A carbon conductive filament-induced robust resistance switching behavior for brain-inspired computing

Area of Science:

• Materials Science
• Neuroscience
• Computer Engineering

Background:

• Memristors are key to brain-inspired computing.
• Active metal electrodes in memristors cause instability, hindering performance.
• Robust conductive filaments are needed for reliable neuromorphic devices.

Purpose of the Study:

• To demonstrate a stable carbon conductive filament-induced memristor.
• To improve device performance and reliability for neuromorphic computing.
• To provide guidelines for memristor research in neuromorphic applications.

Main Methods:

• Fabrication of carbon conductive filament memristors.
• Characterization of switching parameters, retention, and endurance.
• Performance evaluation using MNIST handwriting recognition and ASCII data.
• Analysis using transmission electron microscopy (TEM) and first-principles calculations.

Main Results:

• Achieved low variation coefficients (3.9%/-1.18%) and threshold power (10^-9 W).
• Demonstrated high retention (3 × 10⁶ s) and endurance (10⁷ cycles).
• Reached 96.87% MNIST recognition accuracy and ASCII data functions.

Conclusions:

• Carbon conductive filaments offer a stable and reliable solution for memristor-based neuromorphic computing.
• The demonstrated memristor shows potential for high-performance, low-power brain-inspired systems.
• This work provides a new direction for studying memristors in neuromorphic applications.