Configurable computing: the catalyst for high-performance architectures

Abstract

Recent trends in the cost and performance of application-specific hardware relative to conventional processors discourage investing much time and energy in special-purpose architectures except for niche applications. These trends, however, may be reversed by the increasing complexity of computer architectures and the advent of configurable computing. Configurable computers have attracted considerable attention recently because they promise to deliver the performance of application-specific hardware along with the flexibility of general-purpose computers. In this paper, we discuss some of the forces driving configurable computing, and we argue that new configurable architectures are needed to realize the enormous potential of configurable computing. In particular, we believe that the commercial FPGAs currently used to construct configurable computers are too fine-grained to achieve good cost-performance on computationally-intensive applications that demand high-performance hardware. We then describe a new architecture called RaPiD (Reconfigurable Pipelined Datapaths), which is optimized for highly repetitive, computationally-intensive tasks. Very deep application-specific computation pipelines can be configured in RaPiD that deliver very high performance for a wide range of applications. RaPiD achieves this using a coarse-grained reconfigurable architecture that mixes the appropriate amount of static configuration with dynamic control.