RESUMO
The quantum conductance (QC) behaviors in synaptic devices with stable and tunable conductance states are essential for high-density storage and brain-like neurocomputing (NC). In this work, inspired by the discontinuous transport of fluid in spider silk, a synaptic device composed of a silicon oxide nanowire network embedded with silicon quantum dots (Si-QDs@SiOx) is designed. The tunable QC behaviors are achieved in both the SET and RESET processes, and the QC states exhibit stable retention time exceeding 104 s in the synaptic device and show stable reproducibility after an interval of two months. The synaptic plasticity, including long-term potentiation/depression and Pavlovian conditioning function, is simulated based on the tunable conductance. The mechanism of stable and tunable QC behaviors is analyzed and clarified by beading effect of spider silk in Si-QDs@SiOx nanowires structure. The digit recognition capability of the device is evaluated by simulation using an artificial neural network consisting of the Si-QDs@SiOx-based synaptic device. These results provide insights into the development of neurocomputing systems with high classification accuracy.
RESUMO
Oxide-based memristors composed of Ag/porous SiOx/Si stacks are fabricated using different etching time durations between 0 and 90 s, and the memristive properties are analyzed in the relative humidity (RH) range of 30-60%. The combination of humidity and porous structure provides binding sites to control silver filament formation with a confined nanoscale channel. The memristive properties of devices show high on/off ratios up to 108 and a dispersion coefficient of 0.1% of the high resistance state (CHRS) when the RH increases to 60%. Humidity-mediated silver ion migration in the porous SiOx memristors is investigated, and the mechanism leading to the synergistic effects between the porous structure and environmental humidity is elucidated. The artificial neural network constructed theoretically shows that the recognition rate increases from 60.9 to 85.29% in the RH range of 30-60%. The results and theoretical understanding provide insights into the design and optimization of oxide-based memristors in neuromorphic computing applications.
