A streaming multi-GPU implementation of image simulation algorithms for scanning transmission electron microscopy

Abstract

Simulation of atomic-resolution image formation in scanning transmission electron microscopy can require significant computation times using traditional methods. A recently developed method, termed plane-wave reciprocal-space interpolated scattering matrix (PRISM), demonstrates potential for significant acceleration of such simulations with negligible loss of accuracy. Here, we present a software package called Prismatic for parallelized simulation of image formation in scanning transmission electron microscopy (STEM) using both the PRISM and multislice methods. By distributing the workload between multiple CUDA-enabled GPUs and multicore processors, accelerations as high as 1000 × for PRISM and 15 × for multislice are achieved relative to traditional multislice implementations using a single 4-GPU machine. We demonstrate a potentially important application of Prismatic, using it to compute images for atomic electron tomography at sufficient speeds to include in the reconstruction pipeline. Prismatic is freely available both as an open-source CUDA/C++ package with a graphical user interface and as a Python package, PyPrismatic.

Microscopic model for the stacking-fault potential and the exciton wave function in GaAs

“…The frames were up-sampled 2× prior to non-rigid alignment, followed by template matching parallel to the fault direction. Full multislice image simulations were conducted with the PRISM code [34] for several candidate structures from ab initio calculations. Simulations were performed using a 1 × 4 tiling for crystal thicknesses of 50, 100, and 150 u.c., corresponding to 20, 40, and 60 nm, respectively.…”

Linear-scaling algorithm for rapid computation of inelastic transitions in the presence of multiple electron scattering

“…We also estimate the computation time of a conventional multislice simulation using Eq. (15) for the full nanoparticle with the same probe step size and 2Å slices along the beam direction. If further approximations are made to use only 1/9 of the grid for probe propagation and only evaluating transitions within 4Å of the probe (i.e.…”

“…Instead of the traditional approach of calculating the propagation of the beam for each scan position in a STEM raster independently, the PRISM method uses the multislice algorithm to calculate the scattering matrix for a particular electron microscope experiment and stores it in memory. The scattering matrix, which propagates components of the electron illumination through the imaging object, can then be rapidly applied to the illumination wave function for each scan position to compute the output wave [15].…”

Experimental and Simulation Methods in Scanning Electron Nanobeam Diffraction

“…In this talk, I will also show 4D-STEM strain measurements from amorphous samples. Figure 1d shows the algorithmic steps required for a STEM simulation, using either the multislice method or the newly developed PRISM algorithm [2,3]. First, we build a model of the atomic positions for a given sample, and then use tabulated atomic scattering factors to generate projected potentials.…”

“…Finally, we can save these outputs as 2D, 3D or 4D output files. In this talk I will demonstrate both simulation methods using the Prismatic code [3]. Figure 1e shows a study where experimental and simulated CBED patterns were matched to determine the best fit for sample thickness and composition in an STO-LMO multi-layer sample [4].…”