In this study, the Gaussian mixture model (GMM) was modified and implemented to determine the real-time endpoint of SiO2 plasma etching using optical emission spectrum analysis. Optical emission spectroscopy (OES) signals were collected from the SiO2 plasma etching processes, and the modified GMM was applied to SiO2 etching with relative areas of 8.0, 4.0, and 1.0 %. Consequently, the sensitivity of OES signals was improved by ~5.5 times, and the sensitivity factor of the modified GMM was increased by approximately two times, compared with those of the modified K-means cluster analysis (another clustering technique). In addition, 60 peaks related to the reactants were selected out of 6144 signals to improve the sensitivity of the modified GMM with full-spectrum wavelengths. The modified GMM analysis using the 60 reactant-related peaks exhibited a higher sensitivity (~1.4 times) than that with 6144 full-spectrum OES signals. Thus, the modified GMM can be a suitable and effective clustering technique for etching endpoint detection.

中文翻译：

---

使用改进的高斯混合模型提高 SiO2 等离子体蚀刻终点检测的灵敏度

在本研究中，修改并实施了高斯混合模型 (GMM)，以使用光发射光谱分析确定 SiO2 等离子体蚀刻的实时终点。从 SiO2 等离子体蚀刻过程中收集光发射光谱 (OES) 信号，并将改进的 GMM 应用于 SiO2 蚀刻，相对面积为 8.0、4.0 和 1.0%。因此，与改进的 K 均值聚类分析（另一种聚类技术）相比，OES 信号的灵敏度提高了约 5.5 倍，并且改进的 GMM 的灵敏度因子提高了大约两倍。此外，从 6144 个信号中选择了 60 个与反应物相关的峰，以提高具有全光谱波长的改进 GMM 的灵敏度。使用 60 个反应物相关峰的改进 GMM 分析显示出比 6144 个全谱 OES 信号更高的灵敏度（~1.4 倍）。因此，改进的 GMM 可以是一种适合且有效的用于蚀刻终点检测的聚类技术。