Design Space Analysis for the Acoustic Wave Equation Implementation on FPGA Circuits

Seismic modeling and data processing operates on gigabyte to terabyte-size datasets and involves large CPU clusters running applications for extended periods of time. Accelerating these applications would make a big impact and the emerging new co-processor hardware architectures need to be carefully tested for this purpose. One of these new architectures is based on the reconfigurable circuit field-programmable gate arrays (FPGA). These circuits offer potentially large speedups of applications, but to succeed, the FPGA implementation of algorithms need to be carefully evaluated to discover the tradeoffs in compute capability, memory capacity and memory bandwidth. In many cases, taking the existing software implementation of an algorithm and mapping it to reconfigurable circuits does not result in the fastest implementation on FPGA. Careful evaluation of the various implementation strategies that we call design space analysis serves as an important tool to successfully implement an algorithm on FPGAs. In this abstract we provide a short design space analysis for a 3D acoustic forward modeling algorithm.