Substrate abnormality detection method and substrate abnormality detection device

Abstract

To detect abnormalities in a board when pinning up.SOLUTION: A first video is obtained by imaging from a state where a substrate is placed on a mount table until a state where the substrate has been completely separated from the mount table by a lift pin. An input image based on an image of a test frame extracted from the first video is input to a pre-trained learning model to output a restored image, and it is determined that the substrate has an abnormality when the difference between the input image and the restored image exceeds a predetermined standard. The above learning model is a learning model that has been trained using a plurality of pieces of learning data based on images of a plurality of learning frames extracted from second video obtained by imaging from a state where a normal substrate is placed on the mount table until a state where the normal substrate has been completely separated from the mount table by the lift pin. The images of the plurality of learning frames are images of a plurality of frames in a time series in which the normal substrate is separated by the lift pin.SELECTED DRAWING: Figure 5

Description

[Technical field]

[0001] The present disclosure relates to a substrate abnormality detection method and a substrate abnormality detection device. [Background technology]

[0002] There is known a plasma processing apparatus that performs plasma processing on a substrate placed on a mounting table. In order to suppress a temperature rise of the substrate due to heat generation during plasma processing, the plasma processing apparatus circulates and supplies a coolant whose temperature is adjusted by a chiller device to a coolant passage in the mounting table, and simultaneously supplies a heat transfer gas such as He gas to the rear surface of the substrate through the mounting table. In this way, the plasma processing apparatus controls the temperature of the substrate to be constant. In addition, the plasma processing apparatus fixes the substrate to the mounting table by electrostatically adsorbing the substrate to prevent unnecessary leakage of the heat transfer gas.

[0003] In such a plasma processing apparatus, a technique has been proposed for detecting the presence or absence of an abnormality in the substrate placed on the mounting table. For example, in Patent Document 1, the flow rate of the heat transfer gas is monitored to detect leakage of the heat transfer gas that occurs when an abnormality occurs in the substrate, such as a misalignment of the substrate or a crack in the substrate. When an abnormality in the substrate is detected in this manner, the plasma processing apparatus is immediately stopped to prevent the spread of damage caused by the abnormality in the substrate. [Prior art documents] [Patent documents]

[0004] [Patent Document 1] JP 2012-99634 A Summary of the Invention [Problem to be solved by the invention]

[0005] The technology disclosed herein detects abnormalities in a board during pin-up. [Means for solving the problem]

[0006] One aspect of the technology disclosed herein is a substrate abnormality detection method for detecting the presence or absence of an abnormality in a substrate being separated by a lift pin from a mounting table arranged in a processing vessel of a substrate processing apparatus, the method comprising the steps of: acquiring a first video capturing a state in which the substrate is placed on the mounting table to a state in which separation by the lift pin is completed; inputting an input image based on an image of a test frame extracted from the first video into a pre-trained learning model to output a test frame restored image; and determining that the abnormality exists if the difference between the input image and the test frame restored image exceeds a predetermined standard, wherein the learning model is a learning model trained using a plurality of learning data based on images of a plurality of learning frames extracted from a second video capturing a state in which a normal substrate is placed on the mounting table to a state in which separation by the lift pin is completed, and the images of the plurality of learning frames are images of a plurality of frames along a time series in which the normal substrate is separated by the lift pin. Effect of the Invention

[0007] According to the technology disclosed herein, it is possible to detect abnormalities in a board when the pins are up. [Brief description of the drawings]

[0008] [Figure 1] 1 is a block diagram illustrating a schematic configuration of a substrate abnormality detection system including an information processing device that is a substrate abnormality detection device according to an embodiment of the technology disclosed herein. [Diagram 2] FIG. 2 is a cross-sectional view illustrating a schematic configuration of the film forming apparatus of FIG. [Diagram 3] 3 is a diagram for explaining an image of the inside of the processing vessel captured by the camera in FIG. 2. FIG. [Figure 4] 2 is a block diagram illustrating a schematic hardware configuration of the information processing device in FIG. 1. [Diagram 5] 2 is a block diagram illustrating an outline of a functional configuration of the information processing device in FIG. 1. [Figure 6] 4 is a flowchart showing the procedure of a substrate abnormality determination process executed by the information processing device of FIG. [Figure 7] FIG. 7 is a diagram for explaining the contents of the determination performed in S65 of FIG. [Figure 8] FIG. 13 is a diagram for explaining a configuration in which multiple cameras capture videos at different angles of view to be used for generating learning data. DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS

