Fault Detection and Classification of PECVD equipment Using Wavelet Transform

• Main Authors: CHUN-YING HUANG, 黃俊穎
• Other Authors: Yaw-Jen Chang
• Format: Others
• Language: zh-TW
• Published: 2006
• Online Access: http://ndltd.ncl.edu.tw/handle/66305584643975763701

碩士 === 中原大學 === 機械工程研究所 === 94 === The plasma-enhanced chemical vapor deposition (PECVD) process has been developed for thin film deposition in the semiconductor and TFT-LCD industries for decades. During the process, plasma plays an important role to the success of deposition. It is generated by RF power, gas, pressure and temperature. Any defect factors will influence the forming of the plasma and, moreover, result in the power loss. In this paper, wavelet transform was used for fault detection and classification on plasma. The causes of power loss in KAI800 PECVD equipment, such as the damage of RF generator, the environmental change in the reactor, and the malfunction of electromagnetic valve, were investigated. Due to the merits of wavelet theory on time-frequency and multi-resolution analysis, the raw data collected from KAI800 tool were then decomposed using wavelet basis functions. These parameters include delivery power, reflection power, temperature, pressure, and gas flow. Then, the wavelet coefficients were monitored incorporating with the threshold limits for fault detection. Whenever a fault is detected, it will be classified, based on the causes of power loss, in order to provide a proper diagnostic determination for maintenance. This approach can also promote a higher overall equipment efficiency. In our experiments, faults were classified into seven types. Their causes were further categorized into three problems. Data of normal runs and abnormal runs were collected and examined using wavelet transform. As a result, RF power loss could be correctly detected and classified. In this paper, the wavelet-based fault detection and classification approach is shown as a powerful tool for equipment malfunction.