Plasma control using neural network and optical emission spectroscopy

Byungwhan Kim, Jung Ki Bae, Wan Shick Hong

Research output: Contribution to journalArticlepeer-review

22 Scopus citations


Due to high sensitivity to process parameters, plasma processes should be tightly controlled. For plasma control, a predictive model was constructed using a neural network and optical emission spectroscopy (OES). Principal component analysis (PCA) was used to reduce OES dimensionality. This approach was applied to an oxide plasma etching conducted in a CHF3 CF4 magnetically enhanced reactive ion plasma. The etch process was systematically characterized by means of a statistical experimental design. Three etch outputs (etch rate, profile angle, and etch rate nonuniformity) were modeled using three different approaches, including conventional, OES, and PCA-OES models. For all etch outputs, OES models demonstrated improved predictions over the conventional or PCA-OES models. Compared to conventional models, OES models yielded an improvement of more than 25% in modeling profile angle and etch rate nonuniformtiy. More than 40% improvement over PCA-OES model was achieved in modeling etch rate and profile angle. These results demonstrate that nonreduced in situ data are more beneficial than reduced one in constructing plasma control model.

Original languageEnglish
Pages (from-to)355-358
Number of pages4
JournalJournal of Vacuum Science and Technology A: Vacuum, Surfaces and Films
Issue number2
StatePublished - Mar 2005


Dive into the research topics of 'Plasma control using neural network and optical emission spectroscopy'. Together they form a unique fingerprint.

Cite this