Tuning Bienenstock-Cooper-Munro learning rules in a two-terminal memristor for neuromorphic computing

Area of Science:

• Materials Science
• Neuroscience
• Computer Science

Background:

• The Bienenstock-Cooper-Munro (BCM) learning rule is crucial for artificial synapses in neuromorphic computing, enhancing synaptic modulation balance and reducing energy consumption.
• Current implementations of the BCM rule often rely on complex field-effect transistors, limiting progress in simpler two-terminal memristors due to insufficient tunable parameters.

Purpose of the Study:

• To investigate the feasibility of implementing BCM-like learning rules in a two-terminal memristor by manipulating series resistance.
• To explore different types of BCM-like plasticity and their dependence on device parameters.

Main Methods:

• Utilized a two-terminal BaTiO3 memristor device.
• Systematically adjusted the series resistance to observe changes in synaptic plasticity.
• Employed X-ray photoelectron spectroscopy (XPS) to analyze the underlying physical mechanisms.

Main Results:

• Identified three distinct BCM-like learning rules by varying series resistance: abnormal (low resistance), monotonous (high resistance), and enhanced depression (moderate resistance).
• Demonstrated that these rules are linked to non-monotonous conductance modulation driven by ionized oxygen vacancy migration.
• Achieved spike rate-dependent plasticity (SRDP) and history-dependent plasticity.

Conclusions:

• Successfully implemented diverse BCM-like learning rules in a simple two-terminal memristor by controlling series resistance.
• The findings provide a new pathway for developing energy-efficient artificial synapses and advancing neuromorphic computing hardware.