A Lamellarly Controlled Molecular-Redox-Driven Memristor for Pruned Spiking Neuromorphic Computing

Area of Science:

• Materials Science
• Neuroscience
• Electrical Engineering

Background:

• Low-power memristors and neuromorphic devices are crucial for overcoming energy consumption limitations.
• Precise control over metallic conductive filament (CF) formation at the molecular level remains a significant challenge.

Purpose of the Study:

• To develop a novel molecular material for reconfigurable analog-to-digital memristive operations.
• To achieve precise control of CF growth at the molecular scale for ultralow-power neuromorphic applications.

Main Methods:

• Synthesis of a symmetrical dual-core naphthalene diimide (bis-NDI) molecular material with multi-active, lamellarly ordered redox sites.
• Fabrication and characterization of bis-NDI-based memristors demonstrating analog synaptic behaviors.
• Development of a feedback pruning algorithm for spiking neural networks (SNNs) integrated with the bis-NDI memristor.

Main Results:

• The bis-NDI memristor achieved ultralow-power consumption (90 aJ µm⁻²) and high yield (98%) for analog-to-digital (A-t-D) transition at 0.5 V.
• The material enabled controllable CF growth at the molecular scale, facilitating reconfigurable A-t-D memristive operations.
• Integration with the pruning algorithm reduced connected neurons by up to 92%, maintaining high recognition rates (>90%) for SNNs.

Conclusions:

• The developed bis-NDI molecular material enables efficient, ultralow-power neuromorphic computing through precise molecular-level control of CF formation.
• Co-design of materials and algorithms offers a promising pathway for highly efficient SNNs with reduced systemic energy consumption.
• This work advances the development of next-generation neuromorphic devices and computing paradigms.