[0009] Incidentally, one example of a process in which cracks in a substrate occur is pin-up, which is used to remove static electricity from a substrate that is electrostatically attracted to a mounting table. For example, in pin-up, when a substrate is lifted by lift pins stored in the mounting table, excessive residual charges may remain on the mounting table and attract the substrate, causing the substrate to crack due to the upward force of the lift pins. Also, part of the substrate may get caught while being lifted, causing the substrate to crack.

[0010] However, the technique of Patent Document 1 is premised on the substrate being electrostatically attracted to the mounting table, and therefore is unable to detect cracks in the substrate during pin-up as described above.

[0011] In contrast, the technology disclosed herein inputs an input image based on a test frame image extracted from a video captured from the state in which the substrate is placed on the mounting table (start of pin-up) to the state in which the lift pins have completed separating the substrate (completion of pin-up) into a trained learning model, outputs a restored image, and determines whether or not there is an abnormality in the substrate based on the difference between the input image and the restored image.

[0012] Hereinafter, an embodiment of the technology according to the present disclosure will be described with reference to the drawings.

[0013] Fig. 1 is a block diagram showing a schematic configuration of a substrate abnormality detection system 1 including an information processing device 40 which is a substrate abnormality detection device according to an embodiment of the technology disclosed herein. In Fig. 1, the substrate abnormality detection system 1 is composed of a film forming apparatus 10 and an information processing device 40. In the substrate abnormality detection system 1, the information processing device 40 performs a substrate abnormality determination process shown in Fig. 6 described later to determine the presence or absence of an abnormality in a substrate separated by lift pins from a mounting table arranged inside a processing chamber of the film forming apparatus 10, and notifies the film forming apparatus 10 of the determination result. In this embodiment, the abnormality in the substrate is a crack in the substrate, but may also be an extreme warp or bend in the substrate.

[0014] Fig. 2 is a cross-sectional view showing a schematic configuration of the film forming apparatus 10 shown in Fig. 1. The film forming apparatus 10 shown in Fig. 2 is an inductively coupled plasma processing apparatus (substrate processing apparatus). The film forming apparatus 10 performs a film forming process to form a protective film or the like on a rectangular substrate, for example, a glass substrate G (hereinafter referred to as "substrate G") for an FPD (Flat Panel Display), by using plasma generated from a processing gas.

[0015] The film forming apparatus 10 includes a processing vessel 11 made of a conductive material and having a rectangular cylindrical shape, and the processing vessel 11 accommodates a substrate G. The upper part of the processing vessel 11 is airtightly sealed by a window member 12. Inside the processing vessel 11, a mounting table 13 for mounting the substrate G thereon is disposed at the lower part, and the mounting table 13 faces the window member 12. In the processing vessel 11, a processing space U is formed between the mounting table 13 and the window member 12. In the processing space U, plasma is generated from a processing gas, as described below.

[0016] An electrostatic chuck (not shown) is provided on the upper surface of the mounting table 13. The substrate G placed on the mounting table 13 is attracted and held on the mounting table 13 by the electrostatic chuck. The mounting table 13 also contains lift pins 38 as shown in FIG. 3, which will be described later. The lift pins 38 are mainly used when the substrate G is transferred in and out of the processing vessel 11 through the transfer port 19, which will be described later. More specifically, the lift pins 38 lift the substrate G and transfer the substrate G between the transfer device (not shown), which is installed adjacent to the processing vessel 11 and advances into the processing vessel. When the substrate G electrostatically attracted to the mounting table 13 is to be de-electrified, the lift pins 38 lift the substrate G (pin-up). Furthermore, a temperature control mechanism such as a chiller and a heat transfer gas supply mechanism (neither of which are shown) are provided inside the mounting table 13 in order to control the temperature of the substrate G placed thereon. The mounting table 13 is installed on the bottom surface of the processing vessel 11 via an insulating frame 14.

