The Technology Lane on the Road to a Zettaflops

Given that viable petaflops-level machines are now within sight, the time has come to begin considering how we can reach the next performance milestones. This work spells out the upper limits of performance that currently understood technology might achieve. Our “design target” is a zettaflops or 1021 flops). We explore whether or not a classical approach has any chance of achieving this target of one million petaflops, and find that area and power might very well be insurmountable obstacles even in the absolute best case. We also explore whether or not non-silicon, but still classical, nano-scale devices might move us any closer. In a best case scenario, while significant improvements over silicon are at least possible, a zettaflops still appears improbably ambitious. Finally, we explore the capabilities of reversible logic as it may present one of the few plausible inroads into the monumental power problem that appears before us. Based on these studies, we suggest a forward path for future technology development, and couple it with needed work in architecture and algorithms. 1. http://www.highproductivity.org/