Enhancing stability and iterative learning in neuromorphic memristor via TiN/SiO/TiN interface engineering

Area of Science:

• Materials Science
• Electrical Engineering
• Computer Science

Background:

• Resistive random-access memory (ReRAM) is a promising technology for next-generation computing.
• Interface-type ReRAM devices offer potential for high performance and low power consumption.
• Neuromorphic computing requires efficient and stable synaptic devices.

Purpose of the Study:

• To fabricate and characterize SiO2-based interface-type ReRAM devices.
• To evaluate the performance of these devices for neuromorphic applications, specifically MNIST digit recognition.
• To investigate the impact of endurance and material choice on synaptic device performance.

Main Methods:

• Fabrication of TiN/SiO2/TiN and Pt/SiO2/Pt ReRAM devices.
• Electrical characterization including I-V curves, endurance tests, and retention measurements.
• Evaluation of device linearity and synaptic weight degradation for neuromorphic learning.

Main Results:

• Devices operated below 3 V with <1 mA current, achieving an on/off ratio of ~10.
• Fast switching speeds (1 μs set/reset) and good retention (10^4 s at 85 °C) were demonstrated.
• High linearity enabled 92.21% accuracy in MNIST recognition; TiN/SiO2/TiN showed superior endurance due to an oxygen reservoir.

Conclusions:

• SiO2-based interface-type ReRAM devices exhibit excellent performance for neuromorphic computing.
• The TiN/SiO2/TiN structure provides enhanced endurance and stability, crucial for synaptic applications.
• Gradual switching dynamics and device robustness contribute to efficient learning in neuromorphic systems.