Efficient two-level parallelization approach to evaluate spin relaxation in a strained thin silicon film

Abstract

The evaluation of the spin lifetime in an ultra-thin silicon film is a massive computational challenge because of the necessity of performing appropriate double integration of the strongly scattering momentum-dependent spin relaxation rates. We have tackled the problem by dividing the whole computation range into two levels. Our scheme in each level is based on a hybrid parallelization approach, using the message passing interface MPI and OpenMP. In the first level, the algorithm precalculates the subband wave functions corresponding to fixed energies and archives the results in a file-based cache to reduce memory consumption. In the second level, we compute the spin relaxation time by using the archived data in parallel. This two-level computation approach shows an excellent parallel speedup, and most efficient ways to maximally utilize the computational resources are described. Finally, how an application of shear strain can dramatically increase the spin lifetime is shown.