Thermally stable threshold selector based on CuAg alloy for energy-efficient memory and neuromorphic computing...

Area of Science:

• Materials Science and Engineering
• Electrical Engineering
• Computer Science

Background:

• Cross-point memory arrays offer a solution to the von Neumann bottleneck for high-density data storage and neuromorphic computing.
• Sneak-path currents in these arrays limit scalability and read accuracy, necessitating the development of selector devices.
• The one-selector-one-memristor (1S1R) architecture integrates a selector with each memristor to mitigate sneak-path issues.

Purpose of the Study:

• To demonstrate a thermally stable, electroforming-free selector device based on a CuAg alloy.
• To integrate this selector into a 1S1R cross-point array for improved memory and computing applications.
• To explore the potential of the CuAg alloy selector in neuromorphic computing beyond synaptic applications, specifically in neuron design.

Main Methods:

• Fabrication and characterization of a CuAg alloy-based selector device.
• Integration of the selector with SiO2-based memristors to form a vertically stacked 64x64 1S1R cross-point array.
• Experimental implementation of a selector-based leaky integrate-and-fire neuron circuit.

Main Results:

• The CuAg alloy selector exhibited thermal stability, electroforming-free operation, tunable threshold voltage, and an ON/OFF ratio exceeding 7 orders of magnitude.
• The integrated 1S1R devices demonstrated extremely low leakage currents and suitable switching characteristics for storage-class memory and synaptic weight storage.
• A functional selector-based leaky integrate-and-fire neuron was successfully implemented, showcasing expanded application potential.

Conclusions:

• The CuAg alloy selector is a promising component for overcoming sneak-path current limitations in 1S1R cross-point memory arrays.
• The developed 1S1R array is suitable for next-generation data storage and neuromorphic computing applications.
• CuAg alloy selectors can be effectively utilized in both synaptic and neuronal components of neuromorphic systems.