[0017] A metal frame 15 is provided at the upper end of the sidewall of the processing vessel 11, and a sidewall portion 16 is provided on the upper surface of the metal frame 15. The sidewall portion 16 supports a top plate 17, and the top plate 17 covers the window member 12 from above. A seal member 18 such as an O-ring is provided between the sidewall of the processing vessel 11 and the metal frame 15 to keep the processing space U airtight.

[0018] A loading / unloading port 19 for loading / unloading the substrate G into / from the processing space U and a gate valve 20 for opening / closing the loading / unloading port 19 are provided on one side wall of the processing vessel 11. A monitoring window 21 (observation window) in which an optically transparent member is airtightly fitted is provided on the other side wall of the processing vessel 11, and a camera 22 (imaging device) for monitoring the inside of the processing vessel 11 through the monitoring window 21 is provided on the outside of the processing vessel 11. The camera 22 transmits the captured image of the inside of the processing vessel 11 to a control unit 37 described later. In this embodiment, the image captured by the camera 22 is not a still image but a video. For example, the video captured by the camera 22 includes information on the state of the substrate G from when it is placed on the mounting table 13 until the lift pins 38 described later complete the separation.

[0019] The window member 12 is made of a conductor, for example, a metal, has a rectangular shape, and is divided into a plurality of segments 23, with a partition member 24 made of an insulator disposed between each of the adjacent segments 23. The partition members 24 separate the adjacent segments 23 and electrically insulate them from each other.

[0020] In the film forming apparatus 10, the space surrounded by the window member 12, the side wall portion 16, and the top plate 17 constitutes an antenna chamber 25. In the antenna chamber 25, an inductive coupling antenna 26 is disposed so as to face the substrate G on the mounting table 13 via the window member 12. Moreover, the inductive coupling antenna 26 is disposed across the regions facing each of the divided pieces 23 as a whole, and is formed in a spiral shape so as to go around in the circumferential direction of the window member 12, but the shape of the inductive coupling antenna 26 is not limited to a spiral shape.

[0021] Each divided piece 23 has a number of gas holes 27 formed therein, which open toward the processing space U. A gas supply pipe 28 is connected to each divided piece 23, and the gas supply pipe 28 is connected to a gas supply device 29. A gas diffusion chamber 30 is formed inside each divided piece 23, and the gas supply device 29 introduces processing gas into the gas diffusion chamber 30 via the gas supply pipe 28. The processing gas introduced into the gas diffusion chamber 30 is supplied to the processing space U from each gas hole 27.

[0022] In the film forming apparatus 10, the high frequency power supply 32 is connected to the inductive coupling antenna 26 via a matching box 31. The high frequency power supply 32 supplies high frequency power for generating plasma, for example, 13.56 MHz, to the inductive coupling antenna 26. As a result, in each of the divided pieces 23 constituting the window member 12, an eddy current is induced that circulates from the upper surface (the inductive coupling antenna 26 side) to the lower surface (the processing space U side), and this eddy current forms an induced electric field in the processing space U. Then, this induced electric field excites the processing gas supplied to the processing space U to generate plasma. In the above embodiment, the case where the window member 12 is made of a conductor such as a metal has been described, but the window member 12 may be made of a dielectric material. In this case, the electric field formed by the inductive coupling antenna 26 passes through the window member 12 and acts directly on the processing gas to generate plasma.

[0023] Furthermore, a high frequency power supply 34 is connected to the mounting table 13 via a matching box 33. The high frequency power supply 34 supplies, for example, a high frequency bias power of 3.2 MHz to the mounting table 13. This allows various ions in the plasma in the processing space U to be attracted to the substrate G, thereby performing a film formation process for forming various films on the substrate G. Note that the film formation apparatus 10 may also perform an etching process on the substrate G by attracting various ions to the substrate G. In this case, the film formation apparatus 10 functions as an etching apparatus.

