values).

[0222] Each value defined in each field of the class information field TB31 may be a value (default value) defined in advance in the recipe 131, or may be a value input by the operator operating the input unit 104.

[0223] Next, a method for generating a trained model 136 executed by the substrate processing apparatus 100 of this embodiment will be described with reference to Fig. 14 and Fig. 29. Fig. 29 is a flowchart showing the method for generating a trained model 136 in this embodiment. In detail, Fig. 29 shows the processing executed by the control unit 102 when generating the trained model 136. The method for generating a trained model 136 of this embodiment includes the processing of steps S111 to S118. The processing shown in Fig. 29 is executed for each training dataset 135 created for each class.

[0224] First, the control unit 102 acquires one of the plurality of learning data sets 135 from the storage unit 103 (step S111). At this time, the control unit 102 acquires learning data to which a first flag has been assigned from the plurality of learning data, similar to step S11 in FIG.

[0225] The control unit 102 causes the display unit 105 to display an input screen showing information about the acquired training data set 135 (step S112). Hereinafter, this input screen may be referred to as a "selection instruction input screen." Specifically, the selection instruction input screen displays information showing each training data item included in the acquired training data set 135. More specifically, the information showing the training data indicates the learning target speed information #A and the processing amount (etching amount). Hereinafter, the acquired training data set 135 may be referred to as a "current training data set 135."

[0226] The selection instruction input screen is a screen that allows the worker to select one of the learning data included in the current learning data set 135. The worker operates the input unit 104 to input an instruction to select one of the learning data displayed on the selection instruction input screen. For example, the worker may select one of the learning data based on information indicating the processing amount (etching amount). Hereinafter, the learning data selected by the worker may be referred to as "reference learning data."

[0227] 29, after the selection instruction input screen is displayed (step S112), the control unit 102 determines whether an instruction to select reference learning data has been input (step S113). The control unit 102 waits until an instruction to select reference learning data has been input (No in step S113).

[0228] When the control unit 102 determines that an instruction to select reference training data has been input (Yes in step S113), it selects one of the training data included in the current training dataset 135 as the reference training data in accordance with the instruction from the operator (step S114).

[0229] After selecting the reference training data, the control unit 102 selects training data whose correlation with the reference training data satisfies a predetermined condition from the training data included in the current training data set 135 (step S115).

[0230] In this embodiment, the control unit 102 selects training data based on the root mean square error (RMSE) between the processing amount (etching amount) of the reference training data and the processing amount (etching amount) of each training data included in the current training data set 135. Specifically, the control unit 102 calculates the value of the root mean square error using the following formula, and selects training data whose value of the root mean square error is equal to or greater than a threshold value.

number

[0231] In the above formula, "n" represents the number of measurement points of the thickness measurement unit 8 (thickness measurement device 85), and "y i " indicates the distribution of the processing amount of the reference learning data (the processing amount at each measurement point), and "y i "(Hat)" indicates the distribution of the amount of processing of the learning data (the amount of processing at each measurement point), and i indicates the number of pieces of learning data included in the current learning data set 135.

[0232] In this embodiment, the error (distance) between the processing amount (etching amount) of the reference training data and the processing amount (etching amount) of each training data set included in the current training data set 135 is calculated using the root mean square error (RMSE), but it may be calculated based on other indices. For example, it may be calculated using the mean absolute error (MAE) or the mean square error (MSE).

[0233] 29, after selecting the learning data (step S115), the control unit 102 learns the selected learning data (step S116) to generate learned parameters (step S117). More specifically, the learned parameters are output from the learning program 134. The control unit 102 generates a learned model 136 based on the acquired learned parameters (step S118), and the process shown in FIG. 29 ends.

[0234] In this embodiment, the process shown in Fig. 29 is executed for each class. As a result, a trained model 136 is generated for each class.

[0235] Next, a substrate processing method executed by the substrate processing apparatus 100 of this embodiment will be described with reference to Figures 16 and 30. Figure 30 is a flowchart showing part of the substrate processing method of this embodiment. In detail, Figure 30 shows part of the processing executed by the control unit 102 when etching the substrate W to be processed. The substrate processing method of this embodiment further includes processing of step S120 in addition to the processing of steps S21 to S32 described with reference to Figures 17 and 18.

