TY - JOUR
T1 - Two-Step Read Scheme in One-Selector and One-RRAM Crossbar-Based Neural Network for Improved Inference Robustness
AU - Woo, Jiyong
AU - Yu, Shimeng
N1 - Funding Information:
Manuscript received August 12, 2018; revised September 27, 2018; accepted October 10, 2018. Date of publication October 26, 2018; date of current version November 26, 2018. This work was supported in part by the National Science Foundation (NSF) under Grant NSF-CCF-1552687, in part by NSF/SRC E2CDA, and in part by ASCENT (one of the six SRC/DARPA JUMP Centers). The review of this paper was arranged by Editor J. Kang. (Corresponding author: Shimeng Yu.) J. Woo is with the School of Electrical, Computer and Energy Engineering, Arizona State University, Tempe, AZ 85281 USA (e-mail: jiyong.woo@asu.edu).
Publisher Copyright:
© 2018 IEEE.
PY - 2018/12
Y1 - 2018/12
N2 - Introducing a threshold switching selector in a resistive random access memory (RRAM) is essential for implementing a crossbar array that accurately accelerates neuromorphic computations. But, at an expense, a read voltage (Vread) to be used for inference tasks is inevitably boosted. Therefore, this brief shows the effect of the enlarged Vread on the stability of conductance states of the RRAM relevant to the inference robustness. The multiple conductance states of the analog RRAM achieved by a SPICE simulation are stable under consecutive 106 cycles of nominal Vread. However, each state of the one selector and one RRAM begins to be disturbed at 104 cycles due to the boosted Vread. More importantly, when a certain state exceeds to the next state due to the accumulated Vread stress, a classification accuracy of the neural network is significantly degraded. We, thus, introduce a two-step read scheme that separates the roles of turning on the selector and reading the states. As the selector is turned on rapidly with an additional large pulse, the following Vread can be lowered. As a result, the read disturbance is minimized, and the optimized two-step pulse scheme allows 106 MNIST images to be recognized with >95% accuracy in the neural network.
AB - Introducing a threshold switching selector in a resistive random access memory (RRAM) is essential for implementing a crossbar array that accurately accelerates neuromorphic computations. But, at an expense, a read voltage (Vread) to be used for inference tasks is inevitably boosted. Therefore, this brief shows the effect of the enlarged Vread on the stability of conductance states of the RRAM relevant to the inference robustness. The multiple conductance states of the analog RRAM achieved by a SPICE simulation are stable under consecutive 106 cycles of nominal Vread. However, each state of the one selector and one RRAM begins to be disturbed at 104 cycles due to the boosted Vread. More importantly, when a certain state exceeds to the next state due to the accumulated Vread stress, a classification accuracy of the neural network is significantly degraded. We, thus, introduce a two-step read scheme that separates the roles of turning on the selector and reading the states. As the selector is turned on rapidly with an additional large pulse, the following Vread can be lowered. As a result, the read disturbance is minimized, and the optimized two-step pulse scheme allows 106 MNIST images to be recognized with >95% accuracy in the neural network.
KW - Inference robustness
KW - neuromorphic computing
KW - read disturbance
KW - resistive random access memory (RRAM)
KW - selector device
UR - http://www.scopus.com/inward/record.url?scp=85055700899&partnerID=8YFLogxK
UR - http://www.scopus.com/inward/citedby.url?scp=85055700899&partnerID=8YFLogxK
U2 - 10.1109/TED.2018.2875937
DO - 10.1109/TED.2018.2875937
M3 - Article
AN - SCOPUS:85055700899
SN - 0018-9383
VL - 65
SP - 5549
EP - 5553
JO - IEEE Transactions on Electron Devices
JF - IEEE Transactions on Electron Devices
IS - 12
M1 - 8510826
ER -