Proactive Power Management-Based Supply Regulation with Online Learning for Variation-Tolerant Workload-Aware Droop Mitigation in 28nm CMOS

📋 论文概要

该论文提出一种主动电源管理方案，结合神经网络电压降管理单元、高速功率转换器和在线学习引擎，在28nm SoC上实现工作负载感知的电压降缓解，有效应对电源分配网络和工作负载变化。实验表明，该方法可将最差情况电压降降低59%，限制调节次数减少48倍，稳压器峰值效率超过91%。

💡 主要创新点

• 提出基于在线学习的主动电源管理架构，实时调整电源调节策略以适应PDN和工作负载变化，克服了传统固定模型的局限性。
• 将神经网络电压降管理单元与高速功率转换器集成，实现快速预测和主动补偿电压降，显著降低性能退化。
• 通过在线学习引擎持续优化控制参数，在变容条件下保持高效率和低电压降，实现自适应工作负载感知的电源调节。

🏷 关键词

📄 原文摘要

IBM T. J. Watson Research Center, Yorktown Heights, NY 1 5 Abstract A 28nm SoC solution with integrated proactive power management for droop mitigation is demonstrated combining a neural droop management unit, integrated high speed power converter, and an online learning engine to combat the PDN and workload variations. The 28nm test chip integrated with CPU and accelerators achieves 59% worst-case droop reduction, 48× throttling reduction, and >91% regulator peak efficiency, reducing performance degradation from prior fixed-model or throttling-only schemes. Accelerator-enriched microprocessors with highly dynamic workload cause significant supply droops. Recent works address droop challenges through a variety of schemes including reactive [1], or proactive [2], [3] clock throttling, unified clock and power regulation [4], machine learning (ML)-based power management (PM) [5], softwareassisted workload management [6]. However, there are still several unaddressed challenges,

👥 作者与机构

Xi Chen1, Andrew Liss1, William Covington1, Qiankai Cao1, Yiqi Li1, Kang Wei2, Raveesh Magod3, Muhammad Khellah4, Xin Zhang5, Jie Gu1

Northwestern University, Evanston, IL, 2Texas Instruments, Dallas, TX, 3Indian Institute of Technology Madras, Chennai, India, 4Intel, Hillsboro, OR