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

The rising demand for high-performance, energy-efficient neuromorphic systems has driven the exploration of chalcogenide materials with tunable electronic properties. Here, Ag/GeSe/FTO resistive-switching devices are demonstrated that exhibit Pavlovian associative learning, mimicking classical conditioning through paired electrical stimuli and voltage-dependent adaptation. Amorphous and crystalline GeSe thin films deposited by RF sputtering were systematically compared. Structural analysis confirmed that annealing at 375 °C induces phase crystallization with improved grain integrity and stoichiometric uniformity. Crystalline devices show superior performance, featuring lower SET/RESET voltages, larger hysteresis windows, endurance exceeding 4000 cycles, and retention over 1500 s. Conduction analysis reveals a transition from ohmic to a space-charge-limited transport-mediated mechanism in crystalline GeSe, enabling controlled Ag⁺ ion migration along grain boundaries, leading to confined filament formation and stable rupture dynamics, which significantly enhance switching uniformity and analog synaptic behavior. Neuromorphic behavior includes stable potentiation, depression, paired-pulse facilitation, and inhibition of long-term potentiation, all demonstrating adaptive and nonlinearly tunable learning. Associative learning, analogous to classical conditioning, was successfully reproduced across multiple cycles. Furthermore, an artificial neural network (ANN) implemented using experimentally extracted long-term potentiation/depression characteristics achieved classification accuracies of ∼73% and ∼87% for the Fashion-MNIST and digit-MNIST data sets at 0.9 V, improving to ∼75% and ∼87% at 1.1 V. These results firmly establish crystalline GeSe as a promising, stable, and cost-effective memristive material for energy-efficient neuromorphic and artificial-intelligence hardware applications.