Energy-Efficient Superconducting Computing—Power Budgets and Requirements

TLDR

Large‑scale computing systems face a growing power and energy challenge, and superconducting computing—especially recent single‑flux‑quantum circuit advances—offers a promising alternative. The study explores superconducting system concepts targeting 1–1000 PFLOP/s, limited to existing commercial cryogenic refrigerators and Nb technology, and identifies the need for cryogenic cache and main memory. Results indicate that superconducting computing could be competitive in power and energy efficiency if key component technologies meet specific goals, while highlighting both its advantages and remaining development needs.

Abstract

Large-scale computing system characteristics vary by application class, but power and energy use has become a major problem for all classes. Superconducting computing may be able to serve the needs of these systems significantly better than conventional technology. Recent developments in single flux quantum circuit technology for digital logic include variants with greatly improved energy efficiency. Concepts were investigated for computing systems capable of performance in the range from 1 to 1000 PFLOP/s. The concept systems were constrained to use existing commercial cryogenic refrigerators and Nb superconducting technology. In order to meet the performance goals, cache and main memory capable of operating at cryogenic temperatures will be required. Superconducting computing is shown to be potentially competitive on the basis of power and energy efficiency if key component technologies can meet specific goals. Potential advantages of superconducting computing are identified as well as areas requiring further development.