A GPU/CUDA implementation of the collection-diffusion model to compute SER of large area and complex circuits

This work reports the CUDA implementation of the collection-diffusion model to compute the soft-error rate (SER) of large area and/or complex circuits on graphics processing units (GPU). We detail the time parallelization introduced in the algorithm to accelerate by one order of magnitude the SER calculation. Code performances are evaluated on a NVIDIA Tesla C1060 GPU card for the calculation of the SER of a 65 nm SRAM circuit subjected to an alpha-particle source irradiation.