Substrate conveyance system and substrate conveyance method

The substrate transport system employs machine learning to automatically control transport robots, addressing the inefficiencies of manual teaching and positional deviations, thereby improving operational precision and reducing operator burden.

WO2024248047A1PCT designated stage expired Publication Date: 2024-12-05KAWASAKI JUKOGYO KK
View PDF 6 Cites 0 Cited by

Patent Information

Application Number
PCT/JP2024/019722
Authority / Receiving Office
WO · WO
Patent Type
Applications
Current Assignee / Owner
Priority Date
2023-06-01
Filing Date
2024-05-29
Publication Date
2024-12-05

AI Technical Summary

Technical Problem

Conventional substrate transport methods require manual teaching of transfer operations, which is labor-intensive and inefficient for operators, as they need to correct positional deviations and adapt to different substrate configurations.

Method used

A substrate transport system utilizing a detection unit and a control unit with machine learning capabilities to generate a trained model that automatically controls the transport robot based on detection results, reducing operator effort and improving positional accuracy.

Benefits of technology

The system significantly reduces operator workload by enabling automatic control of substrate transport operations, enhancing precision and efficiency by correcting deviations and adapting to various configurations without manual intervention.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure JP2024019722_05122024_PF_FP_ABST
    Figure JP2024019722_05122024_PF_FP_ABST
Patent Text Reader

Abstract

This substrate conveyance system (100) comprises a detection unit (43) that detects at least one of a substrate (210) conveyed by a substrate conveyance robot (20) and a substrate placement unit (30a) on which the substrate (210) is placed. The substrate conveyance system (100) further comprises a control unit (60) that generates, by machine learning, a trained model that, when inputted with the detection result from the detection unit (43), outputs conveyance control information for controlling the operation of having the substrate (210) conveyed by the substrate conveyance robot (20).
Need to check novelty before this filing date? Find Prior Art

Description

Substrate transport system and substrate transport method

[0001] The present disclosure relates to a substrate transfer system and a substrate transfer method.

[0002] Conventionally, substrate transfer methods for transferring substrates have been disclosed. Japanese Patent No. 6640321 discloses a substrate transfer method in which a transfer robot disposed in a substrate transfer chamber transfers a substrate to a mounting table in a substrate processing chamber where the substrate is processed. The transfer robot has a fork on which the substrate is placed. A position sensor is disposed in the substrate transfer chamber to detect the substrate placed on the fork. In the substrate transfer method disclosed in Japanese Patent No. 6640321, the position sensor detects the substrate placed on the fork, and the transfer path of the substrate by the transfer robot is corrected to correct the positional deviation of the substrate relative to the fork.

[0003] Patent No. 6640321

[0004] However, when performing a substrate transport operation, it is necessary to teach the substrate transport robot, which is the transport robot that transports the substrate, the transport operation. In the above-mentioned Japanese Patent No. 6640321, while the positional deviation of the substrate relative to the fork that holds the substrate is detected by a position sensor, the transport operation of the substrate transport robot must be taught manually. Therefore, since the transport operation must be taught for each configuration of the substrate transport system, such as the position of the substrate placement unit, the time and effort required to teach the substrate transport robot the transport operation is considered to be a burden for the operator. Therefore, it is desirable to reduce the time and effort required by the operator to control the substrate transport operation of the substrate transport robot.

[0005] This disclosure has been made to solve the above-mentioned problems, and one purpose of this disclosure is to provide a substrate transport system and a substrate transport method that can reduce the effort required by an operator to control the substrate transport operation by a substrate transport robot.

[0006] A substrate transport system according to a first aspect of this disclosure includes a substrate transport robot, a detection unit that detects at least one of a substrate transported by the substrate transport robot and a substrate placement portion on which the substrate is placed, and a control unit that uses the detection results of at least one of the substrate and the substrate placement portion by the detection unit as input and generates a trained model through machine learning that outputs transport control information for controlling the substrate transport operation by the substrate transport robot.

[0007] As described above, the substrate transport system according to the first aspect of this disclosure uses, as input, the detection results of at least one of the substrate and the substrate placement unit by the detection unit, and generates, through machine learning, a trained model that outputs transport control information for controlling the substrate transport operation by the substrate transport robot. This makes it possible to automatically control the substrate transport operation based on the output from the trained model generated through machine learning. As a result, the effort required by an operator to control the substrate transport operation by the substrate transport robot can be reduced compared to when the substrate placement unit is manually instructed to perform the substrate transport operation.

[0008] A substrate transport method according to a second aspect of this disclosure detects at least one of a substrate transported by a substrate transport robot and a substrate placement portion on which the substrate is placed, and generates a trained model using machine learning to input the detection results of at least one of the substrate and the substrate placement portion and output transport control information for controlling the substrate transport operation by the substrate transport robot.

[0009] As described above, a substrate transport method according to a second aspect of this disclosure uses machine learning to generate a trained model that uses detection results of at least one of the substrate and the substrate placement unit as input and outputs transport control information for controlling the substrate transport operation by the substrate transport robot. This makes it possible to automatically control the substrate transport operation based on the output from the trained model generated by machine learning. As a result, a substrate transport method can be provided that reduces the effort required by an operator to control the substrate transport operation by the substrate transport robot, compared to manually teaching the substrate placement unit the substrate transport operation.

[0010] According to the present disclosure, it is possible to reduce the effort required by an operator to control the substrate transport operation by a substrate transport robot.

[0011] FIG. 1 is a top view showing a substrate processing system in which a substrate transfer system according to an embodiment of the present disclosure is disposed; FIG. 2 is a block diagram showing the configuration of the substrate transfer system according to an embodiment of the present disclosure; FIG. 3 is a perspective view showing a schematic configuration of the substrate transfer system; FIG. 4 is a diagram showing an example of a captured image of a FOUP captured by an imaging unit arranged in a transfer chamber; FIG. 5 is a diagram showing an example of a captured image of a load lock unit captured by an imaging unit arranged in a transfer chamber; FIG. 6 is a diagram showing an example of a captured image of an aligner captured by an imaging unit arranged in a substrate transfer robot; FIG. 7 is a diagram for explaining output of transfer control information using a trained model; FIG. 8 is a diagram showing an example of positional deviation of a substrate relative to a substrate placement unit; FIG. 9 is a diagram showing an example of positional deviation of a substrate relative to a substrate holding hand; and FIG. 10 is a flow diagram for explaining a substrate transfer method according to the present embodiment.

[0012] Hereinafter, embodiments of the present disclosure will be described with reference to the drawings.

[0013] A substrate transfer system 100 according to the present embodiment will be described with reference to FIGS.

[0014] (Substrate Processing System) As shown in FIG. 1 , the substrate transfer system 100 is disposed in a substrate processing system 200. The substrate processing system 200 includes a processing device 201 that processes a substrate 210. The processing device 201 includes a substrate transfer robot 202, a load lock unit 203, multiple processing module units 204, and a transfer chamber 205. The substrate transfer robot 202 is a horizontal articulated robot that transfers the substrate 210. The load lock unit 203 is a room for loading and unloading the substrate 210. That is, the load lock unit 203 is a chamber in which the substrate 210 is placed so as to transfer the substrate 210 to the processing device 201. The processing device 201 of the substrate processing system 200 includes two load lock units 203. The load lock units 203 are connected to the substrate transfer system 100. The load lock unit 203 includes a substrate placement unit 203 a on which the substrate 210 is placed. In the processing module section 204, processing such as resist coating and etching is performed on the substrate 210. A substrate transfer robot 202 is disposed in the transfer chamber 205. In the transfer chamber 205, the substrate transfer robot 202 transfers the substrate 210 between the load lock section 203 and the processing module section 204. The load lock section 203 is an example of a processing standby section.

