Sol-Gel-Processed Y₂O₃ Multilevel Resistive Random-Access Memory Cells for Neural Networks.

Area of Science:

• Materials Science
• Electrical Engineering
• Nanotechnology

Background:

• Resistive random-access memory (RRAM) offers promising non-volatile memory solutions.
• Yttrium oxide (Y2O3) is explored as a potential dielectric material for RRAM devices.
• Achieving multilevel cell (MLC) functionality is key for increasing memory density.

Purpose of the Study:

• To fabricate and characterize Y2O3-based RRAM devices.
• To investigate the impact of current compliance on device performance and filament formation.
• To evaluate the potential of these RRAM devices for MLC switching and neuromorphic computing.

Main Methods:

• Sol-gel fabrication of Y2O3 thin films on ITO/glass substrates.
• Bipolar resistive switching measurements with varying current compliance.
• Analysis of conductive filament composition and evolution.
• Numerical simulations for neural network applications.

Main Results:

• Y2O3 RRAM devices exhibited bipolar switching without a forming process.
• Adjusting current compliance controlled Ag filament formation and enabled MLC switching.
• Devices achieved three low-resistance states and one high-resistance state for 2-bit/cell capacity.
• Simulated neural networks using these RRAM devices achieved ~88% digit image classification accuracy.

Conclusions:

• Yttrium oxide RRAM devices are suitable for multilevel cell operation.
• Controlled filament engineering is crucial for MLC functionality.
• These RRAM devices show significant potential for practical neuromorphic systems.
• The fabricated devices offer a viable memory component for future AI hardware.