[0236] In this embodiment, after the substrate W to be processed is carried in (step S22), the control unit 102 increments the processed number counter value (step S120). The processed number counter value during processing of the substrate W to be processed indicates the number of processed substrates W to be processed. The control unit 102 stores the incremented processed number counter value in the memory unit 103.

[0237] Specifically, the control unit 102 counts the number of processed substrates W to be processed from the start of use of the processing liquid (etching liquid) used to process (etch) the substrates W to be processed. More specifically, the control unit 102 starts counting from the start of processing (etching) the substrates W to be processed using the etching liquid in the processing liquid container 45 (Figure 25). When the processing liquid container 45 (Figure 25) is replaced with a new processing liquid container 45, the control unit 102 resets the processed number counter value. Here, the processed number counter value is reset to "0".

[0238] Next, the process of acquiring processing speed information (step S25) will be described with reference to Fig. 16 and Fig. 31. Fig. 31 is a flowchart showing the process of acquiring processing speed information.

[0239] 31, after acquiring the target processing amount (target etching amount) (step S24 in FIG. 30), the control unit 102 selects one of the classes based on a predetermined selection rule (step S251). The selection rule is stored in the storage unit 103.

[0240] In this embodiment, the selection rule specifies that one of a plurality of classes is selected depending on the number of processed substrates W to be processed. The control unit 102 selects one of the plurality of classes by identifying the range (classification) corresponding to the processed substrate counter value stored in the memory unit 103 (step S120 in FIG. 30) from among the plurality of ranges (classifications) of the processed substrate counter value described with reference to FIGS.

[0241] The control unit 102 inputs a target processing amount (target etching amount) into the trained model 136 corresponding to the selected class (step S252), thereby causing the trained model 136 to output processing speed information. As a result, the control unit 102 acquires the processing speed information (step S253).

[0242] The fourth embodiment of the present invention has been described above with reference to FIGS. 1, 8, 14, 16, and 24 to 31. According to this embodiment, as with the first to third embodiments, it is possible to reduce the burden on the worker involved in developing the scan speed information. Furthermore, according to this embodiment, it is possible to acquire processing speed information according to the number of substrates W processed since the start of use of the etching liquid. Therefore, it is possible to move the first nozzle 41 at a more appropriate scan speed. Therefore, it is possible to process (etch) the substrates W more appropriately.

[0243] Furthermore, in this embodiment, as described with reference to Fig. 29, the trained model 136 is generated using training data whose correlation with the reference training data satisfies a predetermined condition. As a result, more appropriate processing speed information can be acquired. Therefore, the substrate W can be processed (etched) more appropriately.

[0244] In this embodiment, the trained model 136 is generated using training data whose correlation with the reference training data satisfies predetermined conditions, but the processing shown in Figure 15 may be performed for each class to generate the trained model 136 for each class.

[0245] Furthermore, although machine learning has been described in this embodiment, additional learning can also be performed in the same way.

[0246] [Embodiment 5] Next, a fifth embodiment of the present invention will be described with reference to Figures 1, 8, 16, 24, 25, 27, and 31 to 35. However, differences from the first to fourth embodiments will be described, and descriptions of the same aspects as the first to fourth embodiments will be omitted. Unlike the first to fourth embodiments, the fifth embodiment classifies the trained models 136 according to the number of lots. The number of substrates W included in one lot is, for example, 25.

[0247] First, a method for generating learning data executed by the substrate processing apparatus 100 of this embodiment will be described with reference to Fig. 8 and Fig. 32. Fig. 32 is a flowchart showing the method for generating learning data in this embodiment. In detail, Fig. 32 shows the processing executed by the control unit 102 when generating learning data. The method for generating learning data in this embodiment further includes the processing of steps S130 and S140 in addition to steps S1 to S9 described with reference to Fig. 9.

