Understanding the relationship between true and measured resist feature critical dimension and line edge roughness using a detailed scanning electron microscopy simulator

© 2010 American Vacuum Society. This article may be downloaded for personal use only. Any other use requires prior permission of the author and the American Vacuum Society. The electronic version of this article is the complete one and can be found at: http://dx.doi.org/10.1116/1.3517717DOI: 10.1116/1.3517717Top-down critical dimension scanning electron microscopy (SEM) is still the workhorse metrology tool used for nanoscale structure analysis, such as measurement of photoresist features, during integrated circuit manufacturing. However, the degree to which top-down SEM imaging can accurately be used to quantitatively determine the size, shape, and roughness characteristics of three-dimensional structures such as photoresist features has not been carefully characterized. A rigorous Monte Carlo simulation of scanning electron microscopy has been developed to probe the relationship between the roughness of a three-dimensional feature and the line edge roughness (LER) as measured by SEM. The model uses the differential Mott cross section to compute elastic scattering, while inelastic scattering and secondary electron generation are handled using dielectric function theory. The model can calculate the electron scattering for any arbitrary three-dimensional geometry. Experimental SEM measurements of photoresist nanostructures show good agreement with the simulation output. The critical dimension of the resist determined from SEM best matches the true resist feature width when the line edge is defined using a high image threshold because the roughness on the outer edge of the resist tends to cause an increase in SEM signal that is nonproportional to the amount of material on the outer edge of the feature. LER determined from SEM was found to be significantly smaller than the true resist feature sidewall roughness. The measured LER is typically greater than 50% smaller than the actual sidewall roughness