TY - GEN
T1 - Tracking nanostructural evolution in alloys
T2 - 37th International Conference on Parallel Processing, ICPP 2008
AU - Seal, Sudip
AU - Moody, Michael
AU - Ceguerra, Anna
AU - Ringer, Simon
AU - Rajan, Krishna
AU - Aluru, Srinivas
PY - 2008
Y1 - 2008
N2 - The advent of Local Electrode Atom Probe (LEAP) tomography is revolutionizing materials science by enabling near atomic scale imaging of materials. Analysis of threedimensional atom probe tomography (APT) data holds the promise of relating combinatorial arrangement of atoms to material properties and enable better design and synthesis of complex materials. Existing techniques, which are serial and require O(n2) work for n atoms, do not scale to the hundred million large data sets produced by current generation atom probe microscopes. In this paper, we present an O(n) work autocorrelation based technique that reveals clustering of constituent atoms and spatial associations between them. We present an efficient parallelization of this method and show scaling on a 1,024 node Blue Gene/L. To our knowledge, this is the first parallel algorithm for the analysis of APT data, and together with our linear work autocorrelation technique, is demonstrated to easily scale to billion atom data sets expected in the very near future.
AB - The advent of Local Electrode Atom Probe (LEAP) tomography is revolutionizing materials science by enabling near atomic scale imaging of materials. Analysis of threedimensional atom probe tomography (APT) data holds the promise of relating combinatorial arrangement of atoms to material properties and enable better design and synthesis of complex materials. Existing techniques, which are serial and require O(n2) work for n atoms, do not scale to the hundred million large data sets produced by current generation atom probe microscopes. In this paper, we present an O(n) work autocorrelation based technique that reveals clustering of constituent atoms and spatial associations between them. We present an efficient parallelization of this method and show scaling on a 1,024 node Blue Gene/L. To our knowledge, this is the first parallel algorithm for the analysis of APT data, and together with our linear work autocorrelation technique, is demonstrated to easily scale to billion atom data sets expected in the very near future.
UR - http://www.scopus.com/inward/record.url?scp=55849118129&partnerID=8YFLogxK
U2 - 10.1109/ICPP.2008.73
DO - 10.1109/ICPP.2008.73
M3 - Conference contribution
AN - SCOPUS:55849118129
SN - 9780769533742
T3 - Proceedings of the International Conference on Parallel Processing
SP - 338
EP - 345
BT - Proceedings - 37th International Conference on Parallel Processing, ICPP 2008
Y2 - 9 September 2008 through 12 September 2008
ER -