化學氣相沉積設備之故障偵測與診斷

Fault Detection and Diagnosis of PECVD Equipment

戴賢政(Hsien-Cheng Tai)

指導教授 ： 張耀仁

中原大學/工學院/機械工程學系/碩士(2007年)

https://doi.org/10.6840/cycu200700623

摘要

PECVD在半導體及TFT-LCD製程中，為一極重要的成膜技術，其中電漿乃為最重要的關鍵之一，由於PECVD製程中含有許多無法預測的變數，當電漿受異常狀況(壓力異常、射頻供應器異常、製程氣體供輸異常、成膜環境異常等)影響時，輕者可能造成設備當機，重者可能致使產品的損失及設備稼動率的下降。因此，為了探討此種非穩定製程下的異常變數，針對設備參數的實際訊號進行偵測診斷與分類的模擬分析，再藉由模擬分析的結果診斷出故障種類。
小波理論之小波轉換具對時-頻區域化訊號進行分析的優點，可滿足對PECVD這種隨時間改變的訊號進行波形拆解分析，本論文擷取TFT-LCD中PECVD的實際反射功率訊號進行分析，利用小波近似與細節對原始波形拆解後，透過訊息理論中的失真率計算方法去產生正常反射功率訊號與故障反射功率訊號的波形特徵值，之後，再結合競爭式類神經網路的競爭式學習法去推論不同訊號的集群分類後來達到故障診斷的機制。
在此，除了利用小波理論對波形進行分析外，另外，尚針對反射功率訊號在取樣前後的差分值計算來檢測反射功率訊號在連續時間上的變化差異，並以此差異的加總之總變化率透過競爭式類神經網路進行故障的分類。由結果的比較得知，本論文在電磁閥故障及射頻供應器的故障診斷上可產生較準確的診斷結果。

關鍵字

射頻供應器 ； 化學氣相沉機 ； 失真率 ； 小波理論 ； 競爭式網路

並列摘要

The plasma-enhanced chemical vapor deposition (PECVD) process is always an important technology for the thin-film deposition in the semiconductor and TFT-LCD fabrications. Plasma, in PECVD, is a key parameter for the success of deposition. In case the plasma is subject to any abnormal influences such as deviant pressure, faulty RF generator, unstable gas supply, or abnormal deposition environment, it may lead to equipment breakdown and cost loss. Certainly, the overall equipment effectiveness (OEE) reduces. Therefore, this thesis presents a systematic approach of fault diagnosis for PECVD process.
Wavelet transform can give a time-frequency localization of the signal for further analysis. In this study, the signals collected from PECVD equipment were decomposed using Daubechies wavelet basis functions. The corresponding wavelet and scaling function coefficients were then obtained. In addition, the feature residuals were calculated by the principle of distortion rate which is commonly used in the information theory. Finally, incorporating with the artificial neural network, different faults can be correctly clustered.
In the mean time, beside to use the formula of Wavelet Transform to execute the waveform analysis, I do also aim to the reflection power signal to measure the differential of variance tolerance by difference method in front and post of sampling time point for the data used to exam the reflection power signal in continuity timing deviation. The squawks or malfunction analysis is using summarize differences total change rate through the CL- ANN (Competitive Learning Artificial Neutral Network) works.
From the conclusion to analysis the squawk or malfunction to solenoid valve and radio frequency power supply, it can get more accuracy in judgment or diagnosis.

並列關鍵字

PECVD ； Wavelet Theory ； RF ； Competitive Network ； Distortion Rate

參考文獻

[12] 陳源杰, “應用小波理論於薄膜製程之射頻功率故障偵測＂，中原大學碩士論文，2004.

[2] Kim.B and May.G.S , “Real-time diagnosis of semiconductor manufacturing equipment using a hybrid neural network expert system,” IEEE Trans. Comp. Pack. Manufact. Technol., v 20, n 1, p 39-47, Jan, 1997.

[5] Oohama, Yasutada. , “ Rate-Distortion Theory for Gaussian Multiterminal Source Coding Systems With Several Side Information at the Decoder , ” IEEE Transactions on Information Theory, v 51, n 7, p 2577-2593, July, 2005

[6] Zhang, Jian. ; Yu, Sheng-Sheng. ; Zheng, Xin-Wei. ,“ The Computation Model of Code Error Distortion Based on the Rate-Distortion Theory, ” ICESS 2005 - Second International Conference on Embedded Software and Systems, p 595-598, 2005

[13] 林昆達, “小波理論與類神經網路在橋樑非破壞檢測之應用＂，中原大學碩士論文，2003.