Time-dependent variability in RRAM-based analog neuromorphic system for pattern recognition

Abstract

For the first time, this work investigated the timedependent variability (TDV) in RRAMs and its interaction with the RRAM-based analog neuromorphic circuits for pattern recognition. It is found that even the circuits are well trained, the TDV effect can introduce non-negligible recognition accuracy drop during the operating condition. The impact of TDV on the neuromorphic circuits increases when higher resistances are used for the circuit implementation, challenging for the future low power operation. In addition, the impact of TDV cannot be suppressed by either scaling up with more synapses or increasing the response time and thus threatens both real-time and general-purpose applications with high accuracy requirements. Further study on different circuit configurations, operating conditions and training algorithms, provides guidelines for the practical hardware implementation.

Impact of RTN on Pattern Recognition Accuracy of RRAM-Based Synaptic Neural Network

“…On the other hand, RTN provides useful information on the responsible defect [9,10]. The impact of RTN has been analyzed for CF RRAM [11], but there is a lack of comparative studies on the non-CF (NCF) RRAM whose synaptic application has also drawn extensive interests [4,12].…”

Exploring the Impact of Random Telegraph Noise-Induced Accuracy Loss on Resistive RAM-Based Deep Neural Network

“…RTN in RRAMs can vary with the quality of the fabrication process, which in turn affects the average number of defects causing RTN, n, and the average conductance fluctuation, δg, [9]. By using different n and δg, the accuracy loss for all the DNNs/dataset can be re-assessed.…”

“…The procedure on the integration of RTN into each RRAM is summarized in Fig. 3: For each conductance g0, the trap number, n and the corresponding conductance fluctuation δg are firstly obtained using the method in [9]. For each trap, τc and τe are generated randomly from their lognormal distribution.…”

“…The construction follows their standard structure. The 3-layer perceptron network (MLP) was also constructed for comparison purpose [9]. For all DNNs, the rectified linear unit (ReLU) is used as the activation function after every convolutional and fully-connected layer, which allow us to scale the conductance of the RRAMs to represent synaptic weights.…”

2D Materials Based Optoelectronic Memory: Convergence of Electronic Memory and Optical Sensor

“…The tunable resistance under continuous stimuli of voltage or current inputs is similar to the synaptic weight (generally determined as biological strength of the connection between two neurons). However, the conventional memory synaptic devices for neuromorphic computing are manipulated through the electrical method and are isolated from the sensors (e.g., visual, auditory or olfactory) [9][10][11][12][13][14][15][16][17]. ORAM synaptic devices integrate optical sensors and synaptic RAMs, which can respond to optical stimuli and exhibit light-tunable synaptic behaviors, including light-tunable short-term plasticity (STP), long-term plasticity (LTP), spike-timing-dependent plasticity (STDP), and spatiotemporal learning rule (STLR).…”