High-stable multifunctional dynamically reconfigurable artificial synapses based on hybrid graphene/ferroelectric field-effect transistors

Abstract

In response to the challenges posed by traditional computing architectures in handling big data and AI demands, neuromorphic computing has emerged as a promising alternative inspired by the brain's efficiency. This study focuses on three-terminal synaptic transistors utilizing graphene and P(VDF-TrFE) to achieve dynamic reconfigurability between excitatory and inhibitory response modes, which are crucial for mimicking biological functions. The devices operate by applying different top gate spikes (±25 V and ±10 V) to modulate the polarization degree of P(VDF-TrFE), thereby regulating the carrier type and concentration in the graphene channel. This results in the effective realization of enhancement and inhibition processes in two neural-like states: excitatory and inhibitory modes, accompanied by good neural plasticity with paired-pulse facilitation and spike-time-dependent plasticity. With these features, the synaptic devices achieve brain-like memory enhancement and human-like perception functions, exhibiting excellent stability, durability over 1000 cycles, and a long retention period exceeding 10 years. Additionally, the performance of the artificial neural network is evaluated for handwritten digit recognition, achieving a high recognition accuracy of 92.28%. Our study showcases the development of highly stable, dynamically reconfigurable artificial synaptic transistors capable of emulating complex neural functions, providing a foundation for emerging neuromorphic computing systems and AI technologies.