Bozidar Novakovic

Time-resolved QTBM computational efficiency

B. Novakovic, H. Sahasrabudhe, and G. Klimeck May 2016

Objective:

  • Investigate the efficiency of time-integration in the semi-empirical TB basis.

Approach:

  • Explicit (matrix-vector product) vs. implicit (direct linear solver) methods.
  • Traditional CPU vs. GPU accelerators.
  • Measure time-integration only (matrix preparation/calculation not included).

Results:

  • Implicit methods faster for larger time steps and lower cross sections.
  • Explicit methods better scale and leverage accelerators.
  • Explicit methods conditionally stable.

Impact:

  • Current and future results aim to make realistic simulations feasible.

Impl. Crank-Nicholson (CN), Expl. Askar-Cakmak (AC, leapfrog), Expl. 3rd order Adams-Bashforth (AB3)

Bozidar Novakovic's Pages

Biography Time-resolved device characteristics in the ballistic limit Computational efficiency of the time-resolved simulations in the ballistic limit Efficient time-resolved NEGF based quantum transport algorithm Modeling novel steep subthreshold devices for reduced power consumption and short channel effects Fast convergence algorithms for Poisson and transport equations Accurate I-V curves of Si MOSFET and III-V TFET devices Optimal quadrature techniques for device periodic directions. Optimized energy mesh