Apparatus for processing substrate and method for determining whether substrate processing process is normal

Abstract

The present inventive concept provides a method for determining whether a substrate processing process is normal using a deep learning model. The method includes receiving input of a substrate processing process video, preprocessing the input video, analyzing the preprocessed video using a deep learning model, and determining whether the substrate processing process is normal by comparing the trained model with the real-time substrate processing process video.

Description

Technical Field

[0001] Embodiments of the inventive concept described herein relate to an apparatus for processing a substrate and a method for determining whether a substrate processing process is normal. More specifically, embodiments of the inventive concept described herein relate to a method and apparatus for determining whether a substrate processing process is normal using a deep learning model. Background Art

[0002] Conventional substrate processing equipment uses a vision sensor to determine whether the substrate processing process is normal. However, in this method of using a vision sensor, the bowl-shaped portion may block the vision sensor from sensing the target area, making it difficult to detect whether the substrate processing process is proceeding normally.

[0003] More specifically, conventional technologies use multiple nozzles depending on the process and have a high bowl, making it difficult to detect the presence of a substrate. Furthermore, the nozzles spray aerosol in various patterns depending on the N2 flow rate, and the pattern of the sprayed aerosol is transparent and irregular, making it difficult for conventional visual sensors to diagnose the spray pattern. Furthermore, the sprayed aerosol's diffusion pattern can be affected and irregular depending on the wetting of the substrate, which depends on factors such as the discharge flow rate, discharge position, RPM, and the hydrophilicity or hydrophobicity of the substrate surface, further complicating diagnosis with conventional visual sensors.

[0004] Therefore, a new method is still needed to diagnose substrate processes in situ. Summary of the Invention

[0005] Embodiments of the inventive concept provide a method for in-situ determining whether a substrate treating process is normal.

[0006] The technical objectives of the present invention are not limited to the above-mentioned objectives, and those skilled in the art will understand other unmentioned technical objectives through the following description and accompanying drawings.

[0007] In an embodiment of the present inventive concept, a method of determining whether a substrate treating process is normal using a deep learning model is provided.

[0008] The method includes receiving an input image about substrate processing, preprocessing the input image, and learning the preprocessed input image by using a deep learning model; and comparing the real-time substrate processing process image with the learned data by using the trained deep learning model to determine whether the substrate processing process is normal.

[0009] In an embodiment of the present inventive concept, pre-processing the input image includes identifying a tip region of a nozzle in the input image regarding a substrate process, and ejecting chemicals from the tip region of the nozzle.

[0010] In one embodiment, this may include selecting the identified tip region of the nozzle as the ROI.

[0011] In one embodiment, learning the pre-processed input image using a deep learning model includes learning a ROI based on the flow rate of the chemical.

[0012] In one embodiment, pre-processing the input image includes correcting perspective distortion of the input image.

[0013] In one embodiment, pre-processing the input image includes detecting a shape of the substrate based on the input image corrected for perspective distortion, and detecting a center point of the shape of the substrate.

[0014] In one embodiment, pre-processing the input image further includes converting a coordinate system into a polar coordinate system based on the detected center point.

[0015] In one embodiment, this may include selecting a constant edge range of the substrate as the ROI in a polar coordinate system.

[0016] In one embodiment, learning the pre-processed input image using a deep learning model includes learning a ROI.

[0017] In another embodiment, a substrate processing apparatus for processing a substrate by spraying chemicals is provided.

[0018] The apparatus includes an image capturing unit for capturing images of a substrate and chemicals sprayed on the substrate; and a determining unit for determining whether a substrate treating process is normal by using a deep learning model.

[0019] In one embodiment, the determination unit includes: a preprocessing unit for preprocessing data of an image captured by an image capture unit; a learning unit for learning the preprocessed data by using a deep learning model; and a comparison unit for determining whether the substrate processing process is normal by comparing the real-time substrate processing process image and the learned data in the learning unit.

[0020] In one embodiment, the determination unit determines whether the substrate treating process is normal based on a spraying state of the exhaust chemical on the substrate or a wetting state of the substrate.

[0021] In one embodiment, the pre-processing unit may select a portion of an area of ​​the substrate imaged from the image capturing unit and the chemical sprayed on the substrate as a region of interest (ROI).

[0022] In one embodiment, the pre-processing unit selects a portion of the image of the substrate and the chemical sprayed on the substrate captured by the image capturing unit as the region of interest.

