Thermal Engineering of NbO₂-Based Memristor for Low-Power and High-Capacity Oscillatory Neural Networks

Negative differential resistance (NDR) devices based on transition metal oxides, such as NbO₂ memristors, inherently exhibit multiple nonlinear dynamics that have garnered considerable interest in emulating neuronal functions. However, the challenge of simultaneously reducing switching voltages and currents while maintaining a stable hysteresis window limits the energy efficiency and computational functionality of NbO₂-based oscillatory systems. Here, a thermal engineering strategy is proposed to break this dilemma, in which a SnSe layer with low thermal conductivity and high electrical conductivity is inserted between the NbO₂ layer and the bottom electrode. This SnSe barrier effectively suppresses thermal dissipation, enabling lower switching voltages and currents in SnSe/NbO₂ devices without compromising their hysteresis window. By using such a thermally optimized device to construct oscillator circuits, a 45% reduction in energy consumption per spike is achieved compared to the NbO_y/NbO₂ control sample. Furthermore, the preserved hysteresis window of SnSe/NbO₂ devices enables the construction of oscillatory neural networks (ONNs) with higher oscillator capacity and computational capability than those based on NbO_y/NbO₂ devices. These findings shed light on thermal engineering for the development of low-power NbO₂-based NDR devices, paving the way for energy-efficient neuromorphic systems and high-capacity ONNs.