[0015] The substrate transfer system 100 transfers substrates 210 between a FOUP 101 (Front Opening Unify Pod) and a processing device 201 in a substrate processing system 200 in which processing is performed on the substrates 210. The substrate transfer system 100 is an EFEM (Equipment Front End Module). The FOUP 101 is a container that stores a plurality of substrates 210. The FOUP 101 has a substrate placement portion 101a on which the substrates 210 are placed. In the FOUP 101, the plurality of substrates 210 are arranged vertically side by side. The FOUP 101 is an example of a substrate storage container. The substrates 210 are, for example, disk-shaped silicon wafers.

[0016] 2, the substrate transfer system 100 includes a transfer chamber 10, a substrate transfer robot 20, an aligner 30, an imaging unit 40, an illumination unit 50, and a control unit 60. The substrate transfer system 100 includes a target member 71, a target member 72, and a target member 73. The substrate transfer robot 20 includes a horizontally articulated robot arm 21 and a substrate holding hand 22. The imaging unit 40 includes an imaging unit 41, an imaging unit 42, an imaging unit 43, and an imaging unit 44. The imaging unit 40 is an example of a detection unit.

[0017] As shown in FIG. 3 , the transfer chamber 10 is a rectangular parallelepiped housing. The transfer chamber 10 has a front surface 11 on the Y1 direction side and a back surface 12 on the Y2 direction side. The transfer chamber 10 also has a top surface 13 on the vertically upward side, i.e., the Z1 direction side. The transfer chamber 10 transports substrates 210 within the transfer chamber 10. The transfer chamber 10 houses a substrate transfer robot 20, an aligner 30, an imaging unit 40, an illumination unit 50, and a control unit 60. Three load port units 11a, to which FOUPs 101 are attached, are arranged horizontally on the front surface 11 on the Y1 direction side of the transfer chamber 10. The load port units 11a each have an opening / closing mechanism. The load port units 11a open and close the doors of the attached FOUPs 101. Two load lock units 203 are connected to the back surface 12 on the Y2 direction side of the transfer chamber 10. An opening / closing mechanism is also disposed between the load lock unit 203 and the rear surface 12. Note that Fig. 3 shows a state in which a FOUP 101 is attached to one of the three load port units 11a on the X1 direction side.

[0018] The substrate transfer robot 20 is disposed in the transfer chamber 10. The substrate transfer robot 20 performs a transfer operation of a substrate 210 in the transfer chamber 10. Specifically, in the substrate transfer robot 20, a substrate holding hand 22, which is an end effector that holds the substrate 210, is disposed at the tip of a robot arm 21. The substrate transfer robot 20 performs a transfer operation of the substrate 210 between a FOUP 101 attached to the load port unit 11a and a load lock unit 203. The substrate transfer robot 20 transfers the substrate 210 while the substrate holding hand 22 holds the substrate 210 along a horizontal plane.

[0019] The aligner 30 aligns the substrate 210 transported by the substrate transport robot 20. The aligner 30 is disposed in the transport chamber 10, close to the X2 direction, which is one side in the X direction (left-right direction). The aligner 30 has a substrate mounting part 30a on which the substrate 210 is mounted. The aligner 30 rotates the substrate 210 mounted on the substrate mounting part 30a along a horizontal plane, thereby aligning the rotational position of the substrate 210 along the circumferential direction of the main surface of the substrate 210. The aligner 30 also detects the center position of the substrate 210.

[0020] In the substrate transfer system 100, an unprocessed substrate 210 held by the substrate transfer robot 20 is transferred from the FOUP 101 to the aligner 30. Then, the substrate 210 aligned by the aligner 30 is transferred again to the load lock unit 203 by the substrate transfer robot 20. Furthermore, the substrate 210 after being processed in the processing device 201 is transferred from the load lock unit 203 to the FOUP 101 by the substrate transfer robot 20. The transport operation of the substrate 210 by the substrate transport robot 20 includes both an operation of removing the substrate 210 by removing it from the substrate mounting portion 101a of the FOUP 101, the substrate mounting portion 203a of the load lock portion 203, and the substrate mounting portion 30a of the aligner 30, and an operation of transporting the substrate 210 by placing it on the substrate mounting portion 101a of the FOUP 101, the substrate mounting portion 203a of the load lock portion 203, and the substrate mounting portion 30a of the aligner 30.

[0021] The imaging unit 40 is, for example, a two-dimensional camera. Each of the imaging units 41, 42, 43, and 44 of the imaging unit 40 has an imaging element such as a charge coupled device (CCD) and a complementary metal oxide semiconductor (CMOS). The imaging units 41, 42, and 43 are disposed above the transfer chamber 10. The imaging unit 44 is disposed in the substrate holding hand 22 of the substrate transfer robot 20. Specifically, three imaging units 41 are disposed on the upper side, in the Z1 direction, of the rear surface 12 within the transfer chamber 10 so as to correspond to the three load port units 11a to which the FOUPs 101 are attached. The imaging units 41 capture images of the FOUPs 101. Two imaging units 42 are arranged on the upper side, in the Z1 direction, of the front surface 11 inside the transfer chamber 10, corresponding to the two load lock units 203. The imaging unit 41 images the load lock unit 203. The imaging unit 43 is arranged above the aligner 30 on the ceiling surface 13 inside the transfer chamber 10. The imaging unit 43 images the aligner 30. The imaging unit 44 is arranged on the Z1 direction side of the substrate holding hand 22. The imaging unit 44 is fixed to the substrate holding hand 22 so as to move integrally with the substrate holding hand 22. The imaging units 41, 42, and 43 are examples of transfer chamber imaging units. The imaging unit 44 is an example of a robot imaging unit.

[0022] As shown in Fig. 4 , a captured image 80 is acquired by imaging unit 40. Fig. 4 shows an example of captured image 80 in which FOUP 101 is captured by imaging unit 41 of imaging unit 40. The captured images 80 captured by each of imaging units 41, 42, 43, and 44 of imaging unit 40 are transmitted to control unit 60.

[0023] The image 80 captured by the imaging unit 41 includes the substrates 210 stored in the FOUP 101, the substrate placement portion 101a in the FOUP 101, and the target member 71. The target member 71 is a member that indicates the position of the substrate placement portion 101a in the FOUP 101. In other words, the imaging unit 41 captures an image of the FOUP 101, thereby capturing an image of the substrates 210 stored in the FOUP 101, the substrate placement portion 101a in the FOUP 101, and the target member 71. In the FOUP 101, the target members 71 are arranged, for example, on the front surface 11 within the transfer chamber 10, one on each side of one load port portion 11a in the X direction, which is the left-right direction.

[0024] 5 , a captured image 80 of the load lock unit 203 captured by the imaging unit 42 includes the substrate 210 placed in the load lock unit 203, the substrate placement portion 203a in the load lock unit 203, and the target member 72. That is, the imaging unit 42 captures an image of the load lock unit 203 from inside the transfer chamber 10, thereby capturing an image of the substrate 210 placed in the load lock unit 203, the substrate placement portion 203a in the load lock unit 203, and the target member 72. In the load lock unit 203, the target members 72 are arranged, for example, on the back surface 12 within the transfer chamber 10, two on each side of the load lock unit 203 in the X direction, which is the left-right direction, so that the target members 72 are located at the four corners of the load lock unit 203.

[0025] 6 , a captured image 80 of the aligner 30 captured by the imaging unit 43 includes the substrate 210 placed on the aligner 30, the substrate placement portion 30a of the aligner 30, and the target member 73. That is, the imaging unit 43 captures an image of the aligner 30 from the ceiling surface 13 inside the transfer chamber 10, thereby capturing an image of the substrate 210 placed on the substrate placement portion 30a, the substrate placement portion 30a, and the target member 73. In the aligner 30, the target members 72 are arranged, for example, at the four corners of the substrate placement portion 30a of the aligner 30 within the transfer chamber 10.

