Xilinx 03 (2009-13)

Title: High performance computing with FPGAs

Research Engineer: Thomas Perry
Sponsor: Xilinx
Academic Supervision: Dr Khaled Benkrid and Dr Mark Parsons, University of Edinburgh

Xilinx is the leading supplier of Field Programmable Gate Arrays (FPGAs), which are employed in growing range of applications, from embedded control through to network processing.

More recently, FPGAs have been applied to high performance computing problems and are proving successful for a number of reasons. Lower power consumption addresses the growing cost of powering and cooling the massive compute farms employed by the financial industry, which is set to exceed the cost of hardware in the next few years. Additionally, bit-level configurability of the devices allows the implementation to be better matched to an application, e.g., LUT-based architecture is well matched to the 2-bit operations found in DNA sequencing.

While the benefits are clear, one of the challenges of using FPGAs and an adoption barriers being addressed by Xilinx, involves ensuring hardware platforms and programming environments are as flexible and convenient as conventional computing technologies. The fundamental issues of programming highly parallel systems, something that has always been a part of FPGA design, is now central to multi-core processors, and the solution relevant to both technologies.

The project aim is to apply these emerging technologies to commercial applications. The research will include:

1. Investigation of algorithms and architectures to address the specified application(s)
2. Development of these algorithms on FPGA using the latest tools and platforms
3. Development of a methodology for rapidly creating programs within the specified domain of application
4. Evaluation of that methodology

The contribution of this research to the FPGA and HPC communities will be to demonstrate that FPGAs can be programmed in real-world applications using evolutionary, rather than revolutionary techniques.