[0024] Furthermore, in the film forming apparatus 10, an exhaust port 35 is formed on the bottom surface of the processing vessel 11. An exhaust device 36 such as a turbo molecular pump or a dry pump is connected to this exhaust port 35. When performing the film forming process, the exhaust device 36 maintains the processing space U at a predetermined pressure lower than atmospheric pressure. The film forming apparatus 10 is also provided with a control unit 37. The control unit 37 is composed of a computer having at least a CPU and a memory, and a recipe (program) for performing the predetermined film forming process is recorded in the memory.

[0025] Fig. 3 is a diagram for explaining an image of the inside of the processing vessel 11 captured by the camera 22 in Fig. 2. In Fig. 3, Fig. 3(A) and Fig. 3(B) each show a captured image of one frame among a plurality of frames constituting a moving image captured by the camera 22.

[0026] As described above, the camera 22 captures the state inside the processing vessel 11 through the monitoring window 21, specifically, the state from when the substrate G is placed on the mounting table 13 to when the lift pins 38 have completed the separation, and obtains a video including the captured image of FIG. 3(A) and the captured image of FIG. 3(B). The captured image of FIG. 3(A), which captures the state in which the substrate G is placed on the mounting table 13, includes structural members inside the processing vessel 11, such as the mounting table 13 and the substrate G placed on the mounting table 13. The captured image of FIG. 3(B), which captures the state in which the lift pins 38 have completed the separation, includes the mounting table 13, the substrate G, and the lift pins 38 lifting the substrate G. The camera 22 then transmits the video to the control unit 37. The control unit 37 transmits the received video to the information processing device 40. In this embodiment, a configuration is described in which the camera 22 transmits video to the information processing device 40 via the film forming apparatus 10, but the camera 22 may transmit video directly to the information processing device 40 without passing through the film forming apparatus 10.

[0027] Fig. 4 is a block diagram showing a schematic hardware configuration of the information processing device 40 shown in Fig. 1. The information processing device 40 is, for example, a personal computer. The information processing device 40 includes a CPU 41, a ROM 42, a RAM 43, a HDD 44, a communication I / F 45, and an input / output I / F 46, which are connected to each other via a system bus 47 so as to be able to communicate with each other.

[0028] The CPU 41 is a microprocessor that performs overall control of the information processing device 40. Specifically, the CPU 41 reads out programs stored in the ROM 42 or HDD 44, and executes the programs deployed on the RAM 43. The ROM 42 stores the programs and the like to be executed by the CPU 41. The RAM 43 is used as the main memory of the CPU 41, and is also used as an area for temporarily storing programs and data.

[0029] The HDD 44 stores various data and programs executed by the CPU 41. In the present embodiment, the information processing device 40 is described as having an HDD as an example of a storage device, but the information processing device 40 may have a storage device other than an HDD, such as an SSD.

[0030] The communication I / F 45 is an interface for the information processing device 40 to communicate with an external device. For example, the communication I / F 45 receives a video captured by the camera 22 from the film forming apparatus 10 or the camera 22. Note that communication by the communication I / F 45 is realized by various communication technologies, whether wireless or wired. The input / output I / F 46 is connected to an output device such as a display that displays various information, and an input device such as a mouse or a keyboard that is used by a user.

[0031] Fig. 5 is a block diagram showing a schematic functional configuration of the information processing device 40 in Fig. 1. As shown in Fig. 5, the information processing device 40 includes, as its functional configuration, an acquisition unit 48, a restored image generation unit 49, and a determination unit 50. These units function when the CPU 41 executes a program stored in the ROM 42 or the HDD 44. Note that some or all of these functions may be realized by a hardware circuit.

[0032] The acquisition unit 48 acquires an image from an external device. For example, when determining whether or not there is an abnormality in the substrate G, the acquisition unit 48 acquires a video (first video) taken by the camera 22 from the film forming apparatus 10 or the camera 22. This video is a video taken from a state in which the substrate G is placed on the mounting table 13 (start of pin-up) to a state in which the lift pins 38 complete the separation (completion of pin-up). The acquisition unit 48 extracts one frame of image from the acquired video as an image of a test frame. The extracted image of the test frame is, for example, an image at the completion of pin-up. The acquisition unit 48 also performs a reduction process on the extracted image of the test frame to reduce the image size of the image, and generates and outputs an input image. The input image is output to the restored image generation unit 49 and the determination unit 50, respectively.