[0026] Furthermore, since the substrate transport robot 20 is disposed in the transport chamber 10, the substrate transport robot 20 may be included in the captured image 80 captured by each of the imaging units 41, 42, and 43. In the example of Fig. 6, the substrate transport robot 20 operating to bring the substrate holding hand 22 holding the substrate 210 closer to the aligner 30 is included in the captured image 80 captured by the imaging unit 43. Note that the imaging units 41, 42, and 43 are disposed in a fixed state in the transport chamber 10, and the imaging target areas to be imaged by the imaging units 41, 42, and 43 are fixed.

[0027] 7 , the captured image 80 captured by the imaging unit 44 is an image captured with the imaging direction being the direction from the base end to the tip end of the substrate holding hand 22 of the substrate transport robot 20. For example, when the substrate transport robot 20 performs an operation to remove a substrate 210 from a FOUP 101, the captured image 80 captured by the imaging unit 44 includes the substrate placement portion 101 a of the FOUP 101, the substrate 210 placed on the substrate placement portion 101 a, and the target member 71. The captured image 80 by the imaging unit 44 also includes the tip end portion of the substrate holding hand 22 of the substrate transport robot 20. The imaging direction of the imaging unit 44 is changed in accordance with the operation of the substrate holding hand 22.

[0028] As shown in FIG. 3 , the illumination unit 50 emits illumination light in the transfer chamber 10. The illumination unit 50 includes, for example, a light-emitting diode (LED) that emits yellow light as illumination light. The illumination unit 50 emits the illumination light at the timing when the image capture unit 40 captures an image. The illumination unit 50 emits, for example, diffused illumination light throughout the entire interior of the transfer chamber 10, thereby irradiating the area to be captured by the image capture unit 40 with the illumination light. The illumination unit 50 is disposed on the upper side, i.e., the Z1 direction side, of the transfer chamber 10, and on each of the side surfaces on the X1 direction and the X2 direction. A captured image 80 captured by the image capture unit 40 while illuminated with illumination light from the illumination unit 50 is transmitted to the control unit 60.

[0029] As shown in FIG. 1 , the control unit 60 includes a machine learning unit 61. The control unit 60 controls the operation of each unit of the substrate transfer system 100. In this embodiment, the control unit 60 is a robot control unit that controls the operation of the substrate transfer robot 20. That is, the control unit 60 is a robot controller. The control unit 60 includes, for example, a calculation device such as a CPU (Central Processing Unit). The control unit 60 also includes memories such as a RAM (Random Access Memory) and a ROM (Read Only Memory), and a storage device such as a hard disk. The control unit 60 executes control processing by the calculation device based on programs, parameters, and the like stored in the storage device. In this embodiment, the control unit 60 controls the imaging unit 40 to capture an image 80 and also controls the substrate 210 transport operation by the substrate transfer robot 20. For example, the control unit 60 includes a main CPU that controls the capture of the captured image 80, controls machine learning (described later), and controls the generation of command values ​​for the transport operation of the substrate 210, and a servo CPU that controls, based on the command values ​​for the transport operation from the main CPU, the drive current output to a servo motor that serves as a drive source for operating the substrate transport robot 20. Note that the control unit 60 may be configured from a single CPU.

[0030] (Control of Transport Operation) As shown in FIG. 8 , in this embodiment, the control unit 60 acquires transport control information 81 for controlling the transport operation of the substrate transport robot 20 based on a trained model 62 generated by machine learning using images 80 captured by the imaging unit 40. The trained model 62 receives images 80 captured by each of the imaging units 41, 42, 43, and 44 of the imaging unit 40 as input, and outputs transport control information 81 for controlling the transport operation of the substrate transport robot 20. The transport control information 81 includes instruction information for aligning coordinate axes between the substrate transport robot 20 and the positions of the substrate placement unit 101a, the substrate placement unit 203a, and the substrate placement unit 30a. The transport control information 81 includes information indicating the position of the substrate 210 on the coordinate axes of the substrate transport robot 20, and information indicating the positions of the substrate placement unit 101a, the substrate placement unit 203a, and the substrate placement unit 30a. The transport control information 81 includes, for example, a command value as a control amount for operating the robot arm 21 of the substrate transport robot 20. Specifically, the transport control information 81 includes a command value for controlling the rotation angle of a motor that serves as a drive source for the robot arm 21. The trained model 62 is generated by the machine learning unit 61 of the control unit 60 and is stored in a storage device of the control unit 60.

[0031] <Generation of Trained Model> The control unit 60 generates the trained model 62 through machine learning in the machine learning unit 61. The machine learning unit 61 generates the trained model 62 through machine learning using deep reinforcement learning, such that the trained model 62 receives the captured image 80 as input and outputs the transport control information 81. The machine learning unit 61 performs machine learning using, for example, DQN (Deep Q-Network), which is reinforcement learning using deep learning. In deep reinforcement learning, the machine learning unit 61 generates the trained model 62 by learning the transport control information 81 through repeated trials based on the captured image 80.

[0032] In deep reinforcement learning, the control unit 60 acquires in advance, as target teaching images, for example, captured images 80 in which the center position of the substrate holding hand 22 is located at the center positions of the substrate placement units 101a, 203a, and 30a, or captured images 80 in which the substrate holding hand 22 holds the center position of the placed substrate 210. Then, in deep reinforcement learning by the machine learning unit 61 of the control unit 60, the captured images 80 captured by each of the imaging units 41, 42, 43, and 44 are used as input, and machine learning is performed by operating the substrate transport robot 20 through repeated trials, for example, by evaluating a value indicating the degree of match between the input captured images 80 and the previously acquired teaching images. For example, the degree of match is evaluated based on the positional relationship between each of the target members 71, 72, and 73 and the substrate 210 or the substrate holding hand 22. In deep reinforcement learning, the trained model 62 is generated by repeated trials to obtain a higher evaluation.

[0033] In this embodiment, each of the target members 71, 72, and 73 is a plate-shaped member having a predetermined circular shape in the captured image 80, for example, so that it can be identified by the control unit 60 in machine learning that generates the trained model 62. Furthermore, each of the target members 71, 72, and 73 has a color different from the surrounding area, so that it can be identified in the captured image 80. For example, each of the target members 71, 72, and 73 is black. Furthermore, the control unit 60 detects the board 210 included in the captured image 80.

[0034] The machine learning unit 61 of the control unit 60 performs machine learning using deep reinforcement learning, thereby repeatedly trying and learning the transport operations of the substrate transport robot 20, which include both the operation of unloading the substrate 210 placed on each of the substrate placement units 101 a, 203 a, and 30 a, and the operation of loading the substrate 210 onto each of the substrate placement units 101 a, 203 a, and 30 a. For example, in learning to acquire transport control information 81 for the transport operation for the FOUP 101, an image 80 captured by the imaging unit 41 that images the FOUP 101 and an image 80 captured by the imaging unit 44 arranged on the substrate transport robot 20 are acquired as detection results of the substrate 210 and the substrate placement unit 101 a in the FOUP 101. Then, the machine learning unit 61 uses the captured image 80 as an input, which is the detection result, and performs machine learning using deep reinforcement learning with DQN to train the trained model 62 so as to output the transport control information 81. For example, the control unit 60 trains the trained model 62 by performing a preset number of trial operations.