[0248] In this embodiment, after the learning target substrate W is loaded (step S2), the control unit 102 executes a counting process (step S130). By executing the counting process, the control unit 102 acquires the learning time lot number, which is the number of lots from the start of use of the processing liquid (etching liquid) used to process the learning target substrate W. More specifically, the control unit 102 acquires the learning time lot number from the start of processing (etching) the learning target substrate W using the etching liquid in the processing liquid container 45 (FIG. 25). The control unit 102 stores the learning time lot number acquired by the counting process in the memory unit 103.

[0249] Furthermore, when the control unit 102 acquires the processing amount (etching amount) (step S7), it associates the acquired processing amount with the learning lot number stored in the memory unit 103 and stores it in the memory unit 103 (step S140).

[0250] The timing at which the processes of step S130 and step S140 are performed is not limited to the timing shown in Fig. 32. The timing at which the processes of step S130 and step S140 are performed is not particularly limited as long as it is performed before the next substrate W is loaded.

[0251] Next, an example of the counting process will be described with reference to Fig. 8 and Fig. 33. Fig. 33 is a flowchart showing an example of the counting process. The counting process shown in Fig. 33 includes steps S131 to S134.

[0252] When the control unit 102 starts the counting process, it first increments the processed sheet counter value (step S131). In the following description, the processed sheet counter value may be referred to as a "first counter value."

[0253] The control unit 102 determines whether the incremented first counter value is equal to or less than a certain value (step S132). The certain value indicates the number of substrates W included in one lot.

[0254] If the incremented first counter value exceeds the certain value, the control unit 102 determines that the incremented first counter value is not equal to or less than the certain value (No in step S132), and increments the lot number counter value (step S133). Hereinafter, the lot number counter value may be referred to as the "second counter value." The second counter value when generating the learning data indicates the number of lots during learning.

[0255] After incrementing the second counter value, the control unit 102 changes the first counter value to "1" (step S134) and ends the counting process shown in Fig. 33. Alternatively, if the control unit 102 determines that the incremented first counter value is equal to or less than a certain value (Yes in step S132), it ends the counting process shown in Fig. 33.

[0256] When the processing liquid container 45 (FIG. 25) is replaced with a new processing liquid container 45, the control unit 102 changes the second counter value to "1."

[0257] Next, the process of generating learning data (step S8) will be described with reference to Figures 8 and 27. In this embodiment, the classification rule specifies that the learning data be classified into classes according to the number of lots during learning (second counter value). The control unit 102 classifies the learning data by class with reference to the second counter value associated with the learning data (step S110).

[0258] Next, a training dataset table TB40 for managing the training dataset 135 will be described with reference to Fig. 34. Fig. 34 is a diagram showing an example of the training dataset table TB40 of this embodiment.

[0259] As shown in Fig. 34, the training dataset table TB40 has a class information field TB41 and a training dataset information field TB42. The number of training lots is predefined for each field in the class information field TB41. In this embodiment, the control unit 102 creates the training dataset table TB10 described with reference to Fig. 13 for each class. The control unit 102 associates each field of the training dataset information field TB42 with the training dataset table TB10 corresponding to each field (number of training lots) in the class information field TB41.

[0260] Next, a substrate processing method executed by the substrate processing apparatus 100 of this embodiment will be described with reference to Figures 16 and 35. Figure 35 is a flowchart showing part of the substrate processing method of this embodiment. In detail, Figure 35 shows part of the processing executed by the control unit 102 when etching the substrate W to be processed. The substrate processing method of this embodiment further includes processing of step S150 in addition to the processing of steps S21 to S32 described with reference to Figures 17 and 18.

[0261] In this embodiment, after the substrate W to be processed is loaded (step S22), the control unit 102 executes a counting process (step S150). By executing the counting process, the control unit 102 acquires the number of lots during processing, which is the number of lots from the start of use of the processing liquid (etching liquid) used to process the substrate W to be processed. More specifically, the control unit 102 acquires the number of lots during processing from the start of processing (etching) the substrate W to be processed using the etching liquid in the processing liquid container 45 (FIG. 25). More specifically, the control unit 102 can acquire the number of lots during processing (second counter value) by executing a process similar to the process described with reference to FIG. 33. The control unit 102 stores the number of lots during processing (second counter value) acquired by the counting process in the memory unit 103.

