Improved Resistive and Synaptic Characteristics in Neuromorphic Systems Achieved Using the Double-Forming Process

Area of Science:

• Materials Science and Engineering
• Electrical Engineering
• Computer Science

Background:

• Resistive Random-Access Memory (RRAM) devices are crucial for developing neuromorphic computing systems.
• Zirconium oxide (ZrO_x) based RRAM devices offer promising characteristics for synaptic applications.
• Understanding the forming process is essential for optimizing RRAM device performance and reliability.

Purpose of the Study:

• To investigate and compare the electrical properties of ITO/ZrO_x/TaN RRAM devices.
• To analyze the impact of single- and double-forming processes on device characteristics.
• To emulate synaptic functions and evaluate performance in a deep neural network for pattern recognition.

Main Methods:

• Device characterization using transmission electron microscopy (TEM), dispersive X-ray spectroscopy (DXS), and X-ray photoelectron spectroscopy (XPS).
• Analysis of resistive switching behaviors, endurance, and retention properties for single- and double-forming methods.
• Synaptic simulations using a pulse scheme and emulation of spike-timing-dependent plasticity (STDP) based on the Hebbian rule.

Main Results:

• The existence of TaON interface layers was confirmed, influencing device behavior.
• The double-forming method demonstrated superior uniformity in endurance and retention compared to the single-forming method.
• Successful emulation of potentiation and depression, achieving high recognition accuracy in a deep neural network-based pattern recognition system.

Conclusions:

• The double-forming process is more suitable for reliable RRAM device operation in neuromorphic applications.
• The study provides insights into filament formation mechanisms and energy band diagrams for RRAM devices.
• The demonstrated synaptic plasticity emulation highlights the potential of these RRAM devices for advanced AI hardware.