NSF X-Labs Initiative Topic 5 - Computation at the Limit of Physics
Apply to Internal Competition // Limit: 2 // Tickets Available: 2
Limiting Language
An eligible organization can submit a maximum of two Written Proposals per Topic Announcement for Phase 0 as a lead organization. Senior/Key Personnel may be listed on a maximum of one Written Proposal per Topic Announcement.
Program Description
The Landauer Limit has been posited as a thermodynamic floor on computation, the minimum energy dissipated when irreversibly erasing one bit. Today’s most efficient processors operate at 1-10 × 10-15 joules per switching event, five to six orders of magnitude above that limit. Furthermore, the cost per transistor at leading-edge nodes has been increasing, demonstrating that the economic benefits of silicon density scaling are diminishing. New technologies and paradigms are needed to pursue the computational efficiencies and capabilities required to drive innovation over the next half-century.
NSF X-Labs Teams in this Topic will dramatically narrow the gap between the energy efficiency of conventional Complementary Metal-Oxide-Semiconductor (CMOS)-based computing and the inherent physical limits. Dramatic improvements in computational efficiency might be gained through alternative device technologies, logic frameworks, memory technologies, and other innovations. For example, device level technologies could include sub-threshold and adiabatic CMOS and/or superconducting microelectronics that can significantly reduce energy per operation. Superconducting microelectronics could be particularly compelling, enabling clock frequencies 10X faster than conventional CMOS while reducing switching energy to near the Landauer Limit. Reversible logic frameworks may be a complementary pathway to additional efficiency gains by avoiding information erasure entirely, theoretically enabling operation below the Landauer Limit. In addition to efficiency gains arising from new forms of switches and logic, gains in the efficiency of memories, data movement and related technologies are also of interest.
An NSF X-Lab Mission in this Topic will target platform technologies that function as one or more components of a system and/or an integrated platform. Successful NSF X-Labs will demonstrate platform technologies that achieve 4-5 orders-of-magnitude improvements in effective computational throughput per unit power compared to conventional systems — measured at the system level, not solely at the switching device, memory cell or gate level. Independent of the technology proposed, teams should specify a target computational efficiency metric in their proposal and justify how their platform technology will meet it and how it compares to any known or posited physical limits. They must also present a credible and quantitative argument that their proposed technology could be integrated with complementary technologies to yield the targeted net system energy efficiency gain, after explicitly accounting for overheads associated with data movement, memory access, synchronization, required cooling and control infrastructure, etc.
Examples of challenges not considered in scope for this Topic include development of technologies where the impact is narrow and not widely deployable, fundamental research without potential for application in platform technologies, incremental advancement of the stateof-the-art, and advancement of technologies that are already appropriately developed or supported by existing R&D investments by the U.S. public or private sectors.
Full sponsor guidelines are linked here.