[0035] When learning the operation of loading the substrate mounting part 101a, the operation of the substrate transport robot 20 is learned so that the position of the substrate mounting part 101a indicated by the target member 71 in the captured image 80 matches the position of the substrate holding hand 22 in the captured image 80, or the position of the substrate holding hand 22 detected based on the substrate 210 held by the substrate holding hand 22. When learning the operation of unloading the substrate 210 mounted on the substrate mounting part 101a, the operation of the substrate transport robot 20 is learned so that the position of the substrate mounting part 101a detected based on the target member 71 in the captured image 80, or the substrate 210 mounted on the substrate mounting part 101a, matches the position of the substrate holding hand 22 detected based on the substrate holding hand 22 in the captured image 80. The machine learning unit 61 of the control unit 60 learns the transport control information 81 as a control amount in the transport operation for the load lock unit 203 and the transport control information 81 as a control amount in the transport operation for the aligner 30 using deep reinforcement learning, similar to the FOUP 101.

[0036] That is, in this embodiment, the imaging section 40 images the substrate 210, the substrate placement section 101a and the target member 71, the substrate placement section 203a and the target member 72, the substrate placement section 30a and the target member 73, and the substrate transport robot 20 so as to detect the substrate 210, the substrate placement section 101a, the substrate placement section 203a, the substrate placement section 30a, and the substrate transport robot 20. In other words, in this embodiment, the captured image 80 is a detection result in which the substrate 210, the substrate placement section 101a, the substrate placement section 203a, the substrate placement section 30a, and the substrate transport robot 20 have been detected. Then, the control unit 60 acquires transport control information 81 for controlling the transport operation of the substrate 210 to each of the FOUP 101, the load lock unit 203, and the aligner 30, by using a trained model 62 generated by machine learning based on the images 80 captured by each of the imaging units 41, 42, 43, and 44 obtained as detection results.

[0037] In the present embodiment, the control unit 60 controls the operation of the substrate transport robot 20 based on the transport control information 81 acquired using the trained model 62 generated by machine learning based on the captured image 80. That is, in the substrate transport system 100 of the present embodiment, the transport operation of the substrate transport robot 20 is controlled based on the acquired transport control information 81, and thereby the transport operation of the substrate transport robot 20 is taught by automatic teaching using machine learning based on the captured image 80. Therefore, in the present embodiment, information indicating absolute position coordinates indicating the position of the substrate 210 and the positions of the substrate mounting unit 101 a, the substrate mounting unit 203 a, and the substrate mounting unit 30 a is not taught by an operator, and the transport control information 81, which is output from the trained model 62 to which the captured image 80 is input, is acquired, and thereby a relative positional relationship indicating how the substrate transport robot 20 should be operated appropriately from the state in which the current captured image 80 is acquired, is automatically taught.

[0038] <Correction of deviations in transport operation> In this embodiment, the control unit 60 acquires transport control information 81 by using a trained model 62 generated by machine learning, and controls the transport operation of the substrate transport robot 20 while correcting deviations in the transport operation by performing the transport operation of the substrate 210 based on the acquired transport control information 81.

[0039] For example, as shown in FIG. 9 , the substrate 210 placed on the substrate placement unit 101 a, the substrate placement unit 203 a, and the substrate placement unit 30 a may be misaligned relative to the substrate placement unit 101 a, the substrate placement unit 203 a, and the substrate placement unit 30 a. Note that FIG. 9 illustrates the misalignment of the substrate 210 placed on the substrate placement unit 203 a, and the substrate 210 when no misalignment occurs is indicated by a dotted line. If a transfer operation of the misaligned substrate 210 is performed based on a preset value, it is expected that a deviation will occur in the movement of the substrate 210 during the transfer operation. Therefore, the machine learning unit 61 of the control unit 60 uses the trained model 62 while performing the transfer operation to control the transfer operation of the substrate transfer robot 20 so as to correct the deviation in the transfer operation based on the transfer control information 81 acquired from the captured image 80. The control unit 60 controls the transport operation while correcting the misalignment, thereby performing the transport operation to eliminate the misalignment even if a misalignment occurs in the substrate 210 placed on the substrate placing unit 101a, the substrate placing unit 203a, and the substrate placing unit 30a.

[0040] 10 , there may be a case where the substrate 210 held by the substrate holding hand 22 is misaligned with respect to the substrate holding hand 22 of the substrate transport robot 20. Note that in FIG. 10 , similar to FIG. 9 , the substrate 210 when no misalignment occurs is indicated by a dotted line. Even in such a case, the control unit 60 controls the transport operation so as to correct the misalignment of the transport operation of the substrate transport robot 20 based on the transport control information 81 acquired from the captured image 80 by using the trained model 62 while performing the transport operation, so as to prevent any misalignment in the movement of the substrate 210 during the transport operation.

[0041] When the machine learning unit 61 generates the trained model 62 through machine learning using the captured image 80 as input, the control unit 60 controls the transport operation by the substrate transport robot 20 at a speed slower than the speed of a normal transport operation until learning of the trained model 62 is completed. That is, the control unit 60 acquires transport control information 81 for controlling the transport operation of the substrate transport robot 20 by repeatedly attempting the transport operation while performing machine learning using the captured image 80 as input at a relatively slow speed.

[0042] Furthermore, when learning of the trained model 62 by deep reinforcement learning is completed, the control unit 60 causes the substrate transport robot 20 to perform a transport operation at a relatively high operating speed based on the transport control information 81 output from the trained model 62. For example, the control unit 60 determines that learning of the trained model 62 is completed when a preset number of trial operations are completed.

[0043] For example, when sequentially transporting multiple substrates 210 stored in a FOUP 101, the control unit 60 performs machine learning using deep reinforcement learning to generate a trained model 62 by performing a predetermined number of trials based on the captured images 80 while reducing the operating speed of the transport operation to a relatively low level before transporting the substrates 210. The control unit 60 then performs machine learning again to update the trained model 62 by performing another predetermined number of transport operations while reducing the operating speed of the transport operation to a relatively low level based on the transport control information 81 generated by the trained model 62. After the training of the trained model 62 is completed, the control unit 60 controls the transport operation of the substrate transport robot 20 at a normal operating speed, which is a relatively high operating speed. As an example, the trained model 62 is generated by performing machine learning using a predetermined number of trials before the substrate transport system 100 is shipped from a factory, and the trained model 62 is trained by performing another predetermined number of trials after the substrate transport system 100 is installed.

[0044] (Substrate Transfer Method) Next, a control process of the substrate transfer method by the substrate transfer system 100 of this embodiment will be described with reference to Fig. 11. The control process of the substrate transfer method of this embodiment is executed by the control unit 60 of the substrate transfer system 100.

[0045] First, in step S1, a captured image 80 is acquired. Specifically, a captured image 80 is acquired as a detection result, capturing images of the substrate 210, the substrate placement units 101a, 203a, 30a, the target members 71, 72, 73, and the substrate transport robot 20, so that the substrate 210, the substrate placement units 101a, 203a, 30a, and the substrate transport robot 20 are detected. Note that in step S1, the captured image 80 is acquired while the substrate transport robot 20 is operating at a relatively slow operating speed.

[0046] Next, in step S2, machine learning is performed to generate the trained model 62. Specifically, machine learning is performed using deep reinforcement learning, with the captured image 80 acquired in step S1 as input, so that transport control information 81 for controlling the transport operation of the substrate transport robot 20 is output.

[0047] Next, in step S3, it is determined whether or not learning of the trained model 62 has been completed. If it is determined that learning of the trained model 62 has been completed, the process proceeds to step S4. If it is not determined that learning of the trained model 62 has been completed, the process returns to step S1. In step S3, it is determined whether or not learning of the trained model 62 has been completed, for example, based on whether machine learning has been performed a predetermined number of times.

[0048] In step S4, the operating speed of the substrate transport robot 20 is set to be increased relatively greatly.

[0049] Next, in step S5, the transport operation of the substrate transport robot 20 is started based on the trained model 62 generated by the control processing from step S1 to step S3. That is, the transport operation is started based on the transport control information 81 according to the trained model 62 generated by machine learning using the captured image 80.

