Power Performance of Instruction Pre-Fetch Unit

명령어 선 인출기의 전력 성능

In this paper, we investigate the effect of adopting branch-penalty compensation schemes on the power performance of TLBs(Translation Look-aside Buffers) and instruction caches. We found that the double-buffer branch-penalty compensation scheme can reduce the power consumption of the TLBs and the instruction caches considered by up to 14-21.3%. The power consumption is estimated through simulation at the architectural level, using the Kamble/Ghose method