[0023] In one embodiment, the pre-processing unit is configured to compensate for perspective distortion of the image data.

[0024] In one embodiment, the pre-processing unit is configured to detect a shape of the substrate based on the data corrected for perspective distance distortion, and to detect a center point based on the shape of the substrate.

[0025] In one embodiment, the pre-processing unit converts the coordinate system into a polar coordinate system based on the detected center point.

[0026] In one embodiment, the pre-processing unit selects a constant edge range of the substrate converted in a polar coordinate system as the ROI.

[0027] In one embodiment, training a deep learning model prevents process defects.

[0028] In one embodiment, a preprocessing step before training a deep learning model facilitates more efficient training. BRIEF DESCRIPTION OF THE DRAWINGS

[0029] The above and other objects and features will become apparent from the following description with reference to the following drawings, in which like reference numerals refer to like parts throughout the various figures unless otherwise specified, and in which:

[0030] Figure 1 FIG. 1 illustrates a substrate treating process apparatus according to one embodiment of the present inventive concept in an exemplary manner.

[0031] Figure 2 is a block diagram illustrating a configuration of a determining unit according to one embodiment of the present inventive concept.

[0032] Figure 3 FIGURE 1 illustrates selection of a region of interest for studying an aerosol spray according to one embodiment of the present inventive concept.

[0033] Figure 4 is a flowchart illustrating a method of determining whether an aerosol spray pattern is normal according to one embodiment of the present inventive concept.

[0034] Figure 5 FIGURE 1 illustrates region of interest selection for studying the wetting state of a substrate according to one embodiment of the inventive concept.

[0035] Figure 6 is a flowchart illustrating a method of determining whether a wetting state of a substrate is normal according to one embodiment of the present inventive concept. DETAILED DESCRIPTION

[0036] The present invention is susceptible to various modifications and forms, and specific embodiments thereof will be shown and described in detail in the accompanying drawings. However, the embodiments according to the present invention are not intended to limit the specific disclosed forms, and it should be understood that the present invention includes all permutations, equivalents, and alternatives within the spirit and technical scope of the present invention. In the description of the present invention, when a detailed description of the relevant known technology may obscure the essence of the present invention, it may be omitted.

[0037] The terms used herein are only used for the purpose of describing specific embodiments and are not intended to limit the present inventive concept. As used herein, the singular forms "a", "an" and "said" are also intended to include the plural forms, unless the context clearly indicates otherwise. It will be further understood that the terms "include" and / or "include", when used in this specification, specify the presence of the features, wholes, steps, operations, elements and / or parts, but do not exclude the presence or addition of one or more other features, wholes, steps, operations, elements, parts and / or their groups. As used herein, the term "and / or" includes any and all combinations of one or more of the associated listed items. In addition, the term "exemplary" is intended to refer to an example or illustration.

[0038] It should be understood that although the terms "first," "second," "third," etc. may be used herein to describe various elements, components, regions, layers, and / or portions, these elements, components, regions, layers, and / or portions should not be limited by these terms. These terms are only used to distinguish one element, component, region, layer, or portion from another region, layer, or portion. Thus, a first element, component, region, layer, or portion discussed below may be referred to as a second element, component, region, layer, or portion without departing from the teachings of the present inventive concept.

[0039] Figure 1 A substrate treating process apparatus 1 according to one embodiment of the present inventive concept is illustrated in an exemplary manner.

[0040] Figure 1 The substrate processing apparatus 1 can process a substrate W using a liquid, which is discharged onto the substrate W through a plurality of nozzle outlets 11, 12, 13, and 14. In an embodiment of the present inventive concept, the discharged liquid can be a cleaning liquid or a chemical. According to one embodiment of the present inventive concept, the liquid discharged from the plurality of nozzle outlets 11, 12, 13, and 14 can be sprayed in the form of an aerosol. The discharged liquid can be provided to the surface of the substrate W.

[0041] The substrate processing apparatus 1 according to one embodiment of the present inventive concept may be provided with a fixed nozzle 10 to discharge the processing liquid to the substrate. The fixed nozzle 10 may include a plurality of nozzle outlets 11, 12, 13 and 14. Figure 1 In the embodiment, as an exemplary embodiment, the fixed nozzle 10 includes four nozzle outlets 11, 12, 13 and 14.

