Computational architectures that are optimized to solve non-deterministic polynomial-time hard or complete problems are of use in the development of machine learning, logistical planning and pathfinding. A range of quantum-, optical- and spintronic-based approaches have been explored for solving such combinatorial optimization problems, but they remain complicated to build and to scale. Here we report a scalable ring-oscillator-based integrated circuit for optimization problem solving. Our 1,968-node King’s graph ring oscillator array has five levels of coupling strengths and can achieve up to 95% accuracy for randomly generated combinatorial optimization problems. The measured average power consumption of the Ising chip is 0.042 W and it takes less than 50 oscillation cycles to resolve to the ground state. Our device is resilient to environmental and variation effects. By using a multi-phase phase measurement circuit, we also capture the true phase behaviour within a coupled-oscillator integrated circuit.

经过优化以解决非确定性多项式时间难题或完全问题的计算架构可用于机器学习、逻辑规划和寻路的开发。已经探索了一系列基于量子、光学和自旋电子学的方法来解决此类组合优化问题，但它们的构建和扩展仍然很复杂。在这里，我们报告了一种可扩展的基于环形振荡器的集成电路，用于优化问题解决。我们的 1,968 节点 King 图形环形振荡器阵列具有五个级别的耦合强度，对于随机生成的组合优化问题，可以达到高达 95% 的准确度。Ising 芯片测得的平均功耗为 0.042 W，只需不到 50 个振荡周期即可解析到基态。我们的设备对环境和变化影响具有弹性。通过使用多相相位测量电路，我们还可以捕获耦合振荡器集成电路内的真实相位行为。