Highly Reliable 3D Channel Memory and Its Application in a Neuromorphic Sensory System for Hand Gesture Recognition

Area of Science:

• Materials Science
• Computer Engineering
• Neuroscience

Background:

• Neuromorphic computing utilizes resistive random-access memory (RRAM) for data processing.
• Current RRAM technologies suffer from low reliability and poor conductance tunability due to filament formation, limiting accuracy.
• Software neural networks (SW-NN) offer higher accuracy but are less efficient for large datasets.

Purpose of the Study:

• To develop a highly reliable multilevel RRAM with enhanced conductance tunability.
• To improve the accuracy of neuromorphic computing systems for real-world applications.
• To demonstrate the potential of this RRAM in a neuromorphic sensory system for the metaverse.

Main Methods:

• Fabrication of Cobalt Oxide (CoO)-based multilevel RRAM with an optimized 3D grain boundary (GB) network.
• Characterization of RRAM reliability, including cycle-to-cycle endurance and device-to-device stability.
• Integration of the RRAM into a neuromorphic sensory system with a photoacoustic strain sensor for gesture recognition.

Main Results:

• The 3D GB-channel RRAM (3D GB-RRAM) demonstrated enhanced reliability and stability with minimal variation in I-V characteristics.
• Achieved excellent conductance tunability with high symmetricity (624), low nonlinearity, and a large dynamic range (31.1).
• The neuromorphic system attained high recognition accuracies (97.9% for finger motion, 97.4% for hand gestures), comparable to SW-NN.

Conclusions:

• The optimized 3D GB-RRAM offers superior reliability and performance for neuromorphic computing.
• This RRAM technology is a promising candidate for metaverse applications requiring efficient sensory data processing.
• The developed neuromorphic sensory system showcases the practical potential of advanced RRAM devices.