[0042] Figure 1 The substrate processing apparatus 1 in the embodiment may include a bowl-shaped portion 40 surrounding the substrate W. The bowl-shaped portion 40 may prevent the discharged liquid ejected from the nozzle from scattering.

[0043] The substrate processing apparatus 1 may include an image capturing unit 20 and a determining unit 30 .

[0044] The image capture unit 20 can capture an image of the discharged liquid during discharge from the fixed nozzle 10. Furthermore, the image capture unit can capture the substrate W after the discharged liquid is discharged. In one embodiment, the image capture unit 20 can be a visual camera. The image capture unit 20 can be configured to capture an image of the entire substrate W. The image capture unit 20 can be positioned in a corner of the chamber so that it can capture an image of the entire substrate W. In one embodiment, the image capture unit 20 can perform real-time imaging of the substrate W and the discharged liquid discharged onto the substrate W.

[0045] The determination unit 30 can be connected to the image capture unit 20 and use the image captured by the image capture unit 20 to determine whether the substrate processing process is normal by using a deep learning model. The specific configuration of the determination unit 30 can be referred to Figure 2 Further detailed explanation.

[0046] Figure 2 FIG. 1 shows a configuration of the determination unit 30 according to one embodiment of the present inventive concept.

[0047] In one embodiment, the determining unit 30 may include a pre-processing unit 31 , a learning unit 32 , and a comparing unit 33 .

[0048] The preprocessing unit 31 may preprocess the image data captured by the image capturing unit 20. The learning unit 32 may perform deep learning on the preprocessed data from the preprocessing unit 31. The comparing unit 33 may determine whether the substrate processing is normal by comparing with the learned data from the learning unit 32.

[0049] In one embodiment, the state determined by the determination unit 30 regarding whether the substrate treatment process is normal may be a spray state of discharged chemicals on the substrate or a wetting state of the substrate. In another embodiment of the present inventive concept, the determination unit 30 may determine whether the substrate exists in the chamber.

[0050] In one embodiment, the determination unit 30 can diagnose the aerosol spray pattern ejected from the nozzle. In this way, the N2 aerosol spray pattern image can be used to learn an aerosol network model. Based on the learned aerosol network model, the aerosol spray pattern can be diagnosed.

[0051] According to one embodiment of the present inventive concept, the determination unit 30 can diagnose the wetting state of the substrate, train a wetting network model using images of the substrate in the wetting state and the substrate in the non-wetting state, and determine the wetting state of the substrate based on the trained wetting network model.

[0052] More specifically, in one embodiment, the preprocessing unit 31 may preprocess the data of the image captured by the image capture unit 20. The preprocessing unit 31 may modify the image data to improve the training efficiency before using the image captured by the image capture unit 20 to train the deep learning model. Figure 3 and Figure 5 The detailed preprocessing in the preprocessing unit 31 is explained.

[0053] The learning unit 32 in one embodiment of the present invention can use a deep learning model to learn the pre-processed data. In one embodiment, Tensor Flow developed by the Google Brain Team can be used as the deep learning model.

[0054] TensorFlow is a software library for performing mathematical computations using data flow graphs. Graph nodes represent mathematical computations, while graph edges represent multidimensional data arrays (tensors) that flow between nodes.

[0055] In addition to TensorFlow, other mathematical libraries for general machine learning applications and neural networks can be employed. In one embodiment, the deep learning model is retrained using only positive data and negative data is discarded.

[0056] The comparison unit 33 can determine whether the substrate processing process is normal by using the learned data in the learning unit 32. In one embodiment, the learned data in the learning unit 32 and the real-time imaging data imaged by the image capture unit 20 can be compared in real time to determine whether the current substrate processing process is normal.

[0057] Figure 3 Region of interest (ROI) selection for studying the spraying of aerosol according to one embodiment of the present inventive concept is shown.

[0058] according to Figure 3According to one embodiment of the present inventive concept, the pre-processing unit 31 may select an image of a portion of a substrate and a chemical sprayed on the substrate, captured by the image capturing unit 20, as a region of interest (shown as a rectangular red line). In one embodiment, the region of interest may be a tip region in the substrate processing image, which is a region of chemical sprayed via a nozzle.

[0059] In one embodiment, to determine whether the aerosol spray pattern is normal, the pre-processing unit 31 selects an area near the tip area as a region of interest (ROI) and learns the selected ROI image, thereby enhancing the learning effect. In conventional technology, the spray pattern of the discharged liquid is irregular because it is in the form of mist, the spray pattern of the discharged liquid is transparent, and it is difficult to detect the discharged liquid due to the vibration of the pump.

