Software acceleration techniques for the simulation of scanning electron microscope images

Abstract

A scanning electron microscope (SEM) simulator was developed based on the models used in the MONSEL software. This simulator extends earlier work by introducing an object-oriented framework and adding optimization methods based on precomputation of electron trajectories. Several optimizations enable speedup by factors of 5-100 on a single processor over unoptimized simulations without introducing additional approximations. The speedup for a particular surface depends on the self-similarity of the surface at the scale of the electron penetration depth. We further accelerate by parallelizing the calculations for a total speedup of about 100-2000 on 30 processors. The goal of this work was to create a system capable of simulating a quantitatively accurate SEM image of a relatively unconstrained surface. Results of this work include simulation software, optimization algorithms, performance measurements with various optimizations, and examples of simulated images.