부산대학교 도서관

The function of artificial oscillation neurons (ONs) has been emulated using threshold switching (TS) properties. Thus far, the TS behavior of SiOx materials has been demonstrated by incorporating toxic elements (e.g., As or Te). This study demonstrates TS using a toxic element-free undoped SiOx layer sandwiched by Nb electrodes. We identify that the TS in the SiOx is related to instantly clustered oxygen vacancies ( ${V} _{{0}}$ ) through comparative studies with NbO2-based TS devices and Nb/SiOx interface analyses. The transition from memory switching to nonlinear behavior via the abrupt TS with respect to the Ar/O gas ratio during SiOx deposition further indicates the significance of the amount of ${V} _{{0}}$ . Transient analysis of TS devices shows both volatile output current for triangular pulses and reproducible voltage oscillation for square pulses. The achieved oscillation response is tuned using an input voltage and load resistor; thus, the SiOx-based TS can serve as the ON. Hence, we demonstrate the recognition of 3 $\times $ 5 alphabetic patterns in oscillatory neural networks (ONNs) comprising 15 SiOx-based TSs through MATLAB simulations.