Reducing power density through activity migration

Power dissipation is unevenly distributed in modern microproces-sors leading to localized hot spots with significantly greater die temperature than surrounding cooler regions. Excessive junction temperature reduces reliability and can lead to catastrophic failure. We examine the use of activity migration which reduces peak junc-tion temperature by moving computation between multiple repli-cated units. Using a thermal model that includes the temperature dependence of leakage power, we show that sustainable power dis-sipation can be increased by nearly a factor of two for a given junc-tion temperature limit. Alternatively, peak die temperature can be reduced by 12.4 oC at the same clock frequency. The model pre-dicts that migration intervals of around 20–200 s are required to achieve the maximum sustainable power increase. We evaluate sev-eral different forms of replication and migration policy control