Ostapkevich M.B. Aillet A. Gougeon A. Lorenz R.
An implementation of accelerator for cellular automata models using FPGA technology
Reporter: Ostapkevich M.B.
The description of a matrix coprocessor for cellular automata (CA) models is given in the report. The coprocessor supports classical cellular automata and its certain extensions. CA models make it possible to simulate some natural processes even when their traditional numerical computation methods fail. Such models have a number of advantages in comparison to numerical models, e.g. they do not have problems related to precision and error accumulation. But on the other hand, nearly any real world CA model requires an enormous amount of computations and big size of model data. One of the ways to attain a descent simulation time is to use a massively parallel and pipe-lined implementation of CA simulation using FPGA technology, which is widely applied to the computations for the last two decades. The technology permits to attain a very high performance solution due to the massively parallel fine-grain nature of FPGAs. It is also worthy to note that a typical modern cluster node often includes one or more FPGAs. Thus, hybrid MPI-FPGA implementations can be constructed. The first part of the report is dedicated to a simple scalar processor, which has a modular and extensible architecture and serves as the basis for the construction of the matrix coprocessor. It is followed by the description of architecture and organization of the matrix coprocessor as well as a number of CA models built using this coprocessor.
The work has been partially supported by project contract 14-07-00381.
To reports list