Scanning electron microscope dimensional metrology using a model‐based library

TLDR

The semiconductor electronics industry relies on secondary electron imaging for dimensional measurements critical to process control and failure analysis, yet required tolerances are tighter than the SEM’s spatial resolution, rendering traditional image‑processing techniques insufficient. The study proposes an alternative method that uses Monte Carlo electron‑transport modeling to determine edge positions in SEM images. Specimen geometry is parameterized and iteratively adjusted to minimize least‑squares error against measured images, with a pre‑computed library of Monte Carlo simulations used to interpolate a fast surrogate model for rapid computation. The method achieves measurement results that agree well with cross‑section data and improves same‑site repeatability by up to a factor of three compared to conventional techniques. Published in 2005 by John Wiley & Sons, Ltd.

Abstract

Abstract The semiconductor electronics industry places significant demands upon secondary electron imaging to obtain dimensional measurements that are used for process control or failure analysis. Tolerances for measurement uncertainty and repeatability are smaller than the spatial resolution of edges in the scanning electron microscope (SEM) that is used to perform the measurements. Image processing techniques, historically used to identify edge locations, are inadequate under these conditions. An alternative approach, based upon Monte Carlo electron transport modeling to assign edge positions, has been developed. The specimen shape is parameterized, and parameters are iteratively adjusted to produce the best least squares fit to the measured image. Because Monte Carlo simulators are too slow to be used directly in such an iterative calculation, the Monte Carlo technique is used relatively few times to construct a library of results for parameters spanning the process space of interest. A function that interpolates the library then becomes a surrogate that is used to rapidly compute the model function as needed. This procedure has yielded measurement results from top‐down SEM images that are in good agreement with cross‐section measurements and that have as much as a factor of 3 better same‐site repeatability than the more traditional techniques. Published in 2005 by John Wiley & Sons, Ltd.