A Variation-Tolerant In-eDRAM Continuous-Time Ising Machine Featuring 15-Level Coefficients and Leaked Negative-Feedback Annealing

📋 论文概要

该论文提出了一种基于eDRAM的变容错连续时间Ising机，通过15级系数和泄漏负反馈技术，实现对组合优化问题的高效求解。针对传统离散时间Ising机在系数规模和变容能力上的局限，该设计在连续时间域中实现了更强的容错性和更高的求解质量。

💡 主要创新点

• 首次将eDRAM用于连续时间Ising机，结合泄漏负反馈实现变容错能力，降低工艺偏差影响。
• 采用15级可编程系数，扩展了Ising机的动态范围，提升复杂组合优化问题的求解精度。
• 连续时间工作模式避免了离散时间中的量化误差和同步开销，实现更自然的收敛行为。

🏷 关键词

📄 原文摘要

in various real-world decision-making and planning problems. However, most CO problems (COPs) are NP-hard, demanding substantial computational resources with conventional computers. The Ising machine is promising in addressing them efficiently thanks to its natural convergence behavior. Recently, CMOS-based Ising machines have been developed as low-cost COP solvers, which can be categorized as discrete-time (DT) [1-4] and continuous-time (CT) ones [57]. Benefiting from scalable coefficient levels and available annealing strategies, DT Ising machines can solve complicated COPs and reach a lower Hamiltonian. However, DT solvers usually suffer from limited parallelism to prevent potential spin-state oscillations. Together with the requirement of synchronized clocks, the time-to-solution (TTS) is long.

👥 作者与机构

Jiahao Song*, Zihan Wu*, Xiyuan Tang, Bocheng Xu, Haoyang Luo,

Youming Yang, Yuan Wang, Runsheng Wang, Ru Huang Peking University, Beijing, China *Equally Credited Authors (ECAs) Combinatorial optimization (CO) is essential