Register File Reliability Analysis Through Cycle-Accurate Thermal Emulation
Continuous transistor scaling due to improvements in CMOS devices and manufacturing technologies is increasing processor power densities and temperatures; thus, creating challenges when trying to maintain manufacturing yield rates and devices which will be reliable throughout their lifetime. New microarchitectures require new reliability-aware design methods that can face these challenges without significantly increasing cost and performance. In this paper we present a complete analysis of reliability for the register file architecture of the Leon 3 processor. The analysis conducted is supported by the use of an accurate HW/SW FPGA-based emulation platform that enables a complete design space exploration of thermal and reliability metrics during the execution of an extended set of benchmarks, in a very limited amount of time. The effect of various compiler optimizations and register assignments on the reliability of the register file is then analyzed. Our results quantify the respective effects of these different factors and enable us to design a reliability-aware register file assignment policy that consistently improves the Mean-Time-To-Failure figure (20% on average) for the various types of applications.
IWIA2008-FC45Fd01.pdf
openaccess
294.94 KB
Adobe PDF
546b60234511f22d995462691e388d02