Crystallization-Driven Stable Resistive Switching and Reproducible Synaptic Learning in GeSe-Based Artificial...

Area of Science:

• Materials Science
• Condensed Matter Physics
• Neuroscience

Background:

• The demand for high-performance, energy-efficient neuromorphic systems is increasing.
• Chalcogenide materials are being explored for their tunable electronic properties in neuromorphic applications.

Purpose of the Study:

• To demonstrate Ag/GeSe/FTO resistive-switching devices capable of Pavlovian associative learning.
• To compare the performance of amorphous and crystalline GeSe thin films for memristive applications.
• To establish crystalline GeSe as a viable material for artificial intelligence hardware.

Main Methods:

• RF sputtering deposition of GeSe thin films.
• Annealing to induce phase crystallization.
• Structural and electrical characterization of resistive-switching devices.
• Analysis of conduction mechanisms and ion migration.
• Implementation of an artificial neural network (ANN) using device characteristics.

Main Results:

• Crystalline GeSe devices exhibit superior performance: lower SET/RESET voltages, larger hysteresis, high endurance (>4000 cycles), and long retention (>1500 s).
• Conduction transitions to space-charge-limited transport, enabling controlled Ag+ ion migration and stable filament formation.
• Devices demonstrate stable potentiation, depression, paired-pulse facilitation, and inhibition of long-term potentiation, mimicking synaptic behavior.
• Successful reproduction of associative learning and ANN classification accuracies up to ~87% on MNIST datasets.

Conclusions:

• Crystalline GeSe is a promising, stable, and cost-effective memristive material for energy-efficient neuromorphic and AI hardware.
• The demonstrated associative learning capabilities highlight the potential for advanced computing paradigms.
• The findings support the use of GeSe in next-generation artificial intelligence systems.