AVS 55th International Symposium & Exhibition | |
Plasma Science and Technology | Tuesday Sessions |
Session PS-TuP |
Session: | Plasma Science Poster Session |
Presenter: | M. Klick, Plasmetrex GmbH, Germany |
Authors: | M. Klick, Plasmetrex GmbH, Germany L. Eichhorn, Plasmetrex GmbH, Germany R. Rothe, Plasmetrex GmbH, Germany |
Correspondent: | Click to Email |
The virtual metrology is the prediction of plasma etch rates and critical dimensions is based on measurements of pre-process values and plasma parameters. This can be realized by a self-consistent model of plasma and process or non- self-consistent (empirical) model with plasma parameters measured in real-time and in situ. A self-consistent plasma model must describe the complete plasma process. Already the real-time solution for a self-consistent plasma model is impossible, in particular due to large amount of also chemical mechanisms. The effort can be reduced dramatically by usage of plasma parameters, describing the main physical and chemical mechanisms. The most important issue of development of a chemical / physical model is to identify the key parameters. Tool parameters reflect only the tool properties but not the real process. The most important process parameters, called key parameters, are plasma density and electron collision rate by SEERS (physics), RF-parameters by VI-probe, radical/polymer concentrations by OES (chemistry). Our reduced approach for reactive ion etching as described above assume a combination of a physical (sputter) effect, a pure chemical (surface) reaction, and a physical-chemical mechanism. Despite the pressure is usually kept constant, the real important parameter is the density of the gas (neutrals) which depends on the temperature additionally. The gas temperature is usually not available but replaced here by the electron collision rate which is proportional to the gas density and so reciprocally proportional to the gas temperature. The model was applied to oxide etch with F-chemistry, the unknown coefficients were determined and prediction error was shown to be less than 5%.