Virtual Measurement Combine OCD and FDC Through Optimized Random Forest Machine Learning Algorithm Assists on Post Etching Monitor

Etching is an important process in semiconductor technology. Inconsistency in post Etching wafer uniformity and geometry affects downstream process margin which leads to various yield issues. Post etching monitor such as top, middle or bottom critical dimension is necessary to ensure the process meet the specification and stability. Wafer geometry will be measured and out of control wafers will be reworked or Scraped. To reduce rework and scraped, Etching is normally employed with process control systems like APC (Advanced process Control). The Operation of APC usually relies on conventional measurement methods such as CDSEM or Optical Critical Dimension(OCD) for Feedback or Feedforward. Conventional metrology tools are in charge of ensuring the data quality for APC which need huge sampling data but challenged by the limitation of time and cost. Fault Detection and Classification (FDC) data reflects machine real time status that provide information for diagnosing process quality. However, the selection of FDC parameters is very dependent on the engineering experience, thus it can only be used as an auxiliary means of monitor. In this paper, we combine the traditional OCD measurement with the real-time FDC data of Etching chamber to establish the machine learning prediction model by training the optimized random forest algorithm and carried out virtual measurement finally.