Paper Info
Title
Nonvolatile Binary Cnn Accelerator With Extremely Low Standby Power Using Rram For Iot Applications
Abstract
Recently, with the development of 5G communications technology, a fully interconnected world is coming. Because 5G has the characteristics of low power consumption, high speed, low cost and small delay, the change brought to the Internet of Things industry is dramatic [1]. Artificial intelligence technology has great potential in the field of IoT devices, but the huge computational complexity makes it difficult to be realized on a power-critical device. In this paper, we demonstrate a nonvolatile binary convolutional neural network accelerator. The main contributions of this work are summarized as follows: (1) A nonvolatile binary CNN data path based on RRAM, which can be fully power-gated in standby state; (2) The matrix multiplication and addition is performed by RRAM other than digital logic, with the binary weights stored in the RRAM; (3) Since the accelerator can be fully powered down, the power dissipated during the standby state is almost zero.
Year
DOI
Venue
2019
10.1109/ASICON47005.2019.8983658
2019 IEEE 13TH INTERNATIONAL CONFERENCE ON ASIC (ASICON)
Field
DocType
ISSN
Standby power,Convolutional neural network,Computer science,Internet of Things,Electronic engineering,Boolean algebra,Matrix multiplication,Binary number,Resistive random-access memory,Computational complexity theory
Conference
2162-7541
Citations 
PageRank 
References 
0
0.34
0
Authors
5
Name
Order
Citations
PageRank
Yujie Cai100.68
Keji Zhou251.79
Xiaoyong Xue32110.91
Mingyu Wang413524.90
Xiaoyang Zeng5442107.26