Implementing the Asynchronous Reversible Computing Paradigm in Josephson Junction Circuits.

Abstract

An important figure of merit for practical computing systems is energy-delay efficiency, which characterizes the number of operations a given device can perform per unit energy dissipated, per unit time taken for the operation.This efficiency metric appears to be limited in existing superconducting logic styles, motivating the search for novel, potentially more efficient forms of superconducting logic.One path to improving efficiency might be if we could remove the need for external control driven by slow clocks that is present in existing adiabatic SCE logic styles such as AQFP/RQFP.The Asynchronous Ballistic Reversible Computing (ABRC) paradigm offers an alternative model for ballistic, data-driven reversible computation.In an ongoing project at Sandia, we are attempting to demonstrate how the ABRC model can be implemented using Josephson junction circuits, while minimizing the energy-delay product.In this talk, we review the progress that has been made to date in this effort.An interesting direction for future research is to design physical mechanisms that explicitly invoke quantum processes exhibiting superadiabaticity (a.k.a."shortcuts to adiabaticity") to improve energy-delay efficiency in concrete computing technologies.We briefly review what is known about the potential applicability and limits of this approach.Dissipation-delay Efficiency (DdE) Sandia National Laboratories • The key motivating Figure of Merit (FOM) in the present study.• For a single primitive transition of the digital state of a system between two distinct informational states, consider:• The energy dissipation D incurred by that transition process.Relates to costs associated with supply of energy and cooling.• The delay d, defined as the time interval from start to end of process.Relates to costs associated with achieving a given level of parallel performance.• Then define the dissipation-delay product DdP = D • d.• Note that since refers specifically to energy dissipation, not to energy invested in the signal, in reversible processes, it is not subject to the "quantum speed limit" (QSL) lower bound of -h!No fundamental lower bound to DDP is yet known!-In fact, it would be identically zero for any perfectly-known unitary time-evolution.• Of even more general interest than DdP per se is dissipation as a function of delay, D(d), considered over a range of practical (tolerable) delay values... We'd like to extend the pareto frontier of this function within the useful range.• Dissipation-delay efficiency (DdE) of a computing technology just refers to the reciprocal of DdP, r/Dd = (Dd)-i.