Acceleration of Complex Algorithms on a Fast Reconfigurable Embedded System on Spartan-3

Abstract

Complex algorithms usually require several computation stages. Many embedded microprocessors have not enough computational performance to resolve these algorithms in a reasonable time, so dedicated coprocessors accelerate them although the main drawback is the area devoted to them. A reconfigurable coprocessor can drastically reduce the area, since it accommodates a set of coprocessors whose execution is multiplexed on time, although the reconfiguration speed reduces the overall system performance. Although self-reconfigurable systems are possible on Spartan-3 FPGAs, it requires a hard design task due to the lack of software and hardware support available on higher-cost families. This paper describes the architecture of a fast self-reconfigurable embedded system mapped on Spartan-3, used as computation platform to solve a complex algorithm, such as the image-processing carried out in a fingerprint biometric algorithm. In order to reduce the reconfiguration time, the system uses our custom-made memory and reconfiguration controllers. Moreover, the dynamic coprocessor can access directly to external memory through our memory controller to improve processing time.