[0262] Next, the process of acquiring processing speed information (step S25) will be described with reference to Figures 16 and 31. In this embodiment, the selection rules stipulate that one of a plurality of classes is selected depending on the number of lots at processing. The control unit 102 selects one of the plurality of classes by identifying the number of lots at processing stored in the memory unit 103 (step S150 in Figure 35) from the number of lots at learning time described with reference to Figures 32 to 34.

[0263] The control unit 102 inputs a target processing amount (target etching amount) into the trained model 136 corresponding to the selected class (step S252), thereby causing the trained model 136 to output processing speed information. As a result, the control unit 102 acquires the processing speed information (step S253).

[0264] The trained model 136 may be generated by executing the process shown in FIG. 29 for each class, or may be generated by executing the process shown in FIG. 15 for each class.

[0265] The fifth embodiment of the present invention has been described above with reference to FIGS. 1, 8, 16, 24, 25, 27, and 31 to 35. According to this embodiment, as with the first to fourth embodiments, it is possible to reduce the burden on the worker involved in developing the scan speed information. Furthermore, according to this embodiment, it is possible to acquire processing speed information corresponding to the number of substrates W processed since the start of use of the etching liquid. Therefore, it is possible to process (etch) the substrates W more appropriately.

[0266] Although machine learning has been described in this embodiment, additional learning can also be performed in the same way.

[0267] 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.

[0268] 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.

[0269] For example, in the embodiment described with reference to FIGS. 1 to 35, the thickness measurement unit 8 measured the thickness of the target object TG, but the thickness measurement unit 8 may also measure the thickness of the substrate W. More specifically, the thickness measurement unit 8 may measure the thickness of the substrate W to be learned and the thickness of the substrate W to be processed. Here, the thickness of the substrate W indicates the combined thickness of the substrate body and the substance when a substance is formed on the surface of the substrate body. Note that, when the thickness measurement unit 8 measures the thickness of the substrate W, the etching amount indicates the difference between the thickness of the substrate W before the etching process and the thickness of the substrate W after the etching process. Furthermore, the target thickness (target thickness distribution) indicates the target thickness (target thickness distribution) of the substrate W to be processed.

[0270] 1 to 35, the substrate W is a semiconductor wafer, but the substrate W is not limited to a semiconductor wafer. For example, the substrate W may be a substrate for a liquid crystal display device, 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.

[0271] Furthermore, in the embodiment described with reference to FIGS. 1 to 35, the processing liquid is an etching liquid, but the processing liquid is not limited to an etching liquid. The processing liquid may be any liquid that processes the substrate W. For example, the processing liquid may be a removal liquid that removes the target object TG. By using the removal liquid, it is possible to perform a process of removing a specific film or a process of removing a specific film that has foreign matter mixed in. When a resist removal process is performed on the substrate W, the removal liquid is, for example, a sulfuric acid / hydrogen peroxide mixture (SPM). The resist removal process is a process of removing resist from the surface of a semiconductor substrate.

[0272] Furthermore, in the embodiment described with reference to FIGS. 1 to 35, the process performed by the substrate processing apparatus 100 is an etching process, but the process performed by the substrate processing apparatus 100 is not limited to an etching process. For example, the process may be a film formation process. When the substrate processing apparatus 100 performs a film formation process, the processing liquid is, for example, SPM or ozone water. In this case, an oxide film is formed on the substrate W. Furthermore, the processing amount indicates the amount of film formation.

[0273] Furthermore, in the embodiment described with reference to Figures 1 to 35, a clamping chuck that clamps the substrate W was described as the configuration for holding the substrate W, but a vacuum chuck may also be adopted as the configuration for holding the substrate W.

[0274] 1 to 35, the display unit 105 displays an error screen to notify the operator of an error, but the error may be notified to the operator by voice. In this case, the substrate processing apparatus 100 may be provided with a speaker.

[0275] 1 to 35, the scan speed information indicates the scan speed setting value for each position included in the movement section of first nozzle 41, but the scan speed information may indicate only one scan speed setting value. In this case, first nozzle 41 moves at a constant speed from the start position to the end position of the movement section of first nozzle 41.

