Hf_0.5Zr_0.5O₂-Based Ferroelectric Tunnel Junction as an Artificial Synapse for Speech Recognition

Area of Science:

• Materials Science
• Solid State Physics
• Computer Engineering

Background:

• Ferroelectric tunnel junctions (FTJs) based on Hafnium Oxide (HfO2) are crucial for nonvolatile memory and neuromorphic computing.
• Developing FTJs with enhanced stability, speed, and multi-state capabilities is essential for next-generation electronic devices.

Purpose of the Study:

• To engineer and characterize a novel FTJ memristor with a TiO2 interlayer for improved performance.
• To evaluate the device's potential as an artificial synapse in neuromorphic systems and its application in speech recognition.

Main Methods:

• Fabrication of a Pt/Hf0.5Zr0.5O2/TiO2/TiN FTJ memristor architecture.
• Assessment of resistive switching characteristics, including ON/OFF ratio, speed, retention, and endurance.
• Evaluation of the device as an artificial synapse for multi-state conductance manipulation and its performance in a simulated convolutional neural network for speech recognition.

Main Results:

• The FTJ memristor demonstrated stable resistive switching with a high ON/OFF ratio (5.8 × 10^2) at 50 ns speed and retention over 10^5 s up to 160 °C.
• The TiO2 interlayer significantly improved endurance (2 × 10^8 cycles) and interface properties compared to other HfO2-based FTJs.
• The device achieved highly symmetric 128-state conductance manipulation with low variation (2.75%) and enabled high accuracy (97.6%) in speech recognition tasks, showing remarkable noise immunity (90.2% at 10 dB SNR).

Conclusions:

• The developed TiO2-interlayered FTJ memristor exhibits superior performance, making it highly suitable for multistate nonvolatile memory.
• The device shows significant promise as an artificial synapse for advanced neuromorphic computing applications, including robust speech recognition systems.