Universal Dynamics Analysis of Locally-Active Memristors and Its Applications

Locally-active memristor (LAM) is one of the promising candidates of artificial neurons, indicating it has potential applications in neuromorphic computing. Quantitative theoretical analysis on LAMs can provide benefits for designing related oscillator circuits and systems. This study begins with the aim of assessing the importance of DC V-I characteristic in the performance of LAMs by using small-signal analysis method. The DC V-I curve of the LAM is specified by two parameters involving resistance (conductance) and differential resistance (differential conductance). In addition to these two static parameters, we extract a crucial dynamic parameter to describe the behavior of the LAM. Theoretical analysis demonstrates that the performance of generic current-controlled and voltage-controlled LAMs is closely associated with three crucial parameters, i.e., the above two static and one dynamic parameters. Hence, only based on these three parameters, can one derive the small-signal equivalent circuit of LAMs and determine the oscillation frequency range and condition for simple LAM-based oscillators. By applying the presented universal dynamics analysis results, we further propose a modified mathematical model with higher accuracy to mimic the quasi-static and oscillating behaviors of a real Nb2O5 device, and provide some fundamental guidance for the design of LAM-based high-frequency oscillators.

Automated summary

Locally-active memristor (LAM) has potential applications in neuromorphic computing as an artificial neuron. Quantitative theoretical analysis on LAMs shows that their performance is closely associated with three crucial parameters: resistance, differential resistance, and a dynamic parameter. These parameters can be used to derive the small-signal equivalent circuit of LAMs and determine the oscillation frequency range and conditions for LAM-based oscillators.