Computing with Field-Programmable Gate Arrays

Oskar Mencer, Bell Labs, Imperial College UK
Wayne Luk, Imperial College UK

1/2 Day

Field Programmable Gate Arrays (FPGAs) offer the potential of orders of
magnitude higher performance than a microprocessor for specific domains of
applications. The key to achieving this potential is to move the level of
abstraction of programming FPGAs closer to the level of abstraction of
programming microprocessors.

Topics:

- Applications View:
  Static and dynamic approaches to acceleration of applications in areas
  such as multimedia, graphics, voice and image processing, encryption,
  compression, coding, VLSI CAD and optimization in general (e.g. genetic
  algorithms).

- FPGA Technology-the basic building blocks available within FPGAs. This
  includes FPGA micro architecture, support for reconfiguration. Furthermore,
  the basic building blocks are combined in arithmetic units and module
  generation libraries which form a scalable "instruction set" for
  reconfigurable computing.

- Architecture Generation, including processor-like architectures, stream
  architectures, memory-like architectures, and, pros and cons of bit-level
  architectures versus coarse-grain parallel architectures.

- Compilation technology for FPGAs: State-of-the-Art methods for compiling
  to FPGAs including a description of approaches taken by several
  industrial and academic examples. We compare the hardware designers'
  approach to procedural, object-oriented and functional approaches to the FPGA
  programming problem.