Reliability in the Dark Silicon Era

Information technology is now an indispensable pillar of a modern-day society, thanks to the proliferation of digital platforms in the past several decades. The demand on robust and economical data processing and storage, however, is growing faster than technology can sustain. Moreover, while forecasts indicate that chip density scaling will continue for another decade, the diminishing returns in supply voltage scaling preclude activating all on-chip transistors simultaneously, leading designs towards energy-centric solutions on "dark silicon". In this talk, I will describe this paradigm shift towards energy-centric server designs and its implications on fault tolerance and reliability.