High-Level Synthesis Methodologies for Delay-Area Optimized Coarse-Grained Reconfigurable Coprocessor Architectures

As Very Large Scale Integration (VLSI) process technology continues to scale down transistor sizes, modern computing devices are becoming extremely complex. In order to face this complexity explosion, the shifting of design methodologies towards higher level of abstraction has been proposed. This high level view of the design procedure enables the automated synthesis of applications´ architecture that is written in an application-level description i.e. C/C++. Additionally, it allows designers to explore the tradeoffs between different system and implementation parameters to conclude in an efficient design solution. The work done during this PhD thesis targets the exploration and optimization of the design solutions in a global manner, by focusing on the combined development of novel (i) system-level automated design methodologies/tools and (ii) circuit-level techniques for a specific class of system architectures - reconfigurable systems. Reconfigurable Computing has been proposed as a new paradigm to address the conflicting design requirements for high performance and area efficiency. Towards this direction, fine- and coarse-grained reconfigurable coprocessor architectures have been presented. Unlike fine-grained, coarse-grained architectures (CGA) operate at the word level of granularity exhibiting better power and performance features, close to ASIC solutions. However, a performance-area-power gap still exists for CGAs to overcome ASIC implementations. Thus, new fundamental design problems/questions has been raised. Does this gap be a bridgeable one? How can CGAs shift even closer to ASIC datapaths? In order to address the aforementioned problems, we identified that hardware sharing at the bit-level generates CGAs with performance and area characteristics closer to ASICs than the existing ones. Thus, this thesis proposes new architectural templates and the corresponding high level synthesis methodologies to enable a new shifting on the state-of-the-art of CGAs.