Chaotic dynamics in nanoscale NbO₂ Mott memristors for analogue computing

Area of Science:

• Materials Science
• Computational Neuroscience
• Nonlinear Dynamics

Background:

• Current machine learning models use simplified neuron models lacking biological complexity.
• Biological neurons operate near the 'edge of chaos,' crucial for brain computation, learning, and adaptability.
• Existing methods for simulating chaotic dynamics in neural networks often rely on complex transistor circuits.

Purpose of the Study:

• To experimentally realize controllable chaotic dynamics in a single, scalable electronic device.
• To investigate the potential of these chaotic elements in neural-inspired computational systems.
• To demonstrate improved computational performance using these devices in a Hopfield network.

Main Methods:

• Fabrication of nanoscale niobium dioxide (NbO2) Mott memristors exhibiting nonlinear transport and Mott transitions.
• Incorporation of NbO2 memristors into a relaxation oscillator circuit.
• Observation and analysis of tunable periodic and chaotic self-oscillations generated by the memristors.
• Integration of memristors into a Hopfield computing network.

Main Results:

• NbO2 Mott memristors demonstrate both current-controlled and temperature-controlled negative differential resistance.
• Nonlinear current transport and nanoscale thermal fluctuations generate tunable chaotic oscillations.
• Incorporating these memristors into a Hopfield network significantly enhances convergence efficiency and accuracy for difficult problems.

Conclusions:

• Scalable electronic devices exhibiting controllable chaos are now experimentally realized using NbO2 Mott memristors.
• These memristors offer a pathway for developing more sophisticated neural-inspired computing architectures.
• The chaotic dynamics generated by these devices can prevent synchronization and aid in solving complex optimization problems.