Bioinspired neuromorphic spiking sensors hold great promise for energy-efficient and architecturally simplified neuromorphic networks. Here, an energy-efficient artificial spiking sensor based on a patterned polycrystalline vanadium dioxide (${\text{VO}}_2$) memristor grown by direct-current (dc) reactive magnetron sputtering was developed to achieve high sensitivity and low-energy consumption simultaneously. The memristor demonstrates high endurance in both static and dynamic electrical measurements with low cycle-to-cycle variations and excellent spiking performance when integrated into a 1R (memristor)-1T (transistor) spiking circuit. By optimizing the memristor current and output capacitor, the circuit achieves tunable output frequencies of 181–784 kHz under a 5-V supply voltage and an estimated intrinsic capacitance of approximately 0.51 nF. In-sensor temperature encoding is also achieved under a memristor current of 0.1478 mA without an additional external capacitor, exhibiting an excellent sensitivity of $2.98–13.62\mathrm{kHz}/°\mathrm{C}$, a low total energy consumption of 2.73–5.63 nJ/spike and a signal-to-noise and distortion ratio (SNDR) of 30.95–41.21 dB from 30 to $50°\mathrm{C}$, as well as a high resolution of $0.18°\mathrm{C}$ at $47.5°\mathrm{C}$. Compared with conventional CMOS readout integrated circuits, the energy consumption can be further reduced by lowering the supply voltage and minimizing the intrinsic capacitance through full circuit integration, while the SNDR can be enhanced by mitigating the stochasticity of ${\text{VO}}_2$ and reducing circuit noise, which will require further optimizations of material characteristics and device geometry.

Close

PrevNext

View figure in article

• Metal-insulator transition
• Metrology
• Resistive switching
• Spiking neurons
• Memristors
• Mott insulators
• Solid-state detectors
• Thermometry