基于腔体匹配的刻蚀设备智能化检测技术  

Study on Intelligent Inspection of Etching Equipment Based on Chamber Matching Technology

作　　者：李琛[1] 周涛 LI Chen;ZHOU Tao(Shanghai Integrated Circuits R&D Center Co.,Ltd.,Shanghai 201203,China.)

机构地区：[1]上海集成电路研发中心有限公司,上海201203

出　　处：《集成电路应用》2022年第7期32-37,共6页Application of IC

基　　金：国家科技重大专题课题(2011ZX02702_004)。

摘　　要：阐述腔体匹配(Chamber Matching, CM)是纳米尺度半导体工艺最为关键的挑战之一,随着特征尺寸的不断缩小,半导体工艺对细微误差变得越发敏感,对设备性能的一致性、稳定性提出了更高的要求。在量产Fab中,同一产品、工艺经过不同的腔体需获得一致的工艺结果,但由于刻蚀设备数量多,腔体数量大,腔体情况复杂,结果往往存在偏差。快速发现腔体状态变化,及时根因溯源,甚至提前预知腔体潜在问题对于半导体制造具有显著价值,传统FDC功能监控数据维度较少,分析方法单一,逐渐变得难以适应先进的工艺需求。提出了一套综合而有效的智能化分析方法,融合主成分分析(principle component analysis,PCA)和随机森林对传感器进行重要性排序,定位失配的工艺步骤(step)和问题传感器。针对设备数据特征维度大、干扰性强和冗余度高的特点,采用逐步回归,提炼高质量特征,改善模型的分类效果。最后通过与Fab工程师协同,更换部件,监控效果,完成对腔体问题的全周期监控,从而保证Fab生产高效稳定地进行。This paper discloses that Chamber matching(CM) is one of the most critical challenges in nano scale semiconductor technology. With the continuous reduction of feature size, semiconductor technology becomes more and more sensitive to small errors, and the consistency and stability of device performance are proposed higher requirements. In volume production Fab, the same product and process need to obtain consistent results and quality through different chambers. However, due to the large number of chambers in etch process, as well as the high complex conditions in a single chamber, the results often varies from each other. Quickly discovering condition changes of chambers, tracing the root cause in time, and even predicting potential problems in advance is of great value to semiconductor manufacturing. Traditional FDC function monitoring data has fewer dimensions and often with a single responsive method, which gradually becomes difficult to adapt to advanced nodes’ requirements. It proposes a comprehensive and effective intelligent method, which combines the principal component analysis and random forest to sort the importance of sensors and locate mismatched process steps and problem sensors. As for the large dimension, strong interference and high redundancy of equipment data features, stepwise regression is adopted to extract high-quality features to improve the classification effect of the model. Finally, by cooperating with Fab engineers, the tasks of replacing parts, monitoring the tools’ performance have been done, and the full-cycle monitoring of chamber conditions has been performed to ensure efficient and stable Fab production.

关 键 词：集成电路制造 腔体匹配 智能检测 随机森林 主成分分析 

分 类 号：TP212[自动化与计算机技术—检测技术与自动化装置] TB34[自动化与计算机技术—控制科学与工程]