Multi-way principal component analysis for the endpoint detection of the metal etch process using the whole optical emission spectra

An endpoint detection algorithm based on multi-way principal component analysis (MPCA) is developed for plasma etching processes. Because many endpoint detection techniques use a few manually selected wavelengths, noise renders them ineffective and it is hard to select important wavelengths. Furthermore, process drift and faulty condition should be considered for more robust endpoint detection at the same time. In this paper, MPCA with the whole optical emission spectra is used for effective endpoint detection using a large set of data. And the fault detection was achieved by concept of ‘product’ and ‘mean deviation value’ chart with the result of each wafer’s endpoint detection. The product was defined by the multiples of OES data with loading vector and mean deviation chart was defined by a chart of the difference between the product value of the target wafer and mean value of previous wafers. Therefore, a robust model for endpoint detection can be developed by excluding faulty wafers. This approach is successfully applied to the metal etch process of TiN/Al-0.5%Cu/TiN/Oxide stack in an inductively coupled BCl₃/Cl₂ plasma. The optical emission signal intensities of the 129 wavelengths were measured and saved in a four-dimensional (wavelengths, time, intensity, and wafers) matrix for the subsequent data processing. With this approach the endpoint signal was improved with the whole emission spectra and the process drift was considered by MPCA after information of faulty wafers was discarded.