Energy frugal tags in reprogrammable I-Caches for application-specific embedded processors

In this paper we present a software-directed customization methodology for minimizing the energy dissipation in the instruction cache, one of the most power consuming microarchitectural components of high-end embedded processors. We target particularly the instruction cache tag operations and show how an exceedingly small number of tag bits, if any, are needed to compute the miss/hit behavior for the most frequently executed application loops, thus minimizing the energy needed to perform the tag reads and comparisons. The proposed methodology exploits the fact that the code layout structure of the program loops can be identified after compile and link, and that it typically resides in a very confined memory location, for which very few bits from the effective address can be utilized as a tag. Subsequently, we present an efficient, programmable implementation to apply the suggested energy minimization technique. The experimental results show a significant decrease in energy dissipation for a set of real-life applications.