A publication of the Office of Advanced Simulation & Computing, NA-114, NNSA Defense Programs
December 2007
NA-ASC-500-07—Issue
Return to this issue’s stories
Algorithms Decrease Solution Time Ten-Fold
Sandia National Laboratories’ ASC Algorithms Milestone team, led by Roscoe Bartlett, developed and demonstrated a full vertical integration of numerical algorithms in the Trilinos Library with the Charon semiconductor device modeling code. Compared to prior approaches, these new algorithms have demonstrated up to a ten-fold reduction in time-to-solution while providing more accurate and robust results.
The top images compare typical Charon simulations of a semiconductor device to experimental imagery. The bottom plot shows a time history of the sensitivity of a Charon device model to an input parameter computed using the Trilinos vertically integrated algorithms and the prior point-wise approach. Trilinos algorithms return the entire time history of the sensitivity at 1/10 the computational cost and with greater accuracy.
Algorithms ranging from linear algebra, preconditioners, and iterative linear solvers, all the way up through nonlinear solvers, transient solvers, and optimization were integrated with Charon in a flexible and modular, yet high-performance and scalable manner. The Qualification Alternatives to SPR (QASPR) program is using these new capabilities in Charon to explore the sensitivity of transistor device models to input parameters and, in the future, to optimize parameters based on experimental data. The versatility of these vertically integrated Trilinos components was also highlighted by their application to the SIERRA engineering simulation suite, with the development of an intrusive optimization algorithm used to solve a prototype MEMS actuator design problem.
These new algorithmic capabilities are available in Trilinos version 8.0, released August 31, 2007, and are being deployed to help bridge the research-development-application spectrum for predictive simulation. You can learn more about Trilinos as well as download the full suite of open-source software at http://trilinos.sandia.gov/.
DOE Privacy Disclaimer | Sandia Privacy Disclaimer | SAND 2007-8167 W
Developed and maintained by Sandia National Laboratories for NA.114