Femtojoule-per-operation photonic computer for the subset sum problem

TL;DR

This paper demonstrates a photonic computer that efficiently solves the subset sum problem with extremely low energy per operation, significantly outperforming traditional supercomputers in energy consumption for complex tasks.

Contribution

The authors experimentally develop a photonic computer utilizing photon energy levels and time-of-flight storage to solve the subset sum problem with unprecedented energy efficiency.

Findings

Energy per operation ≤ 10^(-15) J at N=33

Consumes 10^8 times less energy than supercomputers for medium-scale problems

Enhanced energy savings in real-life iterative SSP computations

Abstract

Energy-efficient computing is becoming increasingly important in the information era. However, electronic computers with von Neumann architecture can hardly meet the challenge due to the inevitable energy-intensive data movement, especially when tackling computationally hard problems or complicated tasks. Here, we experimentally demonstrate an energy-efficient photonic computer that solves intractable subset sum problem (SSP) by making use of the extremely low energy level of photons (~10^(-19) J) and a time-of-flight storage technique. We show that the energy consumption of the photonic computer maintains no larger than 10^(-15) J per operation at a reasonably large problem size N=33, and it consumes 10^(8) times less energy than the most energy-efficient supercomputer for a medium-scale problem. In addition, when the photonic computer is applied to deal with real-life problems that involves iterative computation of the SSP, the photonic advantage in energy consumption is further enhanced and massive energy can be saved. Our results indicate the superior competitiveness of the photonic computer in the energy costs of complex computation, opening a possible path to green computing.