High-Performance Synapse Arrays for Neuromorphic Computing via Floating Gate-Engineered IGZO Synaptic Transistors.

Area of Science:

• Materials Science
• Electrical Engineering
• Computer Science

Background:

• Neuromorphic computing, mimicking the human brain, presents an alternative to traditional Von Neumann architecture.
• Artificial synapses are crucial for building hardware neuromorphic systems.
• Indium-gallium-zinc oxide (IGZO)-based synaptic transistors offer low-temperature processing and CMOS compatibility but suffer from poor charge de-trapping and retention.

Purpose of the Study:

• To develop advanced IGZO synaptic transistors with enhanced performance for neuromorphic computing.
• To address the limitations of low charge de-trapping efficiency and insufficient retention in existing IGZO synaptic devices.
• To demonstrate the efficacy of an indium-tin oxide (ITO) floating gate in improving synaptic transistor characteristics.

Main Methods:

• Fabrication of IGZO synaptic transistors incorporating an ITO floating gate (FG).
• Utilizing atomic layer deposition (ALD) for the Al2O3 tunneling layer (TL) to ensure a smooth FG/TL interface.
• Testing an 8x8 synapse array with a half-pulse programming scheme.
• Performing spiking neural network simulations on MNIST and Fashion-MNIST datasets.

Main Results:

• The novel ITO FG design significantly enhances electrical performance and device retention.
• The fabricated 8x8 synapse array demonstrated 100% yield and interference-free programming.
• Spiking neural network simulations achieved high accuracies of 98.31% (MNIST) and 87.76% (Fashion-MNIST), even with device variations and retention considered.

Conclusions:

• IGZO synaptic transistors with ITO floating gates effectively overcome previous performance limitations.
• These improved devices show significant potential for practical neuromorphic computing applications.
• The study highlights a viable pathway for developing high-performance artificial synapses for brain-inspired computing.