[0033] The restored image generating unit 49 inputs the input image output from the acquiring unit 48 into a trained learning model that has trained using multiple images including a normal substrate G as training data, to generate a test frame restored image. The test frame restored image is an image that has the same angle of view as the input image and is a restored image including a normal substrate G that has completed pin-up. The training method of this learning model will be described later. The test frame restored image is output to the determining unit 50. The restored image generating unit 49 may be realized by the CPU 41 alone, or by the CPU 41 and a GPU (Graphics Processing Unit) (not shown) operating in cooperation with each other, or may be realized by the GPU alone.

[0034] The determination unit 50 determines whether or not there is an abnormality in the substrate G by using the difference between the input image output from the acquisition unit 48 and the test frame restored image output from the restored image generation unit 49. For example, the determination unit 50 generates a difference image by subtracting pixel values ​​of the test frame restored image from pixel values ​​of the input image, and determines whether or not the total value of pixel values ​​included in the generated difference image is greater than a predetermined threshold value, and determines that there is an abnormality if it is greater. Note that the method of determining whether or not there is an abnormality in the substrate G is not limited to this method. For example, the presence or absence of an abnormality in the substrate G may be determined by using the difference in luminance values ​​between the input image and the test frame restored image, or the presence or absence of an abnormality in the substrate G may be determined by using the difference in edges (contours) extracted from the input image and the test frame restored image. The result determined by the determination unit 50 is notified to the film forming apparatus 10. The threshold value may be determined by performing a comparison experiment using a normal substrate G and a substrate G with an abnormality, or may be determined based on errors or noise caused by the performance of the system, or may be determined based on other methods. As long as the threshold value can be used to determine whether or not there is an abnormality in the substrate G, the method for setting the threshold value is not limited.

[0035] Next, a learning method of the learning model according to the present embodiment will be described. In the present embodiment, as a preliminary step to executing a process for determining the presence or absence of an abnormality in the substrate G, learning of a learning model used by the restored image generating unit 49 is performed.

[0036] When learning the learning model, the acquisition unit 48 acquires a video (second video) that captures a state in which a normal substrate G is placed on the mounting table 13 until the lift pins 38 complete the separation. The acquisition unit 48 also extracts from the video a plurality of learning frame images along a time series in which the normal substrate G is separated by the lift pins 38. For example, the acquisition unit 48 extracts a total of three images from the acquired video as learning frame images, including two intermediate frame images during the separation of the normal substrate G from the mounting table 13 and a final frame image in which the separation is completed. Next, the acquisition unit 48 generates cut images by cutting out an area including the substrate G from each of the extracted images of the learning frames. Next, the acquisition unit 48 performs a reduction process on the generated cut images to reduce the image size of each of the cut images, thereby generating a plurality of reduced images. The reduction process for reducing the image size is performed to reduce the amount of data to be processed and improve the processing speed. At the very least, the generated reduced image needs to maintain a resolution sufficient to enable determination of the presence or absence of an abnormality.

[0037] In this embodiment, a variational auto-encoder (VAE), which is a learning model used by the restored image generating unit 49, is trained using these multiple reduced images as training data. Specifically, the model is trained to capture the characteristics of the training data, convert (encode) them into latent variables, and generate (decode) data similar to the training data from the latent variables based on the restoration error. When an input image is input to the training model trained in this way to determine whether or not there is an abnormality in the substrate G, a test frame restored image is generated that has the same angle of view as the input image and includes a normal substrate G that has completed pin-up, even if the input image is an image when an abnormality occurs in the substrate G.

[0038] In this embodiment, a process for determining whether or not there is an abnormality in the substrate G is performed using such a trained learning model.

