Proton-Assisted Redox-Based Three-Terminal Memristor for Synaptic Device Applications.

Area of Science:

• Materials Science
• Electrical Engineering
• Computer Science

Background:

• Neuromorphic computing systems leverage emerging technologies like spintronics, 2D materials, and memristive devices.
• Two-terminal memristors face challenges in concurrent signal transmission and memory operations.

Purpose of the Study:

• To present a complementary metal-oxide-semiconductor-compatible three-terminal memristor (3TM).
• To demonstrate its suitability as a synaptic device for artificial neural networks (NNs).

Main Methods:

• Fabrication of a 3TM compatible with complementary metal-oxide-semiconductor technology.
• Investigation of the switching mechanism involving oxygen ion and proton migration.
• Evaluation of synaptic operation, endurance, and dynamic range.
• Simulation and implementation of the 3TM into a four-layer NN model for MNIST recognition.

Main Results:

• The 3TM shows highly linear weight update characteristics and a dynamic range of approximately 15.
• Excellent endurance performance exceeding 256,000 synaptic weight updates was achieved.
• The device demonstrated a stable dynamic range throughout the endurance tests.
• The NN model incorporating the 3TM achieved approximately 92% accuracy in MNIST handwritten digit recognition.

Conclusions:

• The proposed 3TM, with its desirable conductance modulation characteristics, is a promising candidate for hardware implementation of artificial NNs.
• The device's performance, including linearity, dynamic range, and endurance, supports its application in neuromorphic computing.
• The switching mechanism, influenced by protonic defects and humidity, provides insights for future device optimization.