Conventional vs. model-based measurement of patterned line widths from scanning electron microscopy profiles

Scanning electron microscopy (SEM) is a practical tool to determine the dimensions of nanometer-scale features. Conventional width measurements use arbitrary criteria, e.g., a 50 % threshold crossing, to assign feature boundaries in the measured SEM intensity profile. To estimate the errors associated with such a procedure, we have simulated secondary electron signals from a suite of line shapes consisting of 30 nm tall silicon lines with varying width, sidewall angle, and corner rounding. Four different inelastic scattering models were employed in Monte Carlo simulations of electron transport to compute secondary electron image intensity profiles for each of the shapes. The 4 models were combinations of dielectric function theory with either the single-pole approximation (SPA) or the full Penn algorithm (FPA), and either with or without Auger electron emission. Feature widths were determined either by the conventional threshold method or by the model-based library (MBL) method, which is a fit of the simulated profiles to the reference model (FPA + Auger). On the basis of these comparisons we estimate the error in the measured width of such features by the conventional procedure to be as much as several nanometers. A 1 nm difference in the size of, e.g., a nominally 10 nm transistor gate would substantially alter its electronic properties. Thus, the conventional measurements do not meet the contemporary requirements of the semiconductor industry. In contrast, MBL measurements employing models with varying accuracy differed one from another by less than 1 nm. Thus, a MBL measurement is preferable in the nanoscale domain.