[0050] Effect of the Present Embodiment As described above, the substrate transport system 100 uses, as input, the captured image 80 as a result of detection of the substrate 210 and the substrate mounting units 101a, 203a, and 30a, and generates, through machine learning, the trained model 62 that outputs transport control information 81 for controlling the transport operation of the substrate 210 by the substrate transport robot 20. This makes it possible to automatically control the transport operation of the substrate 210 based on the output from the trained model 62 generated through machine learning. As a result, the effort required by an operator to control the transport operation of the substrate 210 by the substrate transport robot 20 can be reduced compared to when the substrate mounting units 101a, 203a, and 30a are manually instructed to transport the substrate 210.

[0051] The control unit 60 generates the trained model 62 through machine learning using deep reinforcement learning. This makes it possible to control the transport operation of the substrate transport robot 20 using the trained model 62 trained through machine learning using deep reinforcement learning, which is reinforcement learning using deep learning. Therefore, in the machine learning for generating the trained model 62, by extracting feature amounts when performing reinforcement learning using deep learning, the effort required by the operator to control the transport operation of the substrate 210 by the substrate transport robot 20 can be further reduced.

[0052] The control unit 60 receives the captured image 80 as a detection result as input, and generates, by machine learning, a trained model 62 that outputs transport control information 81 including teaching information for aligning the coordinate axes of the substrate transport robot 20 with those of the substrate placement unit 101a, the substrate placement unit 203a, and the substrate placement unit 30a. This allows automatic teaching of the transport operation by the substrate transport robot 20 based on the transport control information 81 including the teaching information output from the trained model 62, thereby reducing the effort required by an operator to control the transport operation of the substrate 210 by the substrate transport robot 20.

[0053] The imaging unit 40, which serves as a detection unit, detects the substrate 210, the substrate placement unit 101a, the substrate placement unit 203a, and the substrate placement unit 30a, as well as the substrate transport robot 20, by capturing images of the substrate 210. The control unit 60 receives as input a captured image 80 as a result of the detection of the substrate 210, the substrate placement unit 101a, the substrate placement unit 203a, and the substrate placement unit 30a, and a captured image 80 as a result of the detection of the substrate transport robot 20, and generates a trained model 62 that outputs transfer control information 81. This allows the trained model 62 to learn the substrate 210, the substrate placement unit 101a, the substrate placement unit 203a, and the substrate placement unit 30a, as well as the captured image 80 captured by the substrate transport robot 20, and outputs the transfer control information 81. This improves the accuracy of the transfer control information 81 output by the trained model 62. This allows the transfer operation of the substrate transport robot 20 to be controlled with greater accuracy.

[0054] The control unit 60 also serves as a robot control unit that controls the transport operation of the substrate transport robot 20. The control unit 60 acquires transport control information 81 by using the trained model 62, and controls the transport operation of the substrate transport robot 20 while correcting deviations in the transport operation by performing the transport operation of the substrate 210 based on the acquired transport control information 81. As a result, even if deviations in the transport operation, such as deviations in the operation of the substrate transport robot 20 or deviations in the placement position of the substrate 210, are generated, the control unit 60 can control the transport operation of the substrate transport robot 20 to correct the deviations by using the trained model 62. Therefore, by automatically correcting deviations in the transport operation, the transport operation by the substrate transport robot 20 can be performed with high accuracy.

[0055] The substrate transport system 100 includes an imaging unit 40 that captures an image of at least one of the substrate 210, the substrate mounting units 101a, 203a, and 30a, and the target members 71, 72, and 73 that indicate the positions of the substrate mounting units 101a, 203a, and 30a. The control unit 60 receives an image 80 captured by the imaging unit 40 as an input and generates a trained model 62 that outputs transport control information 81 through machine learning. As a result, the substrate 210, the substrate mounting units 101a, 203a, and 30a, and the target members 71, 72, and 73 can be more easily detected by imaging with the imaging unit 40 than by detection using a position sensor such as a photoelectric sensor. Therefore, the accuracy of identification using the trained model 62 generated by machine learning using the captured image 80 as input can be improved, thereby improving the accuracy of the substrate 210 transport operation by the substrate transport robot 20.

[0056] The substrate transport system 100 includes target members 71, 72, and 73 that indicate the positions of the substrate mounting units 101a, 203a, and 30a. The imaging unit 40 captures images of the target members 71, 72, and 73. The control unit 60 receives a captured image 80 including the target members 71, 72, and 73 acquired as a detection result, and generates a trained model 62 through machine learning that outputs transport control information 81. As a result, even if the shapes of the substrate mounting units 101a, 203a, and 30a are unclear in the captured image 80, the trained model 62 can acquire the transport control information 81 based on the captured image 80 including the target members 71, 72, and 73. Therefore, when the shapes of the substrate mounting portion 101a, the substrate mounting portion 203a, and the substrate mounting portion 30a are unclear in the captured image 80, it is possible to prevent a decrease in the accuracy of control of the transport operation using the trained model 62 generated by machine learning based on the captured image 80.

[0057] The target members 71, 72, and 73 have predetermined shapes in the captured image 80 so that they can be identified in the machine learning that generates the trained model 62. As a result, since the target members 71, 72, and 73 have predetermined shapes, the target members 71, 72, and 73 can be easily detected from the captured image 80. Therefore, the positions of the substrate placement unit 101a, the substrate placement unit 203a, and the substrate placement unit 30a in the captured image 80 can be easily detected, thereby improving the accuracy of the trained model 62 generated in the machine learning. Therefore, the transport operation of the substrate transport robot 20 using the trained model 62 can be controlled with even greater accuracy.

[0058] The substrate transfer system 100 includes a transfer chamber 10 in which a substrate transfer robot 20 is disposed and in which a substrate 210 is transferred. The imaging unit 40 is disposed in the transfer chamber 10. This allows the transfer operation of the substrate transfer robot 20 to be controlled using a trained model 62 based on a captured image 80 captured within the transfer chamber 10. Therefore, the background in the captured image 80 can be kept constant, allowing the transfer operation of the substrate transfer robot 20 to be controlled more accurately using the trained model 62.

[0059] The imaging unit 40 includes imaging units 41, 42, and 43 as transfer chamber imaging units arranged in the transfer chamber 10, and imaging unit 44 as a robot imaging unit arranged in the substrate transfer robot 20. The control unit 60 receives images 80 captured by the imaging units 41, 42, and 43 and images 80 captured by the imaging unit 44 as input, and generates a trained model 62 through machine learning that outputs transfer control information 81. As a result, since the imaging units 40 are arranged in each of the transfer chamber 10 and the substrate transfer robot 20, machine learning can be performed to generate the trained model 62 using the multi-angled images 80 as input. This improves the accuracy of the trained model 62, allowing for more accurate control of the transfer operation of the substrate transfer robot 20 using the trained model 62.

[0060] Imaging units 41, 42, and 43 serving as transfer chamber imaging units are disposed above transfer chamber 10. This prevents imaging units 41, 42, and 43 from interfering with the operation of substrate transfer robot 20, and allows images of substrate placement units 101a, 203a, and 30a to be taken from above. This improves the visibility of substrate placement units 101a, 203a, and 30a in captured image 80, allowing more accurate control of the transfer operation of substrate transfer robot 20 using trained model 62 based on captured image 80.