[0039] Fig. 6 is a flowchart showing the procedure of the substrate abnormality determination process executed by the information processing device 40 of Fig. 1. The substrate abnormality determination process of Fig. 6 is realized by the CPU 41 executing a program stored in the ROM 42 or the HDD 44. The substrate abnormality determination process of Fig. 6 is executed, for example, when the information processing device 40 receives an instruction from a user to start the substrate abnormality determination process.

[0040] 6, first, the CPU 41 acquires a video captured by the camera 22 from the camera 22 or the film forming apparatus 10 (S61). As described above, this video is a video captured from a state in which the substrate G is placed on the mounting table 13 to a state in which the lift pins 38 complete the separation.

[0041] Next, the CPU 41 extracts one frame image from the acquired moving image as a test frame image (S62). The extracted test frame image is the image at the time of completion of pin-up, as described above.

[0042] Next, the CPU 41 performs a reduction process on the image of the test frame to generate an input image (S63). The processes from S61 to S63 realize the function of the acquisition unit 48 in Fig. 5. After that, the CPU 41 inputs this input image into the trained learning model, and generates a test frame restored image (S64). The process of S64 realizes the function of the restored image generation unit 49 in Fig. 5.

[0043] Next, the CPU 41 judges whether or not the difference between the input image and the test frame restored image exceeds a reference value (S65). The process of S65 realizes the function of the judgment unit 50 in Fig. 5. In this embodiment, regardless of whether the input image 71 in Fig. 7(A) including a normal substrate G or the input image 72 in Fig. 7(B) including an abnormal substrate G is input to the trained learning model, a test frame restored image 73 in Fig. 7(C) is generated, which has the same angle of view as the input image and includes a normal substrate G that has completed pin-up.

[0044] 7, the difference between the input image 71 and the test frame restored image 73 is equal to or less than the threshold value. In such a case, in S65, it is determined that the difference between the input image and the test frame restored image does not exceed the reference value, and the CPU 41 notifies the film forming apparatus 10 that there is no abnormality in the substrate G (S66), and ends this process. Upon receiving this notification, the film forming apparatus 10 continues the process that it was executing.

[0045] 7, the difference between the input image 72 and the test frame restored image 73 exceeds the threshold. In such a case, in S65, it is determined that the difference between the input image and the test frame restored image exceeds the reference, and the CPU 41 notifies the film forming apparatus 10 that there is an abnormality in the substrate G (S67), and ends this process. Upon receiving this notification, the film forming apparatus 10 stops the process that was being executed in order to prevent the damage caused by the abnormality in the substrate from spreading.

[0046] According to this embodiment, an input image based on a test frame image extracted from a video captured from a state in which the substrate G is placed on the mounting table 13 to a state in which the lift pins 38 have completed separating the substrate G from the mounting table 13 is input to a trained learning model. Then, a test frame restored image is output, and an abnormality in the substrate, for example, the presence or absence of a crack in the substrate G, is determined based on the difference between the input image and the test frame restored image. This makes it possible to detect an abnormality in the substrate G when the pins are up.

[0047] In this embodiment, the multiple learning frames include at least two intermediate frames during the process of the normal substrate G being separated from the mounting table 13, and a final frame in which the separation is completed. This allows the learning model to learn multiple different states during the process of the normal substrate G being separated from the mounting table 13, thereby improving the accuracy of the test frame restored image output from this learning model.

[0048] On the other hand, in this embodiment, when learning the learning model, learning data based on images of learning frames extracted from a video (third video) that captures the state from when the abnormal substrate G is placed on the mounting table 13 to when the lift pins 38 complete the separation is not used. That is, when learning the learning model, there is no need to prepare learning data for learning the abnormal substrate G. This can reduce the effort required when learning the learning model.

[0049] Furthermore, in this embodiment, since the learning model is a VAE, overlearning can be prevented.

[0050] Moreover, in this embodiment, the learning data is an image generated based on an image obtained by cutting out an area including the substrate G from an image of a learning frame. This allows the learning model to learn while reducing information other than the substrate G, thereby improving the accuracy of the restoration of the substrate G in the test frame restored image output from this learning model. Furthermore, the learning data is a reduced image obtained by performing a reduction process on an image obtained by cutting out an area including the substrate G from an image of a learning frame. This reduces the amount of data processed during learning, and reduces the processing load on the learning model during learning.

