Synergistic Approach of Interfacial Layer Engineering and READ-Voltage Optimization in HfO₂-Based FeFETs for...

Area of Science:

• Materials Science
• Electrical Engineering
• Computer Science

Background:

• Ferroelectric field-effect transistors (FeFETs) are promising for neuromorphic computing.
• Hafnium oxide (HfO2)-based FeFETs offer potential for high-performance synaptic devices.
• Interfacial layer (IL) engineering is crucial for optimizing FeFET characteristics.

Purpose of the Study:

• To enhance the performance of HfO2-based FeFETs through interfacial layer engineering and READ-voltage optimization.
• To evaluate the suitability of FeFETs with different interfacial layers (SiO2 and SiON) as synaptic devices for neuromorphic applications.
• To assess the impact of device degradation on neural network performance for inference-only operations.

Main Methods:

• Fabrication and characterization of FeFET devices with SiO2 and SiON interfacial layers.
• Neuromorphic simulations to evaluate FeFETs as synaptic devices.
• System-level simulations to analyze degradation effects (retention and noise) on neural network accuracy.

Main Results:

• FeFETs with SiO2 IL showed better low-frequency characteristics, while SiON IL exhibited superior WRITE endurance and retention.
• Device WRITE endurance was insufficient for online neural network training, necessitating an inference-only approach.
• System-level simulations indicated retention degradation significantly impacts inference-only operations more than low-frequency noise.
• A neural network utilizing SiON IL-based FeFETs achieved 96% accuracy on the MNIST dataset for inference, with a minor 2.5% deviation from the software baseline.

Conclusions:

• Synergistic engineering of interfacial layers and READ-voltage optimization can improve FeFET performance.
• SiON-based FeFETs are suitable for inference-only neuromorphic applications despite retention degradation challenges.
• Offline training followed by inference-only operation is a viable strategy for FeFET-based neuromorphic systems.
• Further improvements in WRITE endurance and retention are necessary for more demanding neuromorphic tasks.