Compensation for transient chamber wall condition using real-time plasma density feedback control in an inductively coupled plasma etcher

Abstract

Reactive ion etch processing is known to exhibit significant variability in final etch performance due to wall condition. Previous studies have shown that neutral species transients depend strongly upon chamber seasoning. In this article, we show simultaneous measurements of wall-state-induced changes in plasma density and poly-Si etch rate, and demonstrate a real-time feedback control system that corrects for the plasma density variation. We demonstrate that controlling the plasma density to constant value eliminates the classic “first wafer effect” in Cl2 etching of poly-Si in a Lam 9400 transformer coupled plasma (TCP). Chamber conditions studied include fluorination/cleaning by C2F6 plasmas compared to chlorination/deposition from Cl2 plasmas. Transient density changes due to wall condition were measured using a microwave resonance cavity technique called broadband rf. Following chamber fluorination, broadband data show a significant drop in electron density (∼39%) from nominal levels. This is followed by subsequent partial recovery of the nominal density during a 60 s Cl2 etch. Independent measurements correlate strongly with the broadband signals; particularly real-time poly-Si etch rate and SiCl4 etch product concentration. Observed real-time variations in all signals were then compensated using a single input/single output proportional-integral (PI) feedback control algorithm, in which the broadband peak frequency is the system output variable and TCP power is the system actuator. Such PI control not only stabilizes broadband peak frequencies, but also steadies poly-Si real-time etch rate and SiCl4 etch product concentration. This compensation scheme is then applied over multiple runs to reduce etch depth variability due to chamber condition. Results of this control system are shown to attenuate first wafer effects by a factor of 3 and reduce overall etch depth variation from run to run by an additional 33% compared to standard manufacturing practice. When applied to patterned wafer etches, feedback control of plasma density with an oxide hardmask is found not to alter critical dimensions or profile in any observable way. Only total etch depth appears to be affected using density control, as the real-time etch rate is in general increased in comparison to open-loop etching.