TY - JOUR
T1 - Technological Exploration of RRAM Crossbar Array for Matrix-Vector Multiplication
AU - Xia, Lixue
AU - Gu, Peng
AU - Li, Boxun
AU - Tang, Tianqi
AU - Yin, Xiling
AU - Huangfu, Wenqin
AU - Yu, Shimeng
AU - Cao, Yu
AU - Wang, Yu
AU - Yang, Huazhong
N1 - Funding Information:
This work was supported by the National Basic Research 973 Program of China under Grant No. 2013CB329000, the National Natural Science Foundation of China under Grant Nos. 61373026, 61261160501, the Brain Inspired Computing Research of Tsinghua University under Grant No. 20141080934, Tsinghua University Initiative Scientific Research Program, and the Importation and Development of High-Caliber Talents Project of Beijing Municipal Institutions.
Publisher Copyright:
© 2016, Springer Science+Business Media New York.
PY - 2016/1/1
Y1 - 2016/1/1
N2 - Matrix-vector multiplication is the key operation for many computationally intensive algorithms. The emerging metal oxide resistive switching random access memory (RRAM) device and RRAM crossbar array have demonstrated a promising hardware realization of the analog matrix-vector multiplication with ultra-high energy efficiency. In this paper, we analyze the impact of both device level and circuit level non-ideal factors, including the nonlinear current-voltage relationship of RRAM devices, the variation of device fabrication and write operation, and the interconnect resistance as well as other crossbar array parameters. On top of that, we propose a technological exploration flow for device parameter configuration to overcome the impact of non-ideal factors and achieve a better trade-off among performance, energy, and reliability for each specific application. Our simulation results of a support vector machine (SVM) and Mixed National Institute of Standards and Technology (MNIST) pattern recognition dataset show that RRAM crossbar array based SVM is robust to input signal fluctuation but sensitive to tunneling gap deviation. A further resistance resolution test presents that a 6-bit RRAM device is able to realize a recognition accuracy around 90 %, indicating the physical feasibility of RRAM crossbar array based SVM. In addition, the proposed technological exploration flow is able to achieve 10.98 % improvement of recognition accuracy on the MNIST dataset and 26.4 % energy savings compared with previous work. Experimental results also show that more than 84.4 % power saving can be achieved at the cost of little accuracy reduction.
AB - Matrix-vector multiplication is the key operation for many computationally intensive algorithms. The emerging metal oxide resistive switching random access memory (RRAM) device and RRAM crossbar array have demonstrated a promising hardware realization of the analog matrix-vector multiplication with ultra-high energy efficiency. In this paper, we analyze the impact of both device level and circuit level non-ideal factors, including the nonlinear current-voltage relationship of RRAM devices, the variation of device fabrication and write operation, and the interconnect resistance as well as other crossbar array parameters. On top of that, we propose a technological exploration flow for device parameter configuration to overcome the impact of non-ideal factors and achieve a better trade-off among performance, energy, and reliability for each specific application. Our simulation results of a support vector machine (SVM) and Mixed National Institute of Standards and Technology (MNIST) pattern recognition dataset show that RRAM crossbar array based SVM is robust to input signal fluctuation but sensitive to tunneling gap deviation. A further resistance resolution test presents that a 6-bit RRAM device is able to realize a recognition accuracy around 90 %, indicating the physical feasibility of RRAM crossbar array based SVM. In addition, the proposed technological exploration flow is able to achieve 10.98 % improvement of recognition accuracy on the MNIST dataset and 26.4 % energy savings compared with previous work. Experimental results also show that more than 84.4 % power saving can be achieved at the cost of little accuracy reduction.
KW - electronic design automation
KW - machine learning
KW - matrix-vector multiplication
KW - non-ideal factor
KW - resistive switching random access memory (RRAM)
UR - http://www.scopus.com/inward/record.url?scp=84953426381&partnerID=8YFLogxK
UR - http://www.scopus.com/inward/citedby.url?scp=84953426381&partnerID=8YFLogxK
U2 - 10.1007/s11390-016-1608-8
DO - 10.1007/s11390-016-1608-8
M3 - Article
AN - SCOPUS:84953426381
SN - 1000-9000
VL - 31
SP - 3
EP - 19
JO - Journal of Computer Science and Technology
JF - Journal of Computer Science and Technology
IS - 1
ER -