Title :
PHAT: A technology for prototyping parallel heterogeneous architectures
Author :
Wink, Thorsten ; Koch, Andreas
Author_Institution :
Embedded Syst. & Applic. Group, Tech. Univ. Darmstadt, Darmstadt, Germany
Abstract :
This paper presents the Parallel Heterogeneous Architecture Technology (PHAT), a scalable design methodology for prototyping and evaluating heterogeneous arrays of software-programmable VLIW processors and both manually designed and automatically-compiled custom hardware accelerators, using a shared memory architecture for communication. We discuss the trade-offs and breakeven point for switching from bus-based to network-on-chip interconnects, the interface and protocols for connecting distributed on-chip caches and multi-bank out-of-order off-chip-memories, as well as the impact of floorplanning on the quality of results for implementation on Xilinx Virtex 6 LX 760 devices. The capabilities are evaluated at the system-level on the multi-FPGA Convey HC-1ex hybrid-core computer, accessing its high-performance memory system, and integrating r-VEX processor cores with IP blocks for SHA and FFT computations.
Keywords :
cache storage; field programmable gate arrays; integrated circuit design; integrated circuit interconnections; integrated circuit layout; memory architecture; microprocessor chips; network-on-chip; parallel architectures; shared memory systems; FFT computation; IP blocks; PHAT; SHA computation; Xilinx Virtex 6 LX 760 devices; automatically-compiled custom hardware accelerators; bus-based interconnects; distributed on-chip caches; floorplanning impact; heterogeneous software-programmable VLIW processor array evaluation; heterogeneous software-programmable VLIW processor array prototyping; multi FPGA Convey HC-1ex hybrid-core computer; multibank out-of-order off-chip-memories; network-on-chip interconnects; parallel heterogeneous architecture prototyping; r-VEX processor cores; scalable design methodology; shared memory architecture; Aerospace electronics; Computer architecture; Field programmable gate arrays; Hardware; IP networks; Program processors;
Conference_Titel :
Design and Architectures for Signal and Image Processing (DASIP), 2014 Conference on
DOI :
10.1109/DASIP.2014.7115627