Abstract
In this paper, one of the major contributions to the OCD metrology error, resulting from within-wafer variation of the refractive index/extinction coefficient (n/k) of the substrate, is identified and quantified. To meet the required metrology accuracy for the 65-nm node and beyond, it is suggested that n/k should be floating when performing the regression for OCD modeling. A feasible way of performing such regression is proposed and verified. As shown in the presented example, the measured CDU (3σ) with n/k fixed and n/k floating is 1.94 nm and 1.42 nm, respectively. That is, the metrology error of CDU committed by assuming n/k fixed is more than 35% of the total CDU.