Unipolar ferroelectric synapses with subfemtojoule energy consumption

Abstract

Neuromorphic computing demands artificial synapses that combine ultra-low energy consumption with simple modulation schemes. Here, we demonstrate unipolar ferroelectric tunnel junction synapse based on an Au/Cr/HZO/LSMO/STO heterostructure, in which both excitatory and inhibitory plasticity are realized solely by tuning the pulse width, without reversing voltage polarity. This unipolar operation further enables emulation of the Bienenstock–Cooper–Munro learning rule with an enhanced depression effect as well as perceptual functions such as edge detection. Pulse-width-modulated long-term potentiation was leveraged to experimentally emulate Pavlovian conditioning. Moreover, continuous long-term potentiation and long-term depression were achieved by alternately varying the pulse width and were applied to motion detection. This work provides a new pathway for implementing diverse neuromorphic functionalities through simple pulse-width modulation, highlighting the potential of unipolar ferroelectric synapses for efficient in-memory computing systems.