Compact artificial neuron based on anti-ferroelectric transistor

Abstract Neuromorphic machines based on spiking neural networks (SNNs) provide a more fascinating platform over traditional computers on building energy-efficient intelligent systems, in which spiking neuron are pivotal components. Recently, memristive neurons, with promising bio-plausibility and density, have been developed, but with limited reliability or bulky capacitors for integration or additional circuits for reset. Here, we propose a novel anti-ferroelectric field-effect transistor (AFeFET) neuron based on the inherent polarization and depolarization of Hf0.2Zr0.8O2 anti-ferroelectric film to meet these challenges. In this neuron, the intrinsic polarization accumulation effect in the Hf0.2Zr0.8O2 film increases the channel current of AFeFET gradually, which implements the integration feature of the neuronal membrane and avoids using external capacitors. Also, the spontaneous depolarization effect in AFeFET emulates the leaky behavior