Unbiased line width roughness measurements with critical dimension scanning electron microscopy and critical dimension atomic force microscopy

Abstract

With the constant decrease of semiconductor device dimensions, line width roughness (LWR) becomes one of the most important sources of device variability and thus needs to be controlled below 2 nm for the future technological nodes of the semiconductor roadmap. The LWR control at the nanometer scale requires accurate measurements, which are inevitably impacted by the noise level of the equipment that causes bias from true LWR values. In this article, we compare the capability of two metrology tools, the critical dimension scanning electron microscopy (CD-SEM) and critical dimension atomic force microscopy (CD-AFM) to measure the true line width roughness of silicon and photoresist lines. For this purpose, we propose several methods based on previous works to estimate the noise level of those two equipments and thus extract the true LWR. One of the developed methods for the CD-SEM technique generalizes the power spectral densities (PSD) fitting method proposed by Hiraiwa and Nishida with a more universal autocorrelation function, which includes both correlation length and roughness exponent. However, PSD fitting method could not be used with CD-AFM due to the time consuming character of this technique. Hence, other experimental protocols have been set up for CD-AFM in order to accurately characterize the LWR. Our study shows that the CD-SEM technique combined with our PSD fitting method is much more powerful than CD-AFM to get all roughness information (true LWR, correlation length, and roughness exponent) with a good accuracy and efficiency on hard materials such as silicon. Concerning materials degradable under electron beam exposure such as photoresist, the choice is more disputable, since ultimately they are impacted by the electrons. Fortunately, our PSD fitting method allows working with low number of integration frames, which limits the resist degradation. Besides, we have highlighted some limitations of the CD-AFM technique due to the tip diameter. This technique can underestimate LWR if the roughness presents significant amount of high frequency components, as it is the case for photoresist patterns. So far, there is no universal technique to accurately estimate the LWR on any materials. Nevertheless, the CD-SEM protocol we propose opens a way for a better characterization of the photoresist LWR after lithography and a better understanding of the LWR transfer during the plasma etching steps involved in gate patterning processes.