Speaker: Norman O. Birge, Michigan State University
Abstract: Large-scale computing facilities and data centers are using electrical power at an ever increasing rate. Projections suggest that a future “exoscale” computer will require the power output of a typical nuclear power plant – clearly an untenable situation. One approach to addressing this problem is to build a computer out of all superconducting elements, which dissipate very little power. Such a computer would have to be cooled to cryogenic temperatures, so it must be extremely energy-efficient to justify the added complexity and cost associated with cooling.
Superconducting logic circuits based on manipulating single flux quanta have existed for 30 years; what has been missing is a high-density, fast, and energy-efficient cryogenic memory. One approach is to use Josephson junctions containing ferromagnetic (F) materials as the memory element for such a memory. The basic building block is a Josephson junction containing two ferromagnetic layers whose magnetization directions can be switched between being parallel or antiparallel to each other, as in a conventional spin valve. We have demonstrated successful switching of such a junction between the “0” phase state and the “π” phase state, from measurements of two junctions in a SQUID geometry. An alternative approach is to use a Josephson junction containing three ferromagnetic layers, which is designed to carry spin-triplet supercurrent. We have also realized controllable 0 - π switching in such a spin-triplet junction. At the end of the talk I’ll mention what needs to be done to turn these results into a real superconducting computer.