Protecting Big Blue from Rogue Subatomic Particles

Device technology scaling continues to deliver faster and smaller transistors, contributing to IBMs continued leadership in server systems. However, there is also a dark side to device technology scaling. As transistors shrink, the amount of charge required to change the logic state of a memory or a logic circuit (i.e. flip a 1 to a 0 and vice versa) also shrinks. This complication spurs continued effort at IBM to test, characterize, and mitigate these transient bit flips, so called soft errors. In this paper we survey our ongoing work in this realm and introduce our views on trends for soft errors in future device technologies.