Paper PS-FrM3
About the Surface Roughness Generated by Plasma Etching Processes

Friday, October 19, 2007, 8:40 am, Room 606

As the critical dimension of the features decrease for each new generation of CMOS transistors, the thickness of the layers being etched also decreases. When the thickness of the layer approaches 10 nm, a new problematic is emerging: the roughness generated by the plasma etching process could become comparable to the thickness of the layer being etched potentially causing selectivity issues. In this work, we have systematically analyzed the roughness generated in c-Si (100) and in p-Silicon when etched in high-density plasmas over a wide range of conditions (pressure, rf power) using SF₆, CF₄, Cl₂ and HBr chemistries. The roughness is characterized by AFM. In this work, we demonstrate unambiguously that high-density HBr and Cl₂ plasmas DO NOT generate roughness during etching but on the contrary tend to smooth the existing surface roughness if already present. By analyzing the time evolution of the shape of self organised silicon nanopilar (patterned using diblock-copolymers), we show that the smoothening properties of etching plasmas is due to shadowing effects : the "hills" receive a high radical flux than the "valley" and are thus etched faster. In contrast, F-based plasma generates a significant surface roughness whose amplitude increases with the etching time. However, we show that the roughness formed in these conditions is generated by micromasking of silicon by AlF_x particles generated by the sputtering of the (Al₂O₃) reactor walls. A high percentage of Al is indeed detected on the surface after etching in F-based plasmas. As a matter of fact when the chamber walls are intentionally coated by a carbon layer prior to the silicon etching process (thus preventing Al sputtering), the F-based plasmas behave like the other etching chemistries investigated: they rapidly smooth any existing roughness.