Controlled analog Ising machine: An accelerated and accuracy-enhanced intelligent solver for edge computing

Dear Editor,

The proliferation of edge artificial intelligence, intelligent manufacturing, vehicular networks, and smart grids has created an urgent demand for hardware-efficient solvers capable of addressing large-scale combinatorial optimization problems in edge computing systems, amid increasing requirements for computational speed and accuracy to maintain low-latency and real-time features. Based on mature CMOS technology, analog Ising machines (AIMs), which exploit physical dynamical processes inspired by quantum computing to approximate solutions of quadratic unconstrained binary optimization (QUBO) problems, have emerged as a promising computing paradigm beyond the traditional von Neumann architecture. Despite their potential efficiency advantages, existing AIMs suffer from fundamental limitations arising from their uncontrollable autonomous Lyapunov dynamics, with an inherent trade-off between the reachability of ground states and computing speed.

In this letter, we illustrate a controlled AIM (CAIM) framework that fundamentally extends AIMs from autonomous physical systems to feedback-controlled dynamical optimizers. By treating the binarization strength as a time-dependent control variable and integrating the theory of control Lyapunov function (CLF) with asynchronous momentum-based optimization, CAIMs implement an intelligent control algorithm to accelerate computation processes and enhance solution accuracy and robustness in real time. Experiments on benchmark problems show that CAIM consistently outperforms conventional AIMs, making CAIM a scalable optimization architecture that aligns naturally with the requirements of real-time edge intelligence.