[0061] The substrate transfer robot 20 performs a transfer operation of the substrate 210 to a FOUP 101, which serves as a substrate storage container in which the substrate 210 is stored, in the transfer chamber 10. The imaging unit 41 captures an image of at least one of the substrate 210 stored in the FOUP 101, the substrate placement portion 101a in the FOUP 101 on which the substrate 210 is placed, and a target member 71 indicating the position of the substrate placement portion 101a in the FOUP 101. The control unit 60 receives as input a captured image 80 including at least one of the substrate 210 stored in the FOUP 101, the substrate placement portion 101a in the FOUP 101 on which the substrate 210 is placed, and the target member 71 indicating the position of the substrate placement portion 101a in the FOUP 101, and generates, by machine learning, a trained model 62 that outputs transfer control information 81. As a result, by using the trained model 62, transport control information 81 for controlling the transport operation of the substrate transport robot 20 for the FOUP 101 can be obtained, thereby reducing the effort required by the operator to control the transport operation for the FOUP 101.

[0062] The substrate transport robot 20 transports the substrate 210 between a FOUP 101 serving as a substrate storage container and a load lock unit 203 serving as a processing standby unit on which the substrate 210 is placed to be delivered to a processing device 201 that processes the substrate 210. The imaging unit 42 captures an image of at least one of the substrate 210 placed on the substrate placement unit 203 a in the load lock unit 203, the substrate placement unit 203 a in the load lock unit 203, and a target member 72 indicating the position of the substrate placement unit 203 a in the load lock unit 203. The control unit 60 receives as input a captured image 80 including at least one of the substrate 210 placed on the substrate placement unit 203 a in the load lock unit 203, the substrate placement unit 203 a in the load lock unit 203, and the target member 72 indicating the position of the substrate placement unit 203 a in the load lock unit 203, and generates by machine learning a trained model 62 that outputs transport control information 81. This allows the transport operation of the substrate transport robot 20 relative to the load lock unit 203 to be controlled by using the trained model 62 generated by machine learning based on the captured image 80, thereby reducing the effort required by the operator to control the transport operation relative to the load lock unit 203.

[0063] The control unit 60 also serves as a robot control unit that controls the transport operation of the substrate transport robot 20. When generating a trained model 62 through machine learning using a captured image 80 as an input as a detection result, the control unit 60 controls the transport operation while keeping the operating speed of the substrate transport robot 20 relatively slow until learning of the trained model 62 is complete. By doing so, by keeping the operating speed relatively slow while the trained model 62 is being learned, blurring of the captured image 80 acquired for machine learning can be suppressed, thereby improving the accuracy of the trained model 62 generated by machine learning. As a result, the accuracy of control of the transport operation based on the trained model 62 can be improved.

[0064] [Modifications] The embodiments disclosed herein should be considered to be illustrative and not restrictive in all respects. The scope of the present disclosure is defined by the claims, not by the description of the above-mentioned embodiments, and includes all modifications (modifications) within the meaning and scope of the claims.

[0065] For example, in the above embodiment, the control unit 60 that controls the transport operation of the substrate transport robot 20 generates the trained model 62 by performing machine learning using DQN, which is reinforcement learning using deep learning. However, the present disclosure is not limited to this. In the present disclosure, the trained model may be learned by a control device that is a learning device separate from the control unit that controls the operation of the substrate transport robot. In this case, transport control information acquired by the learning device based on the trained model is output to the control unit that controls the transport operation, and the transport operation of the substrate transport robot is controlled based on the transport control information. Furthermore, the machine learning algorithm for generating the trained model is not limited to DQN. For example, reinforcement learning using a multilayer fully connected neural network, which is a deep reinforcement learning method other than DQN, may be performed. Furthermore, machine learning of the trained model may be performed using reinforcement learning such as Q-learning, which does not use deep learning. Furthermore, in machine learning, completion of learning and updating of the trained model may be determined when a predetermined operating time has elapsed, rather than after a predetermined number of trials.

[0066] Furthermore, in the above embodiment, an example was described in which machine learning of the trained model 62 is performed using the captured image 80 as an input, which is a detection result obtained by capturing images of the substrate 210, the substrate placement portion 101a of the FOUP 101, the substrate placement portion 203a of the load lock unit 203, the substrate placement portion 30a of the aligner 30, the target member 71, the target member 72, and the target member 73, and the substrate transport robot 20, by the imaging unit 40 as a detection unit, and which outputs transport control information 81. However, the present disclosure is not limited to this. In the present disclosure, the captured image used for training does not necessarily include the substrate transport robot. Furthermore, the captured image may include at least one of the substrate, the substrate placement portion, and the target member. For example, when learning the substrate unloading operation, machine learning may be performed based on captured images of the substrate so as to detect only the placed substrate. Furthermore, when learning the loading operation for placing a substrate, machine learning may be performed based on an image captured of either the substrate placement portion or the target member to detect the substrate placement portion. That is, the substrate transport system according to the present disclosure does not need to include a target member. The substrate placement portion itself may be detected in the captured image without a target member being provided. Furthermore, instead of imaging by an imaging unit, a position sensor such as a photoelectric sensor that emits detection light may be provided as the detection unit. In this case, a trained model that uses the detection result of the position sensor as input and outputs transport control information is generated by machine learning.

[0067] In the above embodiment, an example has been shown in which the substrate transport robot 20 controls the transport operation including both the operation of loading the substrate 210 and the operation of unloading the substrate 210 by learning both the operation of loading the substrate 210 and the operation of unloading the substrate 210, but the present disclosure is not limited to this. In the present disclosure, the robot may learn transport control information for controlling either the operation of loading the substrate or the operation of unloading the substrate.

[0068] Furthermore, in the above embodiment, an example was described in which the transport operation of the substrate transport robot 20 was controlled while correcting deviations in the transport operation by performing the transport operation of the substrate 210 based on the transport control information 81. However, the present disclosure is not limited to this. In the present disclosure, deviations in the transport operation may be corrected by computational processing, such as a rule-based algorithm, based on the substrate detection results, separate from learning by machine learning. Furthermore, an example was described in which the transport operation was controlled to correct deviations in the transport operation of the substrate transport robot based on transport control information acquired from captured images as the detection results using a trained model while the transport operation was being performed. However, the trained model may not be used while the transport operation was being performed. That is, a control variable for performing the transport operation may be acquired in advance by machine learning, and the control variable based on the transport control information acquired by machine learning may be used as a set value to control the transport operation. That is, the transport operation may be automatically taught by machine learning, and the transport operation itself may be performed based on a set value as a fixed value set by the automatic teaching.

[0069] Although the above embodiment illustrates an example in which the target members 71, 72, and 73 are circular, plate-like members, the present disclosure is not limited thereto. In the present disclosure, the target members may be spherical, cylindrical, or prismatic with a polygonal cross section, or may be a solid having a predetermined shape such as a cross. The target members may also be marker members such as AR markers. The target members may be removed after learning is completed. Target members indicating the position of not only the substrate placement unit but also the substrate transport robot or the substrate holding hand may be placed. In this case, machine learning may be performed by detecting the substrate transport robot based on the detection results or captured images of the target members. Alternatively, target members transported by the substrate transport robot may be placed in place of the substrates, and machine learning may be performed based on the detection results or captured images of the target members placed in place of the substrates. The target members may also be placed at positions on the substrate placement unit where the substrates are placed, such as inside a substrate storage container. The number of target members for one substrate placement portion may be one or more.

[0070] Furthermore, in the above embodiment, an example was shown in which the transfer chamber 10 is provided, in which the substrate transfer robot 20, the imaging unit 40, and the illumination unit 50 are disposed, but the present disclosure is not limited to this. The present disclosure does not necessarily require a transfer chamber. Furthermore, either the imaging unit or the illumination unit may be disposed outside the transfer chamber in which the substrate is transferred. For example, when the substrate transfer method according to the present disclosure is performed in a transfer chamber of a processing apparatus that processes substrates, the ceiling member of the transfer chamber of the processing apparatus may be made of a transparent material such as glass, and the imaging unit and the illumination unit may be disposed above the exterior of the transfer chamber. Furthermore, when the ceiling member of the transfer chamber is made of a transparent material, an external light source may be used, thereby eliminating the need for an illumination unit.

