Polarity-dependent dual-mode AlN-embedded RRAM with improved stochastic switching and synaptic modulation for...

Area of Science:

• Materials Science
• Electrical Engineering
• Computer Science

Background:

• Resistive random-access memory (ReRAM) devices are crucial for next-generation computing.
• Achieving both digital and analog switching in a single ReRAM cell is a key challenge for neuromorphic applications.

Purpose of the Study:

• To develop a single ReRAM cell with polarity-dependent, dual-mode switching capabilities.
• To demonstrate the device's potential for energy-efficient neuromorphic computing.

Main Methods:

• Fabrication of a Pt/Al/TaOx/AlN/Al2O3/Pt ReRAM device with a specific layered structure.
• Characterization of device switching behavior under positive and negative bias conditions.
• Evaluation of analog switching performance using an incremental step pulse with verify algorithm.

Main Results:

• The device exhibits polarity-dependent, dual-mode switching: abrupt digital switching under positive bias and gradual analog switching under negative bias.
• The integrated AlN layer acts as a current limiter, enabling self-compliance and controlled filament formation.
• The device achieved 6-bit resolution for analog switching, >10,000s retention, and emulated synaptic plasticity.
• A multilayer perceptron using device data achieved 93.5% accuracy on the MNIST dataset.

Conclusions:

• The developed ReRAM device offers a promising platform for energy-efficient, hardware-level neuromorphic computing.
• The dual-mode switching and synaptic emulation capabilities are significant advancements for artificial intelligence hardware.
• The device's architecture ensures stability and reliability, suppressing current overshoots.