Local active memristive oscillator enables controllable complex behaviours and frequency domain extraction

Area of Science:

• Materials Science
• Condensed Matter Physics
• Computational Neuroscience

Background:

• Physical non-linearities near the Mott transition are key for neuromorphic computing.
• Current research often simplifies Mott devices, lacking unified models for complex dynamics.

Purpose of the Study:

• Develop a thermodynamic compact model for vanadium oxide devices.
• Propose an injection-based control method for non-linear oscillator behaviors.

Main Methods:

• Utilized electrical measurements and the local active principle.
• Developed a thermodynamic compact model for vanadium oxide devices.
• Investigated non-linear oscillator control for frequency division, stochastic oscillations, and frequency locking.

Main Results:

• Demonstrated an injection-based control method for complex non-linear dynamics.
• Showcased a single device at the edge of chaos for frequency domain information extraction.
• Achieved performance comparable to a two-layer convolutional neural network for a specific task.

Conclusions:

• Bridged physical non-linearities, circuit dynamics, and computational theory.
• Facilitated a shift towards local active devices in neuromorphic computing.
• Advanced dynamic neuromorphic computing through a unified theoretical model.