[0051] In addition, in this embodiment, the input image is a reduced image obtained by performing a reduction process on the image of the test frame. This reduces the processing load for analyzing the input image input to the learning model, and allows the test frame restored image to be output quickly, thereby enabling the presence or absence of an abnormality in the substrate G to be quickly determined.

[0052] In addition, in this embodiment, the information processing device 40, which is a substrate abnormality detection device, is a device separate from the film forming device 10, and therefore can also be used for detecting abnormalities in the substrate G in other substrate processing devices having the same configuration as the film forming device 10. This can save the effort of performing learning in the other substrate processing devices.

[0053] Although the preferred embodiments of the present disclosure have been described above, the present disclosure is not limited to the above-described embodiments, and various modifications and changes are possible within the scope of the gist of the present disclosure.

[0054] For example, in the above-described embodiment, a total of three images are extracted as learning frame images, including two intermediate frame images during the normal substrate G being separated from the mounting table 13 and a final frame image after the separation is completed. However, three or more images, for example, five intermediate frame images during the normal substrate G being separated from the mounting table 13 and a final frame image after the separation may be extracted as learning frame images, for example, six images in total, including five intermediate frame images during the normal substrate G being separated from the mounting table 13. This allows the learning model to learn more diverse states during the process of the normal substrate G being separated from the mounting table 13, and thus the accuracy of the test frame restored image output from this learning model can be further improved.

[0055] In the above-described embodiment, the substrate abnormality detection device is an information processing device 40 separate from the film forming apparatus 10, but the present invention is not limited to this configuration. For example, the film forming apparatus 10 may have the functions of the information processing device 40. In this case, the control unit 37 of the film forming apparatus 10 executes a process equivalent to the above-described substrate abnormality determination process. Even with such a configuration, it is possible to achieve the same effects as the above-described embodiment.

[0056] Furthermore, in the embodiment described above, the trained learning model generates a test frame restored image including the substrate G that has completed pin-up, but the trained learning model may be configured to generate a test frame restored image including the substrate G in the middle of pin-up. For example, the trained learning model may be configured to generate a test frame restored image when the substrate G is pinned up (raised) to a predetermined height. In this case, an image of the substrate G pinned up to a predetermined height is used as the image of the test frame for generating the input image.

[0057] In the above-described embodiment, the configuration is described in which only one camera captures the video used to generate the learning data, but the present invention is not limited to this configuration. For example, as shown in FIG. 8, each of the multiple cameras 22a and 22b that capture images at different angles of view may capture the video used to generate the learning data. In this case, a learning model may be prepared corresponding to each of the videos captured by the multiple cameras 22a and 22b, and each learning model may learn using the corresponding video. This allows the use of learned learning models with different angles of view in detecting an abnormality in the substrate G, thereby preventing the overlooking of an abnormality in the substrate G that cannot be confirmed from a certain angle of view, and improving the accuracy of detecting an abnormality in the substrate G.

[0058] Furthermore, in the above-described embodiment, the VAE has been described as an example of a learning model, but various autoencoders such as AE and CVAE may be used in addition to the VAE. In addition, various generative adversarial networks such as GAN, CGAN, DCGAN, SRGAN, CycleGAN, and VAEGAN, and YOLO (You Only Look Once) for object detection may be used.