[0071] Furthermore, in the above embodiment, an example is shown in which the imaging units 41, 42, and 43 are arranged above the transfer chamber 10 and the imaging unit 44 is arranged in the substrate holding hand 22 of the substrate transfer robot 20. However, the present disclosure is not limited to this. In the present disclosure, the imaging unit does not have to be arranged in the substrate transfer robot. Furthermore, the imaging unit does not have to be arranged in the transfer chamber. Furthermore, when the imaging unit is arranged in the transfer chamber, the imaging unit may be arranged below the transfer chamber. Furthermore, the imaging unit may include a stereo camera or a 3D camera such as a TOF (Time of Flight) camera. Furthermore, multiple imaging units may be arranged for one imaging target. For example, multiple imaging units may be arranged to image one substrate placement unit. Furthermore, the imaging unit may be configured to detect infrared light. In this case, an illumination unit that irradiates infrared light and a target member that reflects infrared light may be arranged. Furthermore, the imaging unit may be arranged inside the substrate storage container. The imaging unit may also be disposed inside the processing standby unit.

[0072] In the above embodiment, the operation speed is relatively slowed down until the learning of the trained model 62 is completed, but the present disclosure is not limited to this. In the present disclosure, the operation speed of the transport operation may not be slowed down even when training of the trained model is performed.

[0073] Furthermore, in the above embodiment, an example was shown in which the substrate transport robot 20 had a horizontally articulated robot arm 21, but the present disclosure is not limited to this. In the present disclosure, the substrate transport robot may have a vertically articulated robot arm instead of a horizontally articulated robot arm. Furthermore, the substrate transport robot may have a linear movement mechanism such as a slide mechanism.

[0074] Furthermore, in the above embodiment, the substrate transfer system 100 of the present disclosure is an EFEM that transfers the substrate 210 to the processing device 201, but the present disclosure is not limited to this. In the present disclosure, the substrate transfer system may be something other than an EFEM. For example, the substrate transfer system may be a stocker or sorter that transfers substrates from one substrate storage container to another. Furthermore, the substrate transfer system may transfer substrates transferred by a belt conveyor or the like to a substrate storage container or a substrate placement unit, rather than to a substrate storage container. Furthermore, the substrate transfer method of the present disclosure may be applied to a system that transfers substrates on the processing device side. In other words, when transferring substrates between a load lock unit and a processing module unit, the substrates may be transferred using the substrate transfer method of the present disclosure.

[0075] In the above embodiment, the illumination unit 50 emits illumination light at the timing when the image capture unit 40 captures an image. However, the present disclosure is not limited to this. In the present disclosure, the illumination unit may emit illumination light continuously. Furthermore, the control unit may control the emission of illumination light based on the captured image through control processing. For example, the control unit may change the amount of light, wavelength, etc. of the illumination light based on the luminance value of the captured image.

[0076] Furthermore, in the above embodiment, an example has been shown in which two illumination units 50 including LEDs that emit yellow light as illumination light are arranged, but the present disclosure is not limited to this. In the present disclosure, the illumination unit may include a light source device other than an LED. Furthermore, the illumination light from the illumination unit may have a wavelength of a predetermined color, such as orange, in addition to yellow. The illumination unit may emit light including multiple wavelengths, such as white light. Furthermore, the illumination light may be light of a wavelength other than visible light, such as infrared light. Furthermore, the number of illumination units may be one, or may be three or more.

[0077] Furthermore, in the above embodiment, an example has been shown in which three FOUPs 101 serving as substrate storage containers are attached to the transfer chamber 10 of the substrate transfer system 100, and three imaging units 40 are arranged corresponding to the FOUPs 101, but the present disclosure is not limited to this. In the present disclosure, the number of substrate storage containers attached to the transfer chamber may be two or less, or four or more. Also, one imaging unit may be arranged for multiple substrate storage containers. In this case, images of substrates being transported to the multiple substrate storage containers are captured by a common imaging unit.

[0078] In the above embodiment, an example was described in which the trained model 62, which uses the captured image 80 captured by the imaging unit 40 as input and outputs the transport control information 81, is trained by machine learning, but the present disclosure is not limited to this. In the present disclosure, image processing such as noise removal, edge enhancement, and contrast change may be performed on the captured image to be input to the trained model before inputting it to the trained model. Furthermore, the output from the trained model may not be transport control information including command values ​​for operating the robot arm, but may be transport control information including coordinates indicating the detected positions of the substrate, substrate placement unit, and substrate transport robot.

[0079] The functions of the elements disclosed herein can be performed using circuits or processing circuits, including general-purpose processors, special-purpose processors, integrated circuits, ASICs (Application Specific Integrated Circuits), conventional circuits, and / or combinations thereof, configured or programmed to perform the disclosed functions. A processor is considered a processing circuit or circuit because it includes transistors and other circuitry. In this disclosure, a circuit, unit, or means is hardware that performs the recited functions or hardware that is programmed to perform the recited functions. The hardware may be hardware disclosed herein or other known hardware that is programmed or configured to perform the recited functions. Where the hardware is a processor, which is considered a type of circuit, the circuit, means, or unit is a combination of hardware and software, and the software is used to configure the hardware and / or processor.

[0080] Aspects It will be appreciated by those skilled in the art that the exemplary embodiments described above are examples of the following aspects.

[0081] (Aspect 1) A substrate transport system comprising: a substrate transport robot; a detection unit that detects at least one of a substrate transported by the substrate transport robot and a substrate placement part on which the substrate is placed; and a control unit that uses a detection result of at least one of the substrate and the substrate placement part by the detection unit as an input and generates a trained model through machine learning that outputs transport control information for controlling the transport operation of the substrate by the substrate transport robot.

[0082] (Aspect 2) The substrate transfer system according to aspect 1, wherein the control unit generates the trained model through machine learning using deep reinforcement learning.

[0083] (Aspect 3) The substrate transport system according to Aspect 1 or Aspect 2, wherein the control unit uses the detection result as an input and generates the trained model by machine learning, the trained model outputting the transport control information including teaching information for aligning coordinate axes between the substrate transport robot and the position of the substrate mounting unit.

[0084] (Aspect 4) A substrate transport system according to any one of Aspects 1 to 3, wherein the detection unit detects the substrate transport robot in addition to at least one of the substrate and the substrate placement unit, and the control unit receives the detection results of at least one of the substrate and the substrate placement unit and the detection results of the substrate transport robot as inputs, and generates the trained model that outputs the transport control information.

[0085] (Aspect 5) A substrate transport system according to any one of Aspects 1 to 4, wherein the control unit also serves as a robot control unit that controls the transport operation of the substrate transport robot, acquires the transport control information by using the trained model, and performs the substrate transport operation based on the acquired transport control information, thereby controlling the transport operation of the substrate transport robot while correcting deviations in the transport operation.

[0086] (Aspect 6) A substrate transport system according to any one of Aspects 1 to 5, wherein the detection unit includes an imaging unit that images at least one of the substrate, the substrate placement unit, and a target member that indicates the position of the substrate placement unit, and the control unit receives as input an image captured by the imaging unit obtained as the detection result by the detection unit, and generates the trained model that outputs the transport control information through machine learning.

[0087] (Aspect 7) A substrate transport system as described in Aspect 6, further comprising a target member indicating the position of the substrate placement portion, wherein the imaging unit images the target member, and the control unit receives the captured image including the target member obtained as the detection result as input, and generates the trained model that outputs the transport control information through machine learning.

[0088] (Aspect 8) The substrate transfer system according to aspect 7, wherein the target member has a predetermined shape in the captured image so as to be identified in machine learning that generates the trained model.

