SASIAF, A Scalable Accelerator for Seismic Imaging on Amazon AWS FPGAs

Oil and gas companies rely on seismic imaging as a tool for drilling prediction, making crucial decisions and evaluating financial parameters of a hydrocarbon field. Reverse Time Migration (RTM) is the most advanced technique in seismic imaging. RTM is a computationally intensive task and has one of the longest execution times in oil and gas field because of its nested loops and high amount of raw data that comes to it. Instead of many research works that have been done on accelerating it on a FPGA and cluster of GPUs, due to high cost of FPGA clusters there are not many accelerators for them. In this paper, we have used Amazon AWS F1 instances and accelerated RTM on 8 FPGAs using time partitioning method and a new architecture called SASIAF. We have shown that we can reach up to 960x speed up comparing with running it sequentially on CPU. We have also compared our work with state-of-the-art FPGA implementation.