[0060] In one embodiment, by excluding the entire image and specifying only a certain area as a region of interest (ROI), it is possible to improve learning efficiency by selecting the nozzle as the ROI and only learning the nozzle area. In addition, by learning each N2 flow rate, it is possible to verify whether the discharge of each flow rate is normal.

[0061] Figure 4 is a flowchart of determining whether an aerosol spray pattern is normal according to one embodiment of the present inventive concept.

[0062] according to Figure 4 , the spray pattern of the substrate and the nozzle can be captured at the image capture unit 20. The captured image is input to the pre-processing unit 31. In the pre-processing unit 31, the aerosol tip area is selected as the area of ​​interest from the input image data and can be provided in a pre-processed state for training the deep learning model. Figure 4 Based on this pre-processing state, the spray pattern for each nozzle volume can be learned. In one embodiment, each of the N2 flow rates from nozzles 10, 20, 30, 40, and 50 can be learned. Whether the N2 input flow rate and the N2 diagnostic flow rate are the same can be determined by comparing the learned data for each flow rate by the deep learning model with the real-time aerosol spray for each flow rate. This can determine whether the aerosol spray pattern ejected from the nozzle is normal.

[0063] Figure 5 Illustrated is the selection of a region of interest for studying the wetting state of a substrate in one embodiment of the present inventive concept.

[0064] The wetting state of the substrate refers to whether the chemical has been evenly applied over the entire area of ​​the substrate.

[0065] according to Figure 5, preprocessing can be performed to learn the wetting state of the substrate. It may be difficult to determine the specific wetting state because the image capture unit 20 is placed at an angle, causing the shape of the substrate to be captured as an ellipse rather than a circle. Therefore, in order to correct the distortion, perspective distortion correction can be performed to form a circle. After the perspective distortion correction process, the substrate is corrected to a circle, and the center point of the substrate can be calculated after the substrate is detected from the image. In addition, after calculating the center point, the coordinate system can be converted into a polar coordinate system. The traditional coordinate system (x, y) can be converted into a Cartesian coordinate system (r, θ). The following equation can be used for conversion.

[0066]

[0067]

[0068] In one embodiment, the coordinates (x, y) representing the wetting state of the substrate can be converted into (r, θ) of a Cartesian coordinate system. Thus, when a certain point of the substrate is not wetted, it can be identified. In one embodiment of the present invention, the range of r and the range of θ of the polar coordinate system can be limited to select a certain area of ​​the substrate that may not be wetted as the region of interest to be learned. In one embodiment, with respect to the wetting state of the substrate converted to the polar coordinate system, as a preprocessing, a constant edge range of the substrate can be selected as the region of interest and the selected region of interest can be learned. Here, the constant edge range refers to a range of positions on the substrate that has a high possibility of substrate wetting defects.

[0069] Figure 6 is a flowchart illustrating a method for determining whether a wetting state of a substrate is normal according to one embodiment of the present inventive concept.

[0070] refer to Figure 6 , the image capturing unit can capture an image of the wetting state of the substrate. Before deep learning, the captured image of the wetting state of the substrate is transmitted to the preprocessing unit for preprocessing. In the preprocessing of the preprocessing unit, perspective distortion correction is first performed, and then the center of the chip and the substrate is detected based on the perspective distortion corrected data. Through perspective distortion correction, the distorted shape of the substrate caused by capturing in an oblique direction can be corrected to a circle. In addition, after detecting the center point of the substrate, the edge portion of the substrate can be detected. By detecting the edge portion and limiting the edge range and then synthesizing them, preprocessed data for learning can be formed. According to Figure 6In this embodiment, by setting r to 140 or 150 and limiting the range of θ to a certain portion, a portion of the substrate where wetting errors may occur can be selected as a region of interest and learned. Using the learned data for the region of interest, the substrate's wetting status can be checked. If substrate wetting is not complete, an alert is issued to the user, allowing maintenance to complete wetting.

[0071] The present inventive concept aims to provide a method for diagnosing substrate treatment processes (e.g., substrate presence, aerosol spray pattern according to N2 flow rate, wafer wetting) based on deep learning technology, which overcomes the problems encountered in conventional methods using vision sensors.

