Dynamic FeO/FeWO nanocomposite memristor for neuromorphic and reservoir computing

Area of Science:

• Materials Science
• Nanotechnology
• Computer Engineering

Background:

• Memristors are key components for miniaturized, energy-efficient computing, essential for neuromorphic and in-memory computing.
• Current memristor applications often demand distinct volatile or nonvolatile operational modes.
• Achieving dual-mode functionality in a single device is critical for versatile AI hardware.

Purpose of the Study:

• To develop a forming-free memristor with dual volatile and nonvolatile switching capabilities.
• To explore the impact of compliance current (CC) and structural engineering on memristor performance.
• To demonstrate the memristor's potential for advanced AI applications like data storage and reservoir computing.

Main Methods:

• Fabrication of a Ag/FeOₓ/FeWOₓ/Pt nanocomposite memristor.
• Adjustment of compliance current (CC) levels to control switching modes (volatile <500 μA, nonvolatile >mA).
• Evaluation of device performance including uniformity, data retention, multilevel switching, and emulation of synaptic functions (potentiation, depression, SRDP).

Main Results:

• The memristor exhibits forming-free, dual-mode switching controlled by CC.
• Excellent performance metrics including low operating voltage (<0.2 V), uniformity, data retention, and multilevel switching.
• Successful emulation of biological synapse functions and memory transitions (STM to LTM).
• Demonstrated effectiveness in reservoir computing for sequence data classification with a 93.4% recognition rate.

Conclusions:

• The Ag/FeOₓ/FeWOₓ/Pt memristor offers a versatile solution for AI hardware due to its dual-mode switching.
• Its ability to mimic synaptic functions and memory processes is highly promising for neuromorphic computing.
• The device's performance in reservoir computing highlights its potential for temporal and sequential data processing in AI.