Experimental and model-based study of the robustness of line-edge roughness metric extraction in the presence of noise

As critical dimensions shrink, line edge roughness (LER) and linewidth roughness (LWR) become of increasing concern. Crucial to the goal of reducing LER is its accurate characterization. LER has traditionally been represented as a single rms value. More recently, the use of power spectral density (PSD), height-height correlation (HHCF), and a versus length plots has been proposed in order to extract the additional spatial descriptors of correlation length and roughness exponent. Here, the authors perform a modeling-based noise-sensitivity study on the extraction of spatial descriptors from line-edge data as well as an experimental study of the robustness of these various descriptors using a large data set of recent extreme-ultraviolet exposure data. The results show that in the presence of noise and in the large data set limit, the PSD method provides higher accuracy in the extraction of the roughness exponent, whereas the HHCF method provides higher accuracy for the correlation length. On the other hand, when considering precision, the HHCF method is superior for both metrics. (C) 2007 American Vacuum Society.