Oxygen vacancy engineering of TaOx-based resistive memories by Zr doping for improved variability and synaptic behavior

Nano t-Se Peninsulas Embedded in Natively Oxidized 2D TiSe2 Enable Uniform and Fast Memristive Switching

Memristive devices, regardless of their potential applications in memory and computing scenarios, still suffer from large cycle-to-cycle and device-to-device variations due to the stochastic growth of conductive filaments (CFs). In this work, we fabricated a crossbar memristor using the 2D TiSe₂ material and then oxidized it into TiO₂ in the atmosphere at a moderate temperature. Such a mild oxidation approach fails to evaporate all Se into the air, and after further annealing using thermal or electrical stimulations, the remnant Se atoms gather near the interfaces and grow into nanosized crystals with relatively high conductivity. The resulting peninsula-shaped nanocrystals distort the electric field, forcing CFs to grow on them, which could largely confine the location and length of CFs. As a result, this two-terminal TiSe₂/TiO₂/TiSe₂ device exhibits excellent resistive switching performance with a fairly low threshold voltage (V_set < 0.8 V, V_reset > 0.55 V) and high cycle-to-cycle consistency, enabling resistive switching at narrow operating variations, e.g., 500 ± 48 and 845 ± 39 mV. Our work offers a new approach to minimize the cycle-to-cycle stochasticity of the memristive device, paving the way for its applications in data storage and brain-inspired computing.

Mimicking Pain-Perceptual Sensitization and Pattern Recognition Based on Capacitance- and Conductance-Regulated Neuroplasticity in Neural Network

Neuromorphic computing, inspired by the biological neuronal system, is a high potential approach to substantially alleviate the cost of computational latency and energy for massive data processing. Artificial synapses with regulable synaptic weights are the basis of neuromorphic computation, providing an efficient and low-power system to overcome the constraints of the von Neumann architecture. Here, we report an ITO/TaOx-based synaptic capacitor and transistor. With the drift motion of mobile-charged ions in the TaOx, the capacitance and channel conductance can be tuned to exhibit synaptic weight modulation. Robust stability in the cycle-to-cycle (C2C) variation is found in capacitance and conductance potentiation/depression weight updating of 0.9 and 1.8%, respectively. Simulation results show a higher classification accuracy of handwritten digit recognition (95%) in capacitance synapses than that in conductance synapses (84%). Besides, the synaptic capacitor consumes much less energy than the synaptic transistor. Moreover, the ITO/TaOx-based capacitor successfully emulates the pain-perceptual sensitization on top of the superior performance, indicating its promising potential in applying the capacitive neural network.