Efficient Implementation of Reverse Time Migration Seismic Imaging on FPGAs

Abstract

AbstractThis work presents an alternative computing platform, Field Programmable Gate Arrays (FPGAs), for one of the widely used applications in the oil and gas industry; seismic imaging using reverse time migration (RTM). As with many of the oil and gas applications, RTM involves many computations performed repeatedly over extremely large number of data points. Conventionally, such computations were implemented as a sequence of temoral steps and parallelized using multi-core CPUs or GPUs at various levels of abstractions (task, thread, and instruction-levels). With FPGAs, computations can be parallelized and implemented as spatially separated steps (i.e. the computing fabric can be configured to carry out all the computations, side-bu-side rather than as temoral steps). In this work, we illustrate the spatial computing capabilities of FPGAs using RTM. Two different implementations with different types of optimizations were developed. Performance optimizations and measurements based on experimental test bed are reported.