Razor II: In Situ Error Detection and Correction for PVT and SER Tolerance

📋 论文概要

本文提出Razor II技术，通过原位延迟误差检测和纠正，消除传统canary电路的安全裕度，并利用观察到的错误率动态调整电源电压，以应对PVT变化和软错误率（SER）的影响。该技术实现了更高的能效和鲁棒性。

💡 主要创新点

• 提出原位延迟误差检测与纠正机制，无需安全裕度，直接监控关键路径时序。
• 基于实时错误率反馈动态调节电源电压，同时容忍PVT变化和软错误。

🏷 关键词

📄 原文摘要

Performance variation due to PVT uncertainty has led to an increased interest in adaptive designs. Traditional methods for adaptive design have used so-called canary circuits that mimic the critical-path delay of the actual design [1-4]. However, canary circuits require safety margins for possible mis-tracking and intradie PVT variations. They also have difficulty responding to rapidly changing conditions. To eliminate these margins, the Razor approach [5] introduces in situ delay-error detection and correction where the supply voltage is tuned using the observed error rate. Razor provides energy reduction in two ways: 1) by lowering the supply voltage to the point of first failure (PoFF), all margins due to global and local PVT variations are eliminated. 2) by purposefully operating below

👥 作者与机构

David Blaauw1, Sudherssen Kalaiselvan2, Kevin Lai1, Wei-Hsiang Ma1,

Sanjay Pant1, Carlos Tokunaga1, Shidhartha Das3, David Bull3 1 University of Michigan, Ann Arbor, MI, 2AMD, Sunnyvale, CA ARM, Cambridge, United Kingdom