Configuration Memory Management for Adaptive Computing Systems


Adaptive computing systems harness the power of FPGAs to provide reconfigurable computing resources for many domains. While these systems currently provide high-performance implementations for automatic target recognition, neural nets, image processing, genetic algorithms, biological database searches, and many other applications, their significant reconfiguration delays limit the potential speedups. In this talk we present new techniques for configuration memory management in ACS systems which can radically reduce these reconfiguration overheads. These techniques include the prefetching of FPGA configurations, which can overlap computation with reconfiguration, as well as configuration compression, which can greatly reduce the bandwidth requirements of these systems. These techniques have the potential to greatly increase the range of applications accelerated by Adaptive Computing.

©2002-2018 U.C. Regents