The Enhanced Performance of Neuromorphic Computing Hardware in an ITO/ZnO/HfO_x/W Bilayer-Structured Memory Device

Area of Science:

• Materials Science
• Electrical Engineering
• Computer Science

Background:

• Neuromorphic systems aim to mimic the human brain's structure and function.
• Resistive Random-Access Memory (RRAM) devices are key components for neuromorphic computing.
• Improving the performance and stability of RRAM is crucial for advancing neuromorphic applications.

Purpose of the Study:

• To investigate the potential of a novel ITO/ZnO/HfOx/W bilayer-structured memory device for neuromorphic systems.
• To evaluate the resistive switching characteristics, endurance, retention, and synaptic plasticity of the proposed device.
• To compare the performance of the bilayer device with traditional RRAM devices.

Main Methods:

• Fabrication of ITO/ZnO/HfOx/W bilayer-structured memory devices.
• Characterization of resistive switching behavior, including ON/OFF ratio, endurance, and retention.
• Analysis of potentiation and depression linearity for neural network applications.
• Observation of spike-timing-dependent plasticity (STDP) behavior.

Main Results:

• The ITO/ZnO/HfOx/W devices exhibited uniform resistive switching characteristics.
• Demonstrated favorable endurance (>10^2) and stable retention (>10^4 s).
• Filament formation/rupture at the ZnO/HfOx interface led to a higher ON/OFF ratio and improved cycle uniformity compared to devices without HfOx.
• Linear potentiation and depression responses were observed, suitable for neural network pattern recognition.
• Spike-timing-dependent plasticity (STDP) behavior was successfully demonstrated.

Conclusions:

• The ITO/ZnO/HfOx/W bilayer structure shows significant potential as a memory component for neuromorphic systems.
• The enhanced performance, particularly the higher ON/OFF ratio and improved uniformity, makes it a promising candidate for future neuromorphic hardware.
• The observed STDP behavior further validates its applicability in mimicking synaptic plasticity for neural networks.