Implementation of numerical methods for nanoscaled semiconductor device simulation using OpenCL

The present work implements solvers with OpenCL of the FGMRES and preconditioned BCGSTAB algorithms. These solvers are integrated in a 3-D simulation tool of nanoscaled MOSFET transistors. Simulations are launched in two different platform devices: NVIDIA Tesla S2050 and Intel Xeon Phi 3120A. The resulting times of execution are compared against the optimized PSPARSLIB version of the FGMRES solver. The results showed that a large computational charge compensate the overhead due to the data transfer time.

keywords: