XNORBIN：一款用于二值卷积神经网络的95 TOp/s/W硬件加速器

Mar, 2018

XNORBIN：一款用于二值卷积神经网络的95 TOp/s/W硬件加速器

XNORBIN: A 95 TOp/s/W Hardware Accelerator for Binary Convolutional Neural Networks

Andrawes Al Bahou, Geethan Karunaratne, Renzo Andri, Lukas Cavigelli, Luca Benini

TL;DR研究表明，使用极端量化可以将CNN的权重和中间特征图二值化，从而节省内存并将能量密集型的乘积操作转换为XNOR和popcount操作，我们提出的二值CNN加速器XNORBIN在低功率嵌入式系统中具有紧密耦合的计算和内存，能够实现95 TOp/s/W的能量效率和2.0 TOp/s/MGE的面积效率。

Abstract

Deploying state-of-the-art cnns requires power-hungry processors and off-chip memory. This precludes the implementation of cnns in low-power embedded systems. Recent research shows →