arXivDaily arXiv每日学术速递 周一至周五更新
arXiv周末暂无论文更新,休息一下吧,周末愉快~~
arXiv 1706.04332cs.NE
收藏

MATIC:有效低电压神经网络加速器的错误周围学习

MATIC: Learning Around Errors for Efficient Low-Voltage Neural Network Accelerators

Sung Kim, Patrick Howe, Thierry Moreau, Armin Alaghi, Luis Ceze, Visvesh Sathe

摘要:

由于对基于深度神经网络(DNN)的服务的需求不断增长,开发用于DNN的专用硬件加速器的努力正在迅速增长。然而,虽然已经开发了在卷积深度神经网络(Conv-DNN)上具有高性能和效率的加速器,但在全连接DNN(FC-DNN)方面取得的进展较少。在本文中,我们提出了MATIC(Memory Adaptive Training with In-situ Canaries),这是一种方法,可以对加速器权重存储器进行积极的电压缩放,以提高DNN加速器的能效。为了在电压超标的情况下实现精确操作,MATIC在记忆自适应训练过程中将破坏性SRAM读取的特性与神经网络的容错能力相结合。此外,使用来自突触权重的位单元作为原位金丝雀来跟踪运行时环境变化,从而消除了PVT相关的电压裕度。MATIC在我们采用65 nm CMOS制造的低功耗DNN加速器上进行了演示,可实现高达60-80 mV的电压超调(与标称电压相比,总能量减少3.3倍)或18.6倍的应用误差减少。

英文摘要:

As a result of the increasing demand for deep neural network (DNN)-based services, efforts to develop dedicated hardware accelerators for DNNs are growing rapidly. However,while accelerators with high performance and efficiency on convolutional deep neural networks (Conv-DNNs) have been developed, less progress has been made with regards to fully-connected DNNs (FC-DNNs). In this paper, we propose MATIC (Memory Adaptive Training with In-situ Canaries), a methodology that enables aggressive voltage scaling of accelerator weight memories to improve the energy-efficiency of DNN accelerators. To enable accurate operation with voltage overscaling, MATIC combines the characteristics of destructive SRAM reads with the error resilience of neural networks in a memory-adaptive training process. Furthermore, PVT-related voltage margins are eliminated using bit-cells from synaptic weights as in-situ canaries to track runtime environmental variation. Demonstrated on a low-power DNN accelerator that we fabricate in 65 nm CMOS, MATIC enables up to 60-80 mV of voltage overscaling (3.3x total energy reduction versus the nominal voltage), or 18.6x application error reduction.

机构、总结、中文摘要均由AI分析而来,可能存在错误,请以论文正文为准。

↑