[0072] In an embodiment of the present invention, using deep learning to learn images that indicate the presence of substrates in complex processing environments can minimize losses by pre-diagnosing chamber damage caused by substrate damage. Furthermore, according to an embodiment of the present invention, an aerosol spray pattern can be learned based on the N2 flow rate by pre-detecting errors in the aerosol spray pattern, thereby preventing process errors. Furthermore, according to an embodiment, the wetting state of the substrate can be pre-detected by learning the wetting state of the substrate based on deep learning, thereby preventing process errors.

[0073] Although preferred embodiments of the present invention have been shown and described so far, the present invention is not limited to the specific embodiments described above, and it should be noted that a person skilled in the art to which the present invention pertains may implement the present invention in various ways without departing from the essence of the present invention as claimed in the claims, and such modifications should not be interpreted as being divorced from the technical spirit or prospects of the present invention. The effects of the present invention are not limited to those described above, and those skilled in the art to which the present invention pertains may clearly understand the effects not mentioned from the description and the accompanying drawings.

Claims

1. A method for determining whether a substrate processing process is normal using a deep learning model, the method comprising: receiving an input image relating to substrate processing; Pre-processing the input image includes: identifying a tip region of a nozzle in the input image related to substrate processing, ejecting a chemical from the tip region of the nozzle from which the chemical is to be ejected; selecting the identified tip region of the nozzle as a region of interest; The method includes learning the pre-processed input image by using a deep learning model, including: learning the region of interest according to the flow rate of the chemical; and comparing the real-time substrate processing process image with the learned data by using the trained deep learning model to determine whether the substrate processing process is normal.

2. The method according to claim 1, wherein preprocessing the input image comprises: The input image is corrected for perspective distortion.

3. The method according to claim 2, wherein preprocessing the input image further comprises: A shape of a substrate is detected based on the perspective distortion corrected input image and a center point of the shape of the substrate is detected.

4. The method according to claim 3, wherein preprocessing the input image further comprises: Convert the coordinate system to polar coordinate system based on the detected center point. The method according to claim 4 , further comprising selecting a constant edge range of the substrate as a region of interest in the polar coordinate system.

6. The method of claim 5, wherein learning the pre-processed input image by using the deep learning model comprises: The region of interest is learned.

7. A substrate processing apparatus for processing a substrate by spraying chemicals, the substrate processing apparatus comprising: an image capturing unit configured to capture an image of the substrate and the chemical sprayed on the substrate; as well as A determination unit, wherein the determination unit determines whether the substrate processing process is normal by using a deep learning model, The determining unit comprises: a preprocessing unit, configured to preprocess data of the image captured by the image capturing unit; a learning unit that performs learning on the preprocessed data using a deep learning model; and a comparing unit, which determines whether the substrate processing process is normal by comparing the data of the real-time substrate processing process image with the learned data in the learning unit, wherein the determining unit determines whether the substrate processing process is normal based on an aerosol spray pattern in a chamber where the substrate process is performed or wetting of the substrate, wherein the pre-processing unit selects a portion of the image of the substrate and the chemical sprayed on the substrate captured by the image capturing unit as a region of interest, wherein the region of interest is a tip region in the image of the substrate process, the tip region being a region where chemicals are ejected from a nozzle, and The learning unit learns the region of interest according to the flow rate of the chemical. 8 . The substrate processing apparatus according to claim 7 , wherein the pre-processing unit is further configured to compensate for perspective distortion of the image data. 9 . The substrate processing apparatus according to claim 8 , wherein the pre-processing unit is further configured to detect a shape of the substrate based on the data corrected for perspective distance distortion, and to detect a center point based on the shape of the substrate. 10 . The substrate processing apparatus according to claim 9 , wherein the pre-processing unit converts a coordinate system into a polar coordinate system based on the detected center point. 11 . The substrate processing apparatus according to claim 10 , wherein the pre-processing unit selects a constant edge range of the substrate as a region of interest in the polar coordinate system.

12. A method for determining whether a substrate is wetted by using a deep learning model, the method comprising: receiving an input image relating to substrate processing; Pre-processing the input image includes: identifying a tip region in the input image related to substrate processing, the tip region being a chemical spraying region sprayed by a nozzle; selecting the identified tip region of the nozzle as a region of interest; Learning the pre-processed input image using a deep learning model includes: learning the region of interest based on the flow rate of the chemical; and The deep learning model compares the real-time substrate processing process image with the learned data and determines whether the substrate processing process is normal.

Images