Synapse with Diverse Plasticity in Ferroelectric BaTiO₃ Thin Films for Neuromorphic Computing

Abstract

Synapse-based artificial neural networks (ANNs) are hopeful in overcoming the von Neumann bottleneck since they can process and store data simultaneously. Here, we present an artificial synaptic device based on a ferroelectric BaTiO3 thin film with a robust weight update and diverse plasticity for ANNs. Specifically, the potentiation and depression effects strongly depend on the spike polarity, amplitude, number, and rate. Moreover, four types of spike timing-dependent plasticities (STDP) and two types of Bienenstock–Cooper–Munro (BCM) learning rules with sliding frequency thresholds are obtained. For BCM learning rules, a normal one with potentiation at a high frequency and depression at a low frequency is obtained under a positive bias and an abnormal one with depression at a high frequency and potentiation at a low frequency is achieved at a negative bias. Furthermore, an ANN is enabled with a recognition accuracy of 92.18%. These results are essential for potential applications of ferroelectric artificial synapses for ANNs.