[0089] (Aspect 9) The substrate transfer system according to any one of Aspects 6 to 8, further comprising a transfer chamber in which the substrate transfer robot is disposed and in which the substrate is transferred, and the imaging unit is disposed in the transfer chamber.

[0090] (Aspect 10) The substrate transport system according to Aspect 9, wherein the imaging unit includes a transport chamber imaging unit disposed in the transport chamber and a robot imaging unit disposed in the substrate transport robot, and the control unit receives the captured image by the transport chamber imaging unit and the captured image by the robot imaging unit as inputs, and generates the trained model by machine learning to output the transport control information.

[0091] (Aspect 11) The substrate transfer system according to aspect 10, wherein the transfer chamber imaging unit is disposed above the transfer chamber.

[0092] (Aspect 12) A substrate transport system according to any one of Aspects 9 to 11, wherein the substrate transport robot, in the transport chamber, performs a transport operation of the substrate to a substrate storage container in which the substrate is stored; the imaging unit images at least one of the substrate stored in the substrate storage container, the substrate placement portion in the substrate storage container on which the substrate is placed, and the target member indicating the position of the substrate placement portion in the substrate storage container; and the control unit receives as input the captured image including at least one of the substrate stored in the substrate storage container, the substrate placement portion in the substrate storage container on which the substrate is placed, and the target member indicating the position of the substrate placement portion in the substrate storage container, and generates by machine learning the trained model that outputs the transport control information.

[0093] (Aspect 13) The substrate transport system described in Aspect 12, wherein the substrate transport robot transports the substrate between the substrate storage container and a processing standby section on which the substrate is placed to be handed over to a processing device that processes the substrate; the imaging section images at least one of the substrate placed on the substrate placement section in the processing standby section, the substrate placement section in the processing standby section, and the target member indicating the position of the substrate placement section in the processing standby section; and the control section receives as input the captured image including at least one of the substrate placed on the substrate placement section in the processing standby section, the substrate placement section in the processing standby section, and the target member indicating the position of the substrate placement section in the processing standby section, and generates the trained model by machine learning that outputs the transport control information.

[0094] (Aspect 14) A substrate transport system according to any one of Aspects 1 to 13, wherein the control unit also serves as a robot control unit that controls the transport operation of the substrate transport robot, and when generating the trained model by machine learning using the detection results as input, controls the transport operation while keeping the operating speed of the substrate transport robot relatively slow until learning of the trained model is completed.

[0095] (Aspect 15) A substrate transport method comprising: detecting at least one of a substrate transported by a substrate transport robot and a substrate placement portion on which the substrate is placed; and generating, by machine learning, a trained model that uses the detection results of at least one of the substrate and the substrate placement portion as input and outputs transport control information for controlling the transport operation of the substrate by the substrate transport robot.

Claims

1. A substrate transport system comprising: a substrate transport robot; a detection unit that detects at least one of a substrate transported by the substrate transport robot and a substrate placement part on which the substrate is placed; and a control unit that uses as input a detection result of at least one of the substrate and the substrate placement part by the detection unit and generates, through machine learning, a trained model that outputs transport control information for controlling the transport operation of the substrate by the substrate transport robot.

2. The substrate transport system of claim 1, wherein the control unit generates the trained model through machine learning using deep reinforcement learning.

3. The substrate transport system of claim 1, wherein the control unit uses machine learning to generate the trained model, which receives the detection results as input and outputs the transport control information including teaching information for aligning coordinate axes between the substrate transport robot and the position of the substrate mounting unit.

4. The substrate transport system of claim 1, wherein the detection unit detects the substrate transport robot in addition to at least one of the substrate and the substrate mounting unit, and the control unit receives as input the detection results of at least one of the substrate and the substrate mounting unit and the detection results of the substrate transport robot, and generates the trained model that outputs the transport control information.

5. The substrate transport system of claim 1, wherein the control unit also serves as a robot control unit that controls the transport operation of the substrate transport robot, acquires the transport control information by using the learned model, and performs the substrate transport operation based on the acquired transport control information, thereby controlling the transport operation of the substrate transport robot while correcting deviations in the transport operation.

6. The substrate transport system of claim 1, wherein the detection unit includes an imaging unit that images at least one of the substrate, the substrate mounting unit, and a target member indicating the position of the substrate mounting unit, and the control unit receives as input an image captured by the imaging unit and obtained as the detection result by the detection unit, and generates the trained model that outputs the transport control information through machine learning.

7. A substrate transport system as described in claim 6, further comprising a target member indicating the position of the substrate mounting portion, wherein the imaging unit images the target member, and the control unit receives as input the captured image including the target member obtained as the detection result, and generates the trained model by machine learning to output the transport control information.

8. The substrate transportation system of claim 7, wherein the target member has a predetermined shape in the captured image so as to be identified in machine learning that generates the trained model.

9. The substrate transport system according to claim 6, further comprising a transport chamber in which the substrate transport robot is disposed and in which the substrate is transported, and the imaging unit is disposed in the transport chamber.

10. The substrate transport system of claim 9, wherein the imaging unit includes a transport chamber imaging unit arranged in the transport chamber and a robot imaging unit arranged in the substrate transport robot, and the control unit receives as input the images captured by the transport chamber imaging unit and the images captured by the robot imaging unit, and generates the trained model that outputs the transport control information through machine learning.

11. The substrate transfer system according to claim 10, wherein the transfer chamber imaging unit is disposed above the transfer chamber.

12. The substrate transport system of claim 9, wherein the substrate transport robot performs a transport operation of the substrate to a substrate storage container in which the substrate is stored in the transport chamber; the imaging unit images at least one of the substrate in a state where it is stored in the substrate storage container, the substrate mounting portion in the substrate storage container on which the substrate is placed, and the target member indicating the position of the substrate mounting portion in the substrate storage container; and the control unit receives as input the captured image including at least one of the substrate in a state where it is stored in the substrate storage container, the substrate mounting portion in the substrate storage container on which the substrate is placed, and the target member indicating the position of the substrate mounting portion in the substrate storage container, and generates the trained model by machine learning that outputs the transport control information.

13. The substrate transport system of claim 12, wherein the substrate transport robot transports the substrate between the substrate storage container and a processing waiting section on which the substrate is placed to be handed over to a processing device that processes the substrate; the imaging section images at least one of the substrate placed on the substrate placement section in the processing waiting section, the substrate placement section in the processing waiting section, and the target member indicating the position of the substrate placement section in the processing waiting section; and the control section receives as input the captured image including at least one of the substrate placed on the substrate placement section in the processing waiting section, the substrate placement section in the processing waiting section, and the target member indicating the position of the substrate placement section in the processing waiting section, and generates the trained model by machine learning to output the transport control information.

14. The substrate transport system of claim 1, wherein the control unit also serves as a robot control unit that controls the transport operation of the substrate transport robot, and when generating the trained model by machine learning using the detection results as input, controls the transport operation with the operating speed of the substrate transport robot kept relatively slow until learning of the trained model is completed.

15. A substrate transport method comprising: detecting at least one of a substrate transported by a substrate transport robot and a substrate placement part on which the substrate is placed; using the detection results of at least one of the substrate and the substrate placement part as input; and generating, through machine learning, a trained model that outputs transport control information for controlling the transport operation of the substrate by the substrate transport robot.

Citation Information

Patent Citations

  • Plasma processor

    JP2007149960A

  • Substrate transfer device and position correction method of hand of the substrate transfer device

    JP2021048270A

  • Transport device and transport method

    JP2021119587A

  • Teaching method, transport system, and program

    JP2022003663A

  • Apparatus, system, and method for improving robot joint coordinate teaching accuracy

    JP2022518980A