SUPERCOMPUTING THE ELECTRON TRANSPORT USING CUDA TECHNOLOGY

Statistical algorithms are presented for modeling the interaction processes between electrons and matter. A software implementation has been developed for hybrid supercomputers making use of NVIDIA (c) CUDA (c) technology. Standard Monte Carlo schemes are modified for effectively exploiting the parallel computing capabilities of graphical processors. Modeling inter-action processes of electrons with objects of a complex non-uniformly scaled structure requires developing fine-grained models of the interaction without using well-known approximations. The model of individual collisions ( MIC) is worked out to describe the interaction of electrons with atoms. MIC does not include the approximations assumed in multiple collision theory or the continuous slow down approach. The distributions of electron characteristics are obtained from tabulated electron cross-section data. Modeling hundreds of thousands of electron collisions with small energy transfers within the scope of MIC requires a huge amount of calculations. Therefore, modern computers based on hybrid architecture are required for computing electron transport using MIC. A corresponding parallelization technique is developed for using these hybrid computers. The discussed examples demonstrate the applicability of the algorithms to investigating the interaction of electrons with X-ray tube targets producing Bremsstrahlung and to researching the process of electron emission from surfaces of objects being under radiation.