<p>Time-multiplexed spinwave Ising Machines have unveiled a route towards miniaturized and low-cost combinatorial optimization solvers but are constrained in the number of spins by nonlinear spinwave dispersion. In contrast, surface acoustic waves offer an intrinsically linear dispersion and high thermal stability. Here, we demonstrate an all-to-all, fully programmable, 50-spin Ising machine using a surface acoustic wave delay line and off-the-shelf microwave components. Our device solves random 50-spin MAX-CUT problems with a single run compute time of 10 ms and a figure of merit of 55 solutions s<sup>−1</sup> W<sup>−</sup><sup>1</sup> reaching success probability of 84% for 99%-accurate solutions on 0.5-density matrices. Moreover, it demonstrates 4–5 orders of magnitude better thermal stability than optical Coherent Ising Machines while having similar scalability potential. Our results illustrate the general merits of wave-based time-multiplexed Ising machines operating in the microwave domain as compact, energy-efficient, and high-performance platforms for commercially feasible combinatorial optimization solvers.</p>

错误:搜索内容不能为空,请输入英文关键词
错误:关键词超出字数限制,请精简
高级检索

A 50-spin surface acoustic wave Ising machine

  • Artem Litvinenko,
  • Roman Khymyn,
  • Roman Ovcharov,
  • Johan Åkerman

摘要

Time-multiplexed spinwave Ising Machines have unveiled a route towards miniaturized and low-cost combinatorial optimization solvers but are constrained in the number of spins by nonlinear spinwave dispersion. In contrast, surface acoustic waves offer an intrinsically linear dispersion and high thermal stability. Here, we demonstrate an all-to-all, fully programmable, 50-spin Ising machine using a surface acoustic wave delay line and off-the-shelf microwave components. Our device solves random 50-spin MAX-CUT problems with a single run compute time of 10 ms and a figure of merit of 55 solutions s−1 W1 reaching success probability of 84% for 99%-accurate solutions on 0.5-density matrices. Moreover, it demonstrates 4–5 orders of magnitude better thermal stability than optical Coherent Ising Machines while having similar scalability potential. Our results illustrate the general merits of wave-based time-multiplexed Ising machines operating in the microwave domain as compact, energy-efficient, and high-performance platforms for commercially feasible combinatorial optimization solvers.