Fundamental energy cost of finite-time computing

TL;DR

This paper analyzes the fundamental energy costs of finite-time irreversible computing, revealing that parallel computers can approach the Landauer limit while serial ones diverge, informing energy-efficient computer design.

Contribution

It provides the first thermodynamic analysis of finite-time computing costs, comparing serial and parallel architectures within nonequilibrium thermodynamics.

Findings

Parallel computing can operate near the Landauer limit in finite time.

Serial computing's energy cost diverges with problem size.

Implications for designing energy-efficient computing systems.

Abstract

The fundamental energy cost of irreversible computing is given by the Landauer bound of $kT \ln2$~/bit. However, this limit is only achievable for infinite-time processes. We here determine the fundamental energy cost of finite-time irreversible computing \er{within the framework of nonequilibrium thermodynamics}. Comparing the lower bounds of energy required by ideal serial and parallel computers to solve a problem of a given size in a given finite time, we find that the energy cost of a serial computer fundamentally diverges with increasing problem size, whereas that of a parallel computer can stay close to the Landauer limit. We discuss the implications of this result in the context of current technology, and for different degrees of parallelization and amounts of overhead. Our findings provide a physical basis for the design of energy efficient computers.