[0059] In the present embodiment, the case where the present invention is applied to a film forming apparatus is described as an example of a substrate processing apparatus, but the present invention is not limited thereto. As long as the apparatus performs a process of separating a substrate placed on a mounting table from the mounting table by a lift pin, the present invention may be applied to a substrate processing apparatus that performs other processes, such as an etching apparatus or an ashing apparatus. In addition, in the present embodiment, the case where an inductively coupled plasma apparatus is used as the substrate processing apparatus is described, but the present invention is not limited thereto, and the present invention may be applied to other types of plasma apparatus, such as a capacitively coupled plasma apparatus or a microwave plasma apparatus. [Explanation of symbols]

[0060] 10 Film deposition equipment 11 Processing vessel 21 Observation Window 22,22a,22b Camera 37 Control Unit 38 Lift pin 40 Information processing device 41 CPU 48 Acquisition Department 49 Restored image generation unit 50 Judgment section 71,72 Input image 73 Test frame restored images

Claims

1. 1. A substrate abnormality detection method for detecting the presence or absence of an abnormality in a substrate that is separated by lift pins from a mounting table disposed in a processing container of a substrate processing apparatus, comprising: acquiring a first moving image capturing a state from when the substrate is placed on the mounting table to when the lift pins complete the separation of the substrate from the mounting table; inputting an input image based on a test frame image extracted from the first moving image into a pre-trained learning model, and outputting a test frame restored image; determining that the abnormality exists when a difference between the input image and the test frame restored image exceeds a predetermined criterion; the learning model is a learning model that has been learned using a plurality of learning data based on images of a plurality of learning frames extracted from a second moving image that captures a state from when a normal substrate is placed on the mounting table to when the separation by the lift pins is completed, A substrate abnormality detection method, wherein the images of the plurality of learning frames are images of a plurality of frames in a time series in which the normal substrate is lifted away by the lift pins.

2. The method for detecting an abnormality in a substrate according to claim 1 , wherein the abnormality is a crack in the substrate.

3. 2. The substrate abnormality detection method according to claim 1, wherein the plurality of learning frames include at least two intermediate frames during which the normal substrate is being separated from the mounting table, and a final frame in which the separation is completed.

4. The substrate abnormality detection method according to claim 3 , wherein the plurality of learning frames includes at least six frames.

5. A substrate abnormality detection method as described in claim 1, wherein the learning model is trained without using learning data based on images of learning frames extracted from a third video captured from the state in which an abnormal substrate is placed on the mounting table to the state in which separation by the lift pins is completed.

6. The substrate anomaly detection method according to claim 1 , wherein the learning model is a VAE.

7. The substrate abnormality detection method according to claim 1 , wherein the learning data is an image generated based on an image obtained by cutting out an area including the substrate from an image of the learning frame.

8. The substrate abnormality detection method according to claim 7, wherein the learning data is a reduced image obtained by performing a reduction process on the cut-out image.

9. The substrate abnormality detection method according to claim 1 , wherein the input image is a reduced image obtained by performing a reduction process on the image of the test frame.

10. 2. The substrate abnormality detection method according to claim 1, wherein the first video and the second video are video images taken through an observation window provided in the processing vessel.

11. A substrate anomaly detection method as described in claim 1, wherein each of a plurality of imaging devices capturing images at different angles of view captures the second video, a learning model is prepared corresponding to each of the second videos captured by the plurality of imaging devices, and each learning model learns using a corresponding second video.

12. 1. A substrate abnormality detection device that detects the presence or absence of an abnormality in a substrate that is separated by lift pins from a mounting table disposed in a processing container of a substrate processing apparatus, a means for acquiring a first moving image capturing an image of a state from when the substrate is placed on the mounting table to when the lift pins complete the separation of the substrate from the mounting table; a means for inputting an input image based on a test frame image extracted from the first moving image into a pre-trained learning model, and outputting a test frame restored image; means for determining that the abnormality exists when a difference between the input image and the test frame restored image exceeds a predetermined criterion; the learning model is a learning model that has been learned using a plurality of learning data based on images of a plurality of learning frames extracted from a second moving image that captures a state from when a normal substrate is placed on the mounting table to when the separation by the lift pins is completed, A substrate abnormality detection device, wherein the images of the plurality of learning frames are images of a plurality of frames in a time series in which the normal substrate is lifted away by the lift pins.

13. The substrate abnormality detection device according to claim 12, wherein the substrate abnormality detection device is an information processing device separate from the substrate processing device.

14. The substrate abnormality detection device according to claim 12 , wherein the substrate processing apparatus includes the substrate abnormality detection device.

Images