IP Library › Granted Patent US 12,731,243
Granted Patent B2
US 12,731,243 · App. 18/075,412 · Granted Sep 8, 2026

Nozzle inspection unit and substrate treatment apparatus including the same

Inventors: Myeong Jun Lim (Gyeonggi-do, KR); Kwang Sup Kim (Chungcheongnam-do, KR); Yeon Chul Song (Seoul, KR); Ji Hoon Yoo (Gyeonggi-do, KR); Jong Min Lee (Gyeonggi-do, KR); Jun Ho Oh (Gyeonggi-do, KR); Young Ho Park (Incheon, KR)
Assignee: SEMES CO., LTD.
G06T7/001B41J2/0451B41J2/04586G06T7/11G06T7/187G06T2207/20021G06T2207/20081G06T2207/20221G06T2207/30144G06T2207/30168
View Patent ↗
Loading inventors, assignments & file history…
Monitor This Case
Get email alerts when status or documents change.
Order Certified Copies
Most orders are placed with the USPTO same day — all within 24 business hours.
Order via The Patent Place →
Pre-filled with this patent's details
Quick Facts
Patent No.
US 12,731,243
App. No.
18/075,412
Granted
Sep 8, 2026
Kind
B2
Abstract

Provided are a nozzle inspection unit configured to generate a large amount of defect data to improve detection accuracy of a defective nozzle, and a substrate treatment apparatus including the same. The nozzle inspection unit includes: a data collection module configured to collect a plurality of image data related to nozzles; a data classification module configured to classify the plurality of image data according to predefined classes; a data merging module configured to merge good image data related to a normal nozzle and defect image data related to a defective nozzle from among the plurality of image data; a data training module configured to train a plurality of merged image data obtained through the data merging module; and a defect data generation module configured to generate a plurality of final image data from the plurality of merged image data based on the training result.

Claims (58)

1 . A nozzle inspection unit comprising:

a controller configured to:

collect a plurality of image data related to nozzles;

classify the plurality of image data according to predefined classes;

generating, after classifying the plurality of image data, a plurality of merged image data by:

first merging a first number of good image data related to a normal nozzle from among the plurality of image data with a second number of defect image data related to a defective nozzle from among the plurality of image data to obtain a first merged image data of the plurality of merged image data, the second number being smaller than the first number; and

generating remaining ones of the plurality of merged image data besides the first merged image data by duplicating and rotating the first merged image data at a third number of random rotation angles such that each of the remaining ones of the plurality of merged image data besides the first merged image data is the first merged image data rotated at one of the third number of random rotation angles;

train the plurality of merged image data after the plurality of merged image data is generated; and

generate a plurality of final image data from the plurality of merged image data based on a training result of the training of the plurality of merged image data.

2 . The nozzle inspection unit of claim 1 , wherein the final image data is image data related to a defect of the nozzle.

3 . The nozzle inspection unit of claim 2 , wherein the controller is configured to generate a number of the final image data greater than a reference number from a number of the defect image data smaller than the reference number.

4 . The nozzle inspection unit of claim 1 , wherein the controller is configured to obtain fake image data similar to the merged image data by training the merged image data.

5 . The nozzle inspection unit of 4 , wherein the controller is configured to obtain the fake image data using a generative adversarial network (GAN).

6 . The nozzle inspection unit of claim 1 , wherein the controller is configured to remove fake image data similar to the merged image data from the plurality of merged image data and generate the final image data based on the remaining merged image data.

7 . The nozzle inspection unit of claim 1 , wherein the controller is configured to segment the good image data into regions and then merge the defect image data into the good image data based on the segmented region.

8 . The nozzle inspection unit of claim 1 , wherein the plurality of image data include the good image data and the defect image data or include the good image data only.

9 . The nozzle inspection unit of claim 1 , wherein the controller is configured to classify the plurality of image data into the good image data and the defect image data.

10 . The nozzle inspection unit of claim 9 , wherein the controller is configured to classify the plurality of image data according to the classes and then classify the image data included in each of the classes into the good image data and the defect image data, or to classify the plurality of image data into the normal image data and the defect image data and then classify each of the normal image data and the defect image data according to the classes.

11 . The nozzle inspection unit of claim 1 , wherein the controller is further configured to, when the plurality of image data include only the normal image data, provide the defect image data.

12 . The nozzle inspection unit of claim 1 , wherein the controller is configured to utilize the final image data in determining a defect of the nozzle.

13 . The nozzle inspection unit of claim 1 , wherein the controller is further configured to:

when the image data of the substrate is obtained, process the image data of the substrate;

detect reference data;

compare and analyze the image data of the substrate and the reference data; and

determine whether the nozzle is in a good condition or defective based on the analysis between the image data of the substrate and the reference data.

14 . The nozzle inspection unit of claim 13 , wherein the controller is further configured to determine a class related to the image data of the substrate from among the predefined classes and detect the reference data from among training data included in the determined class.

15 . The nozzle inspection unit of claim 1 , wherein, during the generating of the remaining ones of the plurality of merged image data besides the first merged image data, the controller is further configured to only rotate one or more content making a duplicated one the first merged image data at the third number of random rotation angles instead of an entirety of the duplicated one of the first merged image data at the third number of random rotation angles, the one or more content containing only one or more defects identified in the second number of defect image data.

16 . The nozzle inspection unit of claim 1 , wherein

the plurality of image data is collected by photographing substrates that are currently being treated on a substrate treatment apparatus containing the nozzle inspection unit,

the controller only generates the plurality of merged image data using the second number of the defect image data related to the defective nozzle if the plurality of image data that is collected contains the defect image data related to the defective nozzle, and

in an instance where the plurality of image data that is collected only contains the good image data related to the normal nozzle, the controller is further configured to collect supplemental defect image data related to the defective nozzle from a source different from and external to the substrate treatment apparatus, the supplemental defect image date not being associated with the substrates that are currently being treated on the substrate treatment apparatus.

