Abstract

Semiconductor manufacturers are increasingly reliant on optical emission spectroscopy (OES) to source information on plasma characteristics and process change. However, OES data is characterized by high dimension and by highly correlated variables. This makes it difficult to interpret process behaviour using OES measurements. It is therefore desirable to obtain more compact representations of the data using dimensionality reduction techniques such as Forward Selection Component Analysis (FSCA). In this paper we investigate non-linear extensions of FSCA based on polynomial expansions and Extreme Learning Machines and show, through a combination of simulated examples and OES recordings from a semiconductor plasma etch process, that they can yield more compact representations that classical FSCA.

Full Text
Published version (Free)

Talk to us

Join us for a 30 min session where you can share your feedback and ask us any queries you have

Schedule a call