Low Power HfOx/TaOx stacked memristors with nanocolumn electrode for neuromorphic computing

Area of Science:

• Materials Science and Engineering
• Electrical Engineering
• Computer Science (Artificial Intelligence)

Background:

• Resistive Random Access Memory (RRAM) shows promise for emulating biological synapses in neuromorphic computing.
• Existing RRAM devices face challenges in power consumption and conductance linearity for efficient neural network implementation.

Purpose of the Study:

• To develop and evaluate a stacked HfOx/TaOx RRAM device with a nanocolumn electrode for low-power neuromorphic applications.
• To investigate the impact of the nanocolumn electrode on device physics, switching characteristics, and system-level performance.

Main Methods:

• Fabrication of a stacked HfOx/TaOx RRAM device featuring a nanocolumn electrode.
• Finite element method simulations to analyze reset and set processes, oxygen vacancy concentration, and temperature distribution.
• System-level validation using a crossbar array integrated with device characteristics, trained on the MNIST dataset via backpropagation.

Main Results:

• The nanocolumn electrode enhances the local electric field, leading to superior conductive filament control and reduced reset/set voltages.
• Reduced programming pulses and lower power consumption were observed due to the nanocolumn electrode's influence on switching processes.
• Improved conductance linearity (long-term potentiation/long-term depression) and a system-level recognition accuracy of 89.72% on MNIST, surpassing plane electrode devices (88.75%).

Conclusions:

• The nanocolumn electrode design significantly enhances RRAM device performance for neuromorphic computing.
• This memristor technology offers a pathway towards more efficient and accurate artificial neural networks.
• The study validates the potential of nanocolumn electrode induced memristors from device physics to system simulation for practical neuromorphic systems.