[0276] Furthermore, in the embodiment described with reference to FIGS. 1 to 35 , the control unit 102 associates training data with a flag (first flag or second flag) and stores the data in the storage unit 103. However, if the control unit 102 determines that the training data (processing amount) cannot be used for machine learning or additional training, the control unit 102 may delete the training data from the storage unit 103. For example, the control unit 102 may delete the speed information field TB11 and the processing amount field TB12 corresponding to the training data to which the second flag has been assigned from the training dataset table TB10 when generating the training dataset 135 or when performing machine learning. Similarly, the control unit 102 may delete the speed information field TB21 and the processing amount field TB22 corresponding to the additional training data to which the second flag has been assigned from the additional training dataset table TB20 when generating the additional training dataset or when performing additional training.

[0277] 1 to 35, the control unit 102 determines whether or not the learning data is usable for machine learning or additional learning after generating the learning data, but the control unit 102 may determine whether or not the processing amount is usable for machine learning or additional learning before generating the learning data. In this case, if the control unit 102 determines that the processing amount is not usable for machine learning or additional learning, the control unit 102 proceeds to the next process (next step) without generating learning data using the processing amount.

[0278] Furthermore, in the embodiment described with reference to Figures 1 to 35, the control unit 102 determined whether or not learning data (processing amount) is available, but the process of determining whether or not learning data (processing amount) is available may be omitted.

[0279] Furthermore, in the embodiment described with reference to Figures 1 to 35, the training dataset 135 and the additional training dataset are stored in the memory unit 103 and the memory unit 202 in table format, but the format in which the training dataset 135 and the additional training dataset are stored in the memory unit 103 and the memory unit 202 does not have to be in table format.

[0280] 1 to 35, the learning target speed information and processing speed information defined the movement speed of the first nozzle 41, but the learning target speed information and processing speed information may define the relative movement speed between the first nozzle 41 and the substrate W. When the substrate W rotates, the learning target speed information and processing speed information define the relative movement speed between the surface of the rotating substrate W and the first nozzle 41. For example, the relative movement speed indicates the sum of the speed component (vector) of the first nozzle 41 and the speed component (vector) of the portion of the rotating substrate W that faces the first nozzle 41. Here, the speed component of the substrate W indicates the speed in the circumferential direction.

[0281] Furthermore, in the embodiment described with reference to FIGS. 1 to 35, the first nozzle 41 turns, but the first nozzle 41 may move linearly.

[0282] Furthermore, in the embodiment described with reference to Figures 1 to 35, the first nozzle 41 was a scan nozzle, but the first nozzle 41 may also be a fixed nozzle. In this case, the processing unit 1 is provided with a substrate moving mechanism that moves the substrate W, instead of the nozzle moving mechanism 6. The learning target speed information and processing speed information specify the movement speed of the substrate W, instead of the movement speed of the first nozzle 41. The trained model 136 outputs substrate speed information indicating the movement speed of the substrate W. The control unit 102 controls the substrate moving mechanism based on the substrate speed information. Similar to the scan speed information, the substrate speed information may indicate the movement speed of the substrate W set for each position (each substrate position) that divides the movement section in which the substrate W moves into multiple sections. Note that the substrate W may rotate or move linearly.

[0283] 1 to 35, only the first nozzle 41 moves during processing of the substrate W, but the first nozzle 41 and the substrate W may also move. In this case, the processing unit 1 further includes a substrate moving mechanism for moving the substrate W in addition to the nozzle moving mechanism 6. The speed information to be learned and the speed information during processing define the moving speed of the substrate W together with the moving speed of the first nozzle 41. The trained model 136 outputs substrate speed information that defines the moving speed of the substrate W in addition to the scan speed information. The control unit 102 controls the nozzle moving mechanism 6 based on the scan speed information and controls the substrate moving mechanism based on the substrate speed information.

[0284] 1 to 35, the substrate processing apparatus 100 includes the thickness measuring unit 8, but the thickness measuring unit 8 may be omitted. When the substrate processing apparatus 100 does not include the thickness measuring unit 8, the thickness is measured by a thickness measuring device external to the substrate processing apparatus 100.

