請用此 Handle URI 來引用此文件:
http://tdr.lib.ntu.edu.tw/jspui/handle/123456789/88544
標題: | 噴射式大氣電漿系統應用: 斜板沉積均勻霧度薄膜、無光罩透明導電圖樣製程、電漿光譜與薄膜性質關聯性分析 Applications of Atmospheric Pressure Plasma Jet System: Oblique Angle Deposition of Hazy TCOs, Laser-Assisted Maskless Deposition of Patterned TCOs, and Analysis of the Correlation between Plasma Spectra and TCO Properties |
作者: | 李允中 Yun-Chung Lee |
指導教授: | 莊嘉揚 Jia-Yang Juang |
關鍵字: | 噴射式大氣電漿,鎵摻雜氧化鋅,傾斜基板,雷射輔助,光譜分析,機器學習, Atmospheric Pressure Plasma Jet System,Ga-doped Zinc Oxide,Tilted substrate,Laser-assisted,Spectroscopic analysis,Machine learning, |
出版年 : | 2023 |
學位: | 博士 |
摘要: | 本論文是針對本研究團隊長期開發研究的噴射式大氣電漿鍍膜系統(Atmospheric pressure plasma jet system, APPJ)進行新的應用開發,並對其成因與原理進行探討與分析。研究內容共可分為三個部份,一、以傾斜基板輔助鍍製均勻超高霧度薄膜;二、結合雷射系統鍍製透明導電圖樣;三、透過機器學習分析電漿光譜資訊特徵與薄膜特性之相關性。
首先,本研究結合了傾斜基板和NTUAPPJ系統,只需單一步驟便能沉積出具有超過80%的均勻超高霧度(非均勻性約為1.35%)、高光學穿透率(88%,不包含基板)和電阻率1.96 × 10E-3 Ωcm的ZnO:Ga薄膜。從研究結果顯示,高霧度是由於偏向氣流將前驅物帶往下游處所形成的吸附顆粒所引起的,本研究將此現象稱做為「預沉積」。而上游未受預沉積影響的「虛擬區域」則做為下游目標樣本的先導,由於沉積條件不同所以不列入均勻霧度計算。與現有常見的光散射製程不同,本研究提出的方法僅需一個步驟便製作出均勻霧度的高品質薄膜,無需額外的前/後處理或其它製程變數。 第二部分,結合雷射輔助NTUAPPJ系統,開發出一種單步驟、無遮罩、不須墨水及奈米材料的製程,能在玻璃基板上沉積透明導電的ZnO:Ga圖樣。利用電腦輔助控制振鏡運作,使連續型二氧化碳雷射能沿著指定CAD圖像路徑對玻璃基板掃描加熱。而當APPJ對整個區域噴塗前驅物時,便會沿著指定CAD圖樣上生長出ZnO:Ga圖樣。沉積的ZnO:Ga圖樣厚度約100 nm,可見光穿透度非常高,並且具有非常低的電阻率(7.89 × 10E-4 Ωcm)。同樣的,此製程僅需單一步驟、無須額外的前/後處理或準備額外的前驅物溶液。整個製程在常溫常壓下完成,無需加熱基板,具有快速、低成本、可連續鍍製透明導電圖樣的潛力,以作為透明電極的應用。 最後,本研究利用光譜分析儀蒐集NTUAPPJ運作中的氮氣電漿以及添加前驅物的電漿光譜。先是對光譜特徵進行觀察,確立部分會對光譜特徵產生影響的原因,建立出光譜取樣實驗的準則。接著透過機器學習模型對大量的光譜資料進行特徵分析,先以主成分分析模型將光譜特徵可視化,觀察光譜資料的分布情況。接著根據樣本的製程參數(載氣進氣管深、前驅物消耗量)以及薄膜特徵(薄膜厚度、片電阻值、電阻率)等,對樣本以及對應的光譜資料進行排序分類,再運用七種監督式機器學習模型對部分光譜特徵進行學習,最後對未經過模型學習的光譜資料進行辨識,觀察模型能將光譜資料歸類至對應分類的準確率。再經過兩種自定義的前處理之後,辨識結果顯示,光譜特徵與薄膜厚度呈現高度的相關性,在隨機森林分類器中獲得最高89%的辨識率;同時,對於其餘四種特徵的辨識率則落在60%上下,與光譜特徵之間的關聯性並不突出。 This dissertation aims to develop new applications for the Atmospheric Pressure Plasma Jet System (APPJ) developed by our research group with detailed discussion and analysis of those applications' principles, pros, and cons. The dissertation consists of three parts: (1) using tilted substrates to deposit uniformly hazy films, (2) using a laser-assisted APPJ system to fabricate transparent conductive patterns, and (3) analysis of the correlation between plasma spectroscopic information and film characteristics using machine learning techniques. First, we combine tilted substrates with the NTU-APPJ system to deposit ZnO:Ga films with a uniform ultrahigh haze of over 80% (non-uniformity approximately 1.35%), the high optical transmittance of 88% (referenced to the substrate) and resistivity of 1.96 × 10E-3 Ωcm. The results show that the high haze is caused by the phenomenon, referred to as the "pre-deposition" of adsorbed particles carried downstream by the inclined airflow. The upstream unaffected area, or the "dummy area," serves as a guide part for downstream samples since the deposition conditions differ. Unlike conventional light scattering processes, our method achieves uniform and high-quality films in a single step without the need for additional pre- or post-processing or changing process variables. Second, we built a laser-assisted NTU-APPJ system to develop a single-step, maskless, particle-free, and ink-free process for depositing transparent conductive ZnO:Ga patterns on glass substrates. Controlling the galvanometer using computer-aided design images, a continuous CO2 laser scans over the glass substrate to heat it. When the APPJ sprays precursor materials onto the entire area, ZnO:Ga patterns grow according to the specified CAD design. The deposited ZnO:Ga patterns are approximately 100 nm thick, visually transparent, and exhibit a remarkably low resistivity of 7.89 × 10E-4 Ωcm. Similarly, this process requires only a single step without additional pre- or post-processing or the preparation of additional precursor solutions. The entire process is conducted at ambient conditions without substrate preheating, offering the potential for rapid, low-cost fabrication of transparent conductive patterns and electrodes. At last, we collect spectroscopic data from the NTU-APPJ system operating with nitrogen plasma and precursor-added plasma using a spectroscopic analyzer. We first observe the spectral features and identify factors influencing them to establish criteria for spectral sampling experiments. Next, we analyze the spectral data using machine learning models, sorting and classifying them based on corresponding sample process parameters (depth of carrier gas inlet, precursor consumption) and film characteristics (film thickness, sheet resistance, resistivity). Using principal component analysis, we visualize the spectral features and propose two spectral data preprocessing methods. We train and recognize the spectral categories using seven supervised machine-learning models to identify their corresponding parameter characteristics. The recognition results demonstrate the highest correlation between spectral features and film thickness, achieving a recognition accuracy of up to 89% using the random forest classifier. |
URI: | http://tdr.lib.ntu.edu.tw/jspui/handle/123456789/88544 |
DOI: | 10.6342/NTU202302041 |
全文授權: | 同意授權(限校園內公開) |
顯示於系所單位: | 機械工程學系 |
文件中的檔案:
檔案 | 大小 | 格式 | |
---|---|---|---|
ntu-111-2.pdf 目前未授權公開取用 | 10.11 MB | Adobe PDF | 檢視/開啟 |
系統中的文件,除了特別指名其著作權條款之外,均受到著作權保護,並且保留所有的權利。