A 12.4TOPS/W @ 136GOPS AI-IoT System-on-Chip with 16

📋 论文概要

该论文提出了一款面向AI-IoT应用的SoC，基于RISC-V架构并集成了2-to-8bit精度可变的DNN加速器，通过自适应体偏置技术实现了30%的能效提升。该芯片在12.4 TOPS/W的能效下提供136 GOPS的性能，有效解决了物联网设备在低功耗下运行多样化AI任务的需求。

💡 主要创新点

• 实现了2-to-8bit精度可变的DNN加速，支持动态精度调整以适应不同任务需求。
• 采用自适应体偏置技术，在保证性能的同时降低功耗，提升能效30%。
• 基于RISC-V的异构架构设计，融合了高效能DNN加速与通用计算能力。

🏷 关键词

📄 原文摘要

Francesco Conti1, Davide Rossi1, Gianna Paulin2, Angelo Garofalo1, Alfio Di Mauro2, Georg Rutishauer2, Gianmarco Ottavi1, Manuel Eggimann2, Hayate Okuhara1, Vincent Huard3, Olivier Montfort3, Lionel Jure3, Nils Exibard3, Pascal Gouedo3, Mathieu Louvat3, Emmanuel Botte3, Luca Benini1,2 University of Bologna, Bologna, Italy ETH Zürich, Zürich, Switzerland 3 Dolphin Design, Meylan, France 1 2 Emerging Artificial Intelligence-enabled Internet-of-Things (AI-IoT) SoCs [1-4] for augmented reality, personalized healthcare and nano-robotics need to run a large variety of tasks within a power envelope of a few tens of mW: compute-intensive but bitprecision-tolerant Deep Neural Networks (DNNs), as well as signal processing and control requiring high-precision floating-point. Performance and energy constraints vary greatly between different applications and even within different stages of the same application. We present Marsellus (Fig. 22.1.1), an all-digital AI-IoT end-node

👥 作者与机构

RISC-V, 2-to-8b Precision-Scalable DNN Acceleration and, 30%-Boost Adaptive Body Biasing