CDI-Type II: Dimensionality-Reduction and Reconstruction Tools for Atom Probe Tomography

Iowa State University, Ames IA

Investigators

Abstract

This award supports a project to establish a new computational paradigm and infrastructure that will enhance three-dimensional (3-D) atomic reconstruction for an emerging new instrumentation technology; atom probe tomography (APT). The project will significantly advance research in parallel computing, computational geometry and graph-theoretic applications of extremely large and noisy data sets. APT is a powerful microscopy tool that enables spatial resolution of hundreds of millions of atoms at the sub-nanoscale. The APT is the only instrument capable of mapping the 3-D composition of a metal, semiconductor, or an insulator with atomic resolution. The 3-D reconstruction of this direct space information provides unprecedented capabilities for characterizing materials at the atomic level. The quantitative analysis of APT images could provide the ability to directly observe and interpret atomistic scale solid state phenomena that until now, have only been visualized via modeling and computer simulation. The unique APT images, transformed as cloud data points, provide an ideal template for developing graphtheoretic and differential geometric techniques that allow (1) assimilation of dynamic, noisy and/or incomplete data during reconstruction, (2) development of parallel algorithms for 3D reconstruction, and (3) extraction of multi-scale geometric features and statistics from the 3D data. This research will yield new computational tools and algorithms that can effectively utilize computing platforms for noise insensitive reconstruction, which can significantly improve our abilities to quantitatively analyze APT data. These advancements will be possible due to developments in uncertainty quantification, model reduction of complex systems, and scalable algorithms. Data-driven model reduction strategies that dynamically account for the noisy data during the 3D reconstruction and efficiently extract atomic coordination and radial distribution functions will be developed to directly link atomic-scale structure and chemical data with materials properties and behavior. A virtual organization (VO) will be established, targeting the rapidly growing global community of APT users and the computational sciences community that is involved in modeling and applications of very large data sets. The VO will be initially hosted at Iowa State University (ISU) and later linked to the NSF TeraGrid to ensure a sustainable and accessible cyberinfrastructure. Educational and outreach activities will include (1) interdepartmental graduate courses on computational nanosciences, (2) summer research programs on data mining and parallel computing for undergraduate students , (3) development of novel teaching materials for high school and first-year courses that demonstrate the connections between data mining and materials science --- this will be class tested through ISU's Science Bound program, which actively engages students from underrepresented groups; (4) workshops on quantitative APT that will be linked to ISU's NSF-ADVANCE program to promote participation by women faculty and other under-represented groups; and (5) leverage international collaborations based on our NSF International Materials Institute.