[0285] 1 to 35, the target throughput (target etching amount) was selected as the explanatory variable, but other factors that affect the throughput may be added to the explanatory variables. Therefore, the learning dataset 135 may indicate the relationship between the throughput and other factors that affect the throughput, in addition to the relationship between the throughput (etching amount) and the scan speed information. In this case, the target throughput and other factors that affect the throughput are input to the trained model 136 as explanatory variables, and the trained model 136 outputs the scan speed information. The other factors that affect the throughput may include, for example, at least one of the rotation speed of the substrate W, the temperature of the processing liquid, the concentration of the processing liquid, and the discharge flow rate of the processing liquid.

[0286] 1 to 35, scan speed information (processing speed information) was selected as the dependent variable, but other factors that affect the processing amount may also be added to the dependent variable. Therefore, the trained model 136 may output, as the dependent variable, other factors that affect the processing amount in addition to the scan speed information. For example, the trained model 136 may output, in addition to the scan speed information, at least one of the rotation speed of the substrate W, the temperature of the processing liquid, the concentration of the processing liquid, and the discharge flow rate of the processing liquid.

[0287] 1 to 35, the training program 134 is a program for executing an algorithm that trains the training dataset 135 (plurality of training data) to generate trained parameters and incorporates the trained parameters into the inference program 133 to generate the trained model 136. However, the training program 134 may also be a program for executing an algorithm that trains the training dataset 135 (plurality of training data) to generate a trained model including trained parameters. In other words, the process of generating trained parameters and incorporating the trained parameters into the inference program 133 may be omitted. Alternatively, the training program 134 may be a program for executing an algorithm that trains the training dataset 135 (plurality of training data) to generate a trained model that does not include trained parameters. [Industrial Applicability]

[0288] The present invention is useful in the field of substrate processing. [Explanation of symbols]

[0289] 1 Processing Unit 2 chambers 3 Spin chuck 4. Etching solution supply unit 5 Spin motor section 6 Nozzle movement mechanism 8 Thickness measurement section 9 Probe movement mechanism 41 No. 1 nozzle 100 Substrate processing apparatus 101 Control device 102 Control section 103 Storage section 200 Learning Device 201 Processing section 202 Storage section TG Object W substrate

Claims

1. A method for generating learning data for additional learning using a substrate processing apparatus that supplies a processing liquid to a substrate and processes the substrate based on an output of a trained model generated by training learning data, comprising: calculating a difference between a thickness distribution of the substrate before processing and a target thickness distribution of the substrate, and acquiring a target processing amount, which is a target amount of processing amount indicating an amount of the substrate processed by the processing; performing the process on the substrate to obtain the processing amount; determining whether the processing amount and the target processing amount match within an allowable range; a step of assigning a flag to the processing amount indicating that the processing amount is to be used as learning data for the additional learning when it is determined that the processing amount does not match the target processing amount within the allowable range; A method for generating training data for additional learning, including:

2. A method for generating a trained model used in a substrate processing apparatus that supplies a processing liquid to a substrate to be processed based on an output of the trained model and performs processing on the substrate, comprising: supplying the processing liquid to a learning target substrate and performing the processing on the learning target substrate; acquiring a processing amount indicating an amount of processing of the learning target substrate by the processing; generating learning data including the processing amount of the learning target substrate; a learning step of classifying the training data into classes based on predetermined classification rules, and learning the training data for each class to generate the trained model; Including, A method for generating a trained model, wherein the classification rules stipulate that the training data is classified into each class according to the number of processed substrates to be trained.

3. A method for generating a trained model used in a substrate processing apparatus that supplies a processing liquid to a substrate to be processed based on an output of the trained model and performs processing on the substrate, comprising: supplying the processing liquid to a learning target substrate and performing the processing on the learning target substrate; acquiring a processing amount indicating an amount of processing of the learning target substrate by the processing; generating learning data including the processing amount of the learning target substrate; a learning step of classifying the training data into classes based on predetermined classification rules, and learning the training data for each class to generate the trained model; Including, In the step of performing the process on the learning substrate, a learning lot number is acquired, which is the number of lots since the start of use of the processing liquid used in the process; A method for generating a trained model, wherein the classification rules stipulate that the training data is classified into each class according to the number of training lots.

Images