The original paper is in English. Non-English content has been machine-translated and may contain typographical errors or mistranslations. ex. Some numerals are expressed as "XNUMX".
Copyrights notice
The original paper is in English. Non-English content has been machine-translated and may contain typographical errors or mistranslations. Copyrights notice
Une structure efficace de mémoire vive résistive (ReRAM) est développée pour accélérer le réseau neuronal convolutif (CNN) alimenté par le calcul en mémoire. Un nouveau circuit de cellules ReRAM est conçu avec une accessibilité bidirectionnelle (2D). L'ensemble du système de mémoire est organisé sous la forme d'un tableau 2D, dans lequel des cellules de mémoire spécifiques peuvent être consultées de manière identique par localisation de colonne et de ligne. Pour les calculs en mémoire des CNN, seules les cellules pertinentes d'un sous-réseau identique sont accessibles par des opérations de lecture 2D, ce qui est difficilement implémenté par les cellules ReRAM conventionnelles. De cette manière, l'accès redondant (colonne ou ligne) des structures ReRAM conventionnelles est évité afin d'éliminer les mouvements de données inutiles lorsque les CNN sont traités en mémoire. D'après les résultats de la simulation, l'efficacité énergétique et la bande passante de la structure de mémoire proposée sont respectivement 1.4x et 5x par rapport à une architecture ReRAM de pointe.
Yan CHEN
Hunan University,Nara Institute of Science and Technology
Jing ZHANG
Hunan University
Yuebing XU
Hunan University
Yingjie ZHANG
Hunan University
Renyuan ZHANG
Nara Institute of Science and Technology
Yasuhiko NAKASHIMA
Nara Institute of Science and Technology
The copyright of the original papers published on this site belongs to IEICE. Unauthorized use of the original or translated papers is prohibited. See IEICE Provisions on Copyright for details.
Copier
Yan CHEN, Jing ZHANG, Yuebing XU, Yingjie ZHANG, Renyuan ZHANG, Yasuhiko NAKASHIMA, "A ReRAM-Based Row-Column-Oriented Memory Architecture for Convolutional Neural Networks" in IEICE TRANSACTIONS on Electronics,
vol. E102-C, no. 7, pp. 580-584, July 2019, doi: 10.1587/transele.2018CTS0001.
Abstract: An efficient resistive random access memory (ReRAM) structure is developed for accelerating convolutional neural network (CNN) powered by the in-memory computation. A novel ReRAM cell circuit is designed with two-directional (2-D) accessibility. The entire memory system is organized as a 2-D array, in which specific memory cells can be identically accessed by both of column- and row-locality. For the in-memory computations of CNNs, only relevant cells in an identical sub-array are accessed by 2-D read-out operations, which is hardly implemented by conventional ReRAM cells. In this manner, the redundant access (column or row) of the conventional ReRAM structures is prevented to eliminated the unnecessary data movement when CNNs are processed in-memory. From the simulation results, the energy and bandwidth efficiency of the proposed memory structure are 1.4x and 5x of a state-of-the-art ReRAM architecture, respectively.
URL: https://global.ieice.org/en_transactions/electronics/10.1587/transele.2018CTS0001/_p
Copier
@ARTICLE{e102-c_7_580,
author={Yan CHEN, Jing ZHANG, Yuebing XU, Yingjie ZHANG, Renyuan ZHANG, Yasuhiko NAKASHIMA, },
journal={IEICE TRANSACTIONS on Electronics},
title={A ReRAM-Based Row-Column-Oriented Memory Architecture for Convolutional Neural Networks},
year={2019},
volume={E102-C},
number={7},
pages={580-584},
abstract={An efficient resistive random access memory (ReRAM) structure is developed for accelerating convolutional neural network (CNN) powered by the in-memory computation. A novel ReRAM cell circuit is designed with two-directional (2-D) accessibility. The entire memory system is organized as a 2-D array, in which specific memory cells can be identically accessed by both of column- and row-locality. For the in-memory computations of CNNs, only relevant cells in an identical sub-array are accessed by 2-D read-out operations, which is hardly implemented by conventional ReRAM cells. In this manner, the redundant access (column or row) of the conventional ReRAM structures is prevented to eliminated the unnecessary data movement when CNNs are processed in-memory. From the simulation results, the energy and bandwidth efficiency of the proposed memory structure are 1.4x and 5x of a state-of-the-art ReRAM architecture, respectively.},
keywords={},
doi={10.1587/transele.2018CTS0001},
ISSN={1745-1353},
month={July},}
Copier
TY - JOUR
TI - A ReRAM-Based Row-Column-Oriented Memory Architecture for Convolutional Neural Networks
T2 - IEICE TRANSACTIONS on Electronics
SP - 580
EP - 584
AU - Yan CHEN
AU - Jing ZHANG
AU - Yuebing XU
AU - Yingjie ZHANG
AU - Renyuan ZHANG
AU - Yasuhiko NAKASHIMA
PY - 2019
DO - 10.1587/transele.2018CTS0001
JO - IEICE TRANSACTIONS on Electronics
SN - 1745-1353
VL - E102-C
IS - 7
JA - IEICE TRANSACTIONS on Electronics
Y1 - July 2019
AB - An efficient resistive random access memory (ReRAM) structure is developed for accelerating convolutional neural network (CNN) powered by the in-memory computation. A novel ReRAM cell circuit is designed with two-directional (2-D) accessibility. The entire memory system is organized as a 2-D array, in which specific memory cells can be identically accessed by both of column- and row-locality. For the in-memory computations of CNNs, only relevant cells in an identical sub-array are accessed by 2-D read-out operations, which is hardly implemented by conventional ReRAM cells. In this manner, the redundant access (column or row) of the conventional ReRAM structures is prevented to eliminated the unnecessary data movement when CNNs are processed in-memory. From the simulation results, the energy and bandwidth efficiency of the proposed memory structure are 1.4x and 5x of a state-of-the-art ReRAM architecture, respectively.
ER -