17 . A nozzle inspection unit comprising:

a controller configured to:

collect a plurality of image data related to nozzles;

classify the plurality of image data according to predefined classes;

generating, after classifying the plurality of image data, a plurality of merged image data by:

first merging a first number of good image data related to a normal nozzle from among the plurality of image data with a second number of defect image data related to a defective nozzle from among the plurality of image data to obtain a first merged image data of the plurality of merged image data, the second number being smaller than the first number; and

generating remaining ones of the plurality of merged image data besides the first merged image data by duplicating and rotating the first merged image data at a third number of random rotation angles such that each of the remaining ones of the plurality of merged image data besides the first merged image data is the first merged image data rotated at one of the third number of random rotation angles;

train the plurality of merged image data after the plurality of merged image data is generated; and

generate a plurality of final image data from the plurality of merged image data based on a training result of the training of the plurality of merged image data,

wherein the controller is further configured to:

generate a number of the final image data greater than a reference number from a number of the defect image data smaller than the reference number,

obtain fake image data similar to the merged image data by training the merged image data, wherein the controller obtains the fake image data using a GAN, and

remove the fake image data similar to the merged image data from the plurality of merged image data and generates the final image data based on the remaining merged image data.

18 . A substrate treatment apparatus comprising:

a support configured to support a substrate while the substrate is treated;

an inkjet head unit including a plurality of nozzles and configured to jet a substrate processing liquid onto the substrate using the nozzles;

a gantry unit having the inkjet head unit installed thereon and configured to move the inkjet head unit on the substrate; and

a controller configured to inspect the nozzles,

wherein the controller is configured to:

collect a plurality of image data related to the nozzles;

classify the plurality of image data according to predefined classes;

generating, after classifying the plurality of image data, a plurality of merged image data by:

first merging a first number of good image data related to a normal nozzle from among the plurality of image data with a second number of defect image data related to a defective nozzle from among the plurality of image data to obtain a first merged image data of the plurality of merged image data, the second number being smaller than the first number; and

generating remaining ones of the plurality of merged image data besides the first merged image data by duplicating and rotating the first merged image data at a third number of random rotation angles such that each of the remaining ones of the plurality of merged image data besides the first merged image data is the first merged image data rotated at one of the third number of random rotation angles;

train plurality of merged image data after the plurality of merged image data is generated; and

generate a plurality of final image data from the plurality of merged image data based on the training result.

19 . The substrate treatment apparatus of claim 18 , wherein the substrate treatment apparatus is configured to perform pixel printing on the substrate.

Assignments (1)
ASSIGNMENT OF ASSIGNOR'S INTEREST Recorded May 1, 2026
From: LIM, MYEONG JUN; KIM, KWANG SUP; SONG, YEON CHUL; YOO, JI HOON; LEE, JONG MIN; OH, JUN HO; PARK, YOUNG HO
To: SEMES CO., LTD.
Reel/Frame 074546/0021 →
Priority Claims (1)
KR 10-2021-0189464 · Dec 28, 2021 · national
Continuity (1)
Related Publication 20230206424A1 · Jun 29, 2023
References Cited (30)
US 11605003B2 · Ki et al. · 2023 [cited by applicant]
US 11928810B2 · Okada · 2024 [cited by applicant]
US 12260548B2 · Okada · 2025 [cited by applicant]
US 20070182955A1 · Tomita · 2007 [cited by examiner]
US 20190057499A1 · Dekel-Klein · 2019 [cited by examiner]
US 20190248153A1 · Muehl · 2019 [cited by examiner]
US 20200005070A1 · Ambikapathi · 2020 [cited by examiner]
US 20210031507A1 · Haik · 2021 [cited by examiner]
US 20210073972A1 · Wu · 2021 [cited by examiner]
US 20210183036A1 · Gurudath · 2021 [cited by examiner]
US 20210287353A1 · Soltwedel · 2021 [cited by examiner]
US 20210304383A1 · Okada · 2021 [cited by examiner]
US 20210312607A1 · Lin · 2021 [cited by examiner]
US 20210366101A1 · George Boehm, Jr. · 2021 [cited by examiner]
US 20220234064A1 · Bremer · 2022 [cited by examiner]
US 20220270360A1 · Mendlovic · 2022 [cited by examiner]
US 20220281177A1 · Gu · 2022 [cited by examiner]
US 20220309634A1 · Atwood · 2022 [cited by examiner]
US 20220374720A1 · Qu · 2022 [cited by examiner]
US 20230004814A1 · Cai · 2023 [cited by examiner]
US 20230073418A1 · Chhaya · 2023 [cited by examiner]
US 20230264483A1 · Levant · 2023 [cited by examiner]
US 20240095983A1 · Goodwin · 2024 [cited by examiner]
CN 113447494 · 2021 [cited by applicant]
KR 1020060112855 · 2006 [cited by applicant]
KR 1020160083055 · 2016 [cited by applicant]
KR 1020190021967 · 2019 [cited by applicant]
KR 102021138 · 2019 [cited by applicant]
Office Action dated Mar. 18, 2026 for Chinese Patent Application No. 202211531878.9 and its English translation by Google Translate. [cited by applicant]
Notice of Allowance dated Apr. 16, 2025 for Korean Patent Application No. 10-2021-0189464 and its English translation from Global Dossier. [cited by applicant]