High-Level Resource Binding and Allocation for Power and Performance Optimization

Field-programmable Gate Arrays (FPGA);behavioral synthesis;glitch power;power reduction;high-level synthesis;process variation;low power

Cromar, Scott A. ; Chen ; Deming

关键词: Field-programmable Gate Arrays (FPGA);    behavioral synthesis;    glitch power;    power reduction;    high-level synthesis;    process variation;    low power;   

Others  :  https://www.ideals.illinois.edu/bitstream/handle/2142/11971/cromar_scott.pdf?sequence=2&isAllowed=y

美国|英语

来源: The Illinois Digital Environment for Access to Learning and Scholarship

【 摘 要 】

While technology scaling has presented many new and exciting opportunities, new design challenges have arisen. Smaller feature sizes have led to increased density and large variations in the delay and power characteristics of on-chip devices. Additionally, with the increasing desirability of low-power chips, decreasing power consumption has become a significant priority. Major sources of dynamic power consumption in modern chips include glitches (i.e., spurious signal transitions), the reduction of which are challenges to circuit designers. High-level synthesis has been touted as a solution to these problems, as it can both significantly reduce the number of man hours required for a circuit design, and offer greater opportunities for optimization of design goals, by raising the level of abstraction. In this thesis, we present two resource binding and allocation algorithms that take advantage of the optimization opportunities available at the higher level of abstraction.The first is a new variation-aware high-level synthesis binding and module selection algorithm, named FastYield, which takes into consideration multiplexers, functional units, registers, and interconnects. FastYield connects with the lower levels of the design hierarchy through its inclusion of a timing-driven floorplanner guided by a statistical static timing analysis engine which is used to modify and enhance the synthesis solution. FastYield is able to incorporate spatial correlations of process variations in its optimization, which are shown to affect performance yield. FastYield is shown to achieve a significant reduction in clock period, and significant gain in performance yield, when compared to a variation-unaware and layout-unaware algorithm.The second is a glitch-aware, high-level binding algorithm for power, area, and multiplexer reduction targeting field programmable gate arrays (FPGAs), called HLPower. HLPower employs a glitch-aware dynamic power estimation technique derived from an FPGA technology mapper. High-level binding results are converted to VHSIC hardware description language (VHDL), and synthesized with Altera’s Quartus II software, targeting the Cyclone II FPGA architecture. Power characteristics are evaluated with the Altera PowerPlay Power Analyzer. The binding results of HLPower are compared to LOPASS, a state-of-the-art low-power high-level synthesis algorithm for FPGAs. Experimental results show that HLPower significantly reduces toggle rate and area, resulting in a large decrease in dynamic power consumption.