Integrated Design of Electrically Configurable Ferroelectric and Redox-Based Memristors for Hardware-Implemented...

Area of Science:

• Neuromorphic Engineering
• Materials Science
• Computer Science

Background:

• Reservoir computing (RC) provides efficient time-series data processing with reduced computational demands.
• Existing RC systems often require complex architectures and high training costs.
• Memristor-based hardware offers a promising avenue for compact and energy-efficient neuromorphic computing.

Purpose of the Study:

• To develop and demonstrate a hardware-implemented reservoir computing system using novel multifunctional memristors.
• To integrate ferroelectric-based memristors (FM) as volatile reservoir layers and redox-based memristors (RM) as non-volatile readout layers within a single fabrication process.
• To evaluate the performance and scalability of these multifunctional memristors for neuromorphic applications.

Main Methods:

• Fabrication of dual-functional memristors capable of electrical conversion between FM and RM states.
• Comprehensive electrical characterization, including low-frequency noise analysis and synaptic weight linearity evaluation.
• Implementation of both offline and online training schemes for reservoir computing tasks.

Main Results:

• Achieved high linearity factors (<2.3% cycle-to-cycle variation) for synaptic weight tuning during potentiation and depression.
• Ferroelectric-based memristor demonstrated stable reservoir states with low cycle-to-cycle variation (3.52%).
• Offline training yielded 93.3% accuracy on the Modified National Institute of Standards and Technology dataset; online training achieved 88.2% accuracy using incremental pulse schemes.

Conclusions:

• Multifunctional memristors are practically viable for building scalable and versatile neuromorphic systems.
• The integrated FM/RM structure offers seamless operation without additional fabrication steps.
• This work establishes a strong foundation for next-generation computing technologies leveraging memristor-based reservoir computing.