TY - GEN
T1 - Characterizing Endurance Degradation of Incremental Switching in Analog RRAM for Neuromorphic Systems
AU - Zhao, Meiran
AU - Wu, Huaqiang
AU - Gao, Bin
AU - Sun, Xiaoyu
AU - Liu, Yuyi
AU - Yao, Peng
AU - Xi, Yue
AU - Li, Xinyi
AU - Zhang, Qingtian
AU - Wang, Kanwen
AU - Yu, Shimeng
AU - Qian, He
N1 - Publisher Copyright:
© 2018 IEEE.
PY - 2018/7/2
Y1 - 2018/7/2
N2 - Resistive random access memory (RRAM) is attractive for neuromorphic computing systems as synaptic weights. In the neural network training, incremental switching occurs between the analog conductance states, thus the analog RRAM devices have unique endurance degradation behaviors compared to the convention digital memory application. In this work, a fast measurement platform is developed to characterize the endurance of incremental switching in analog RRAM. It is found that under weak weight update pulses, the incremental switching cycles of RRAM can be increased for more than 5 orders of magnitude compared with full window switching under strong programming pulses. The 1011-cycle endurance of analog RRAM is proved to be sufficient for training neural networks online for various datasets (from MNIST to ImageNet). However, the nonlinearity and dynamic range of analog RRAM degrade during cycling, which may influence the learning accuracy of the neural network when it re-trains with new datasets.
AB - Resistive random access memory (RRAM) is attractive for neuromorphic computing systems as synaptic weights. In the neural network training, incremental switching occurs between the analog conductance states, thus the analog RRAM devices have unique endurance degradation behaviors compared to the convention digital memory application. In this work, a fast measurement platform is developed to characterize the endurance of incremental switching in analog RRAM. It is found that under weak weight update pulses, the incremental switching cycles of RRAM can be increased for more than 5 orders of magnitude compared with full window switching under strong programming pulses. The 1011-cycle endurance of analog RRAM is proved to be sufficient for training neural networks online for various datasets (from MNIST to ImageNet). However, the nonlinearity and dynamic range of analog RRAM degrade during cycling, which may influence the learning accuracy of the neural network when it re-trains with new datasets.
UR - http://www.scopus.com/inward/record.url?scp=85061791185&partnerID=8YFLogxK
UR - http://www.scopus.com/inward/citedby.url?scp=85061791185&partnerID=8YFLogxK
U2 - 10.1109/IEDM.2018.8614664
DO - 10.1109/IEDM.2018.8614664
M3 - Conference contribution
AN - SCOPUS:85061791185
T3 - Technical Digest - International Electron Devices Meeting, IEDM
SP - 20.2.1-20.2.4
BT - 2018 IEEE International Electron Devices Meeting, IEDM 2018
PB - Institute of Electrical and Electronics Engineers Inc.
T2 - 64th Annual IEEE International Electron Devices Meeting, IEDM 2018
Y2 - 1 December 2018 through 5 December 2018
ER -