研究目的
To propose an energy-efficient VLSI array architecture of a coarse-grained Recon?gurable Processing Unit (RPU) targeting computation-intensive multiple-standard video decoding; proposing a Line-Switched Mesh Connect (LSMC) routing scheme which reduced the chip area signi?cantly; introducing a Hierarchical Con?guration Context (HCC) organization scheme which reduced the implementation overhead and the energy dissipation resulted from fast recon?guration of the PEs.
研究成果
The proposed RPU architecture achieves significant improvements in performance and energy efficiency for multiple-standard video decoding applications. It also demonstrates applicability for other computation-intensive applications, though with limitations for control-intensive tasks.
研究不足
The architecture is not a perfect solution for all general-purpose computation tasks. For certain kernel algorithms, the proposed architecture could not gain any performance boosts compared with general-purpose CPUs. The performance advantages do not hold for other high-end general-purpose CPUs.