Fatigue mechanism of NbOx neuron device is unveiled

Simulating the functionality of biological neurons is a prerequisite for establishing Spiking Neural Networks (SNNs) and forms the foundation for achieving future neuromorphic computing. Devices exhibiting threshold switching characteristics, owing to their highly nonlinear response to electrical bias, are widely employed in constructing artificial neuron devices. Threshold switching devices based on niobium oxide (NbOx) materials possess negative differential resistance characteristics, allowing for high switching speeds (<2 ns) while maintaining low power consumption (~100 fJ), making them naturally suitable for constructing artificial neurons.

However, in practical operation, the application of cyclic electrical stress inevitably leads to changes in material properties. Of particular concern is the drift of the characteristic transition voltage, which significantly compromises the robustness of neural networks, often necessitating additional signal modulation to compensate in applications. Additionally, changes in the transition window alter the oscillation frequency of neuron device pulse emission, ultimately leading to device malfunction. Currently, research on device transition mechanisms primarily focuses on the macroscopic response of individual devices, such as the threshold transition process. There is still a lack of systematic research and abstract model descriptions of device degradation phenomena induced by cyclic electrical stress. This greatly hampers further improvement in device reliability.

Addressing the aforementioned critical issues, we constructed a threshold transition microscopic physical model for NbOx-based neuron devices, introducing the oxygen vacancy migration process induced by electric fields and thermal effects. This model not only effectively describes the threshold transition behavior and corresponding oxygen vacancy dynamics in NbOx devices but also successfully explains the performance degradation mechanism, especially the drift of threshold voltage (Vth) and holding voltage (Vhold), with simulation results highly consistent with experiments. The underlying mechanism is attributed to the comprehensive effects of electric field, temperature, and oxygen vacancy concentration gradients on the redistribution of oxygen vacancies in NbOx materials. Finally, we propose a method to mitigate device degradation by modulating the oxygen vacancy migration process, the effectiveness of which has been validated through simulation and experiments.

The device reliability can be enhanced via optimizing the operation method:

Reference

Fatigue of NbOx-based locally active memristors—
(1) Part I: Experimental Characteristics
Yanting Ding, Yu Li, Shujing Jia, Pei Chen, Xumeng Zhang*, Wei Wang, Yang Li, Yunxia Hao, Jinshun Bi, Tiancheng Gong, Hao Jiang*, Ming Wang, Qi Liu*, Ningsheng Xu, and Ming Liu;

(2) Part II: mechanisms and modeling
Yu Li, Yanting Ding, Xumeng Zhang*, Shujing Jia, Wei Wang, Yang Li, Ming Wang, Hao Jiang*, Qi Liu*, Ningsheng Xu, and Ming Liu,
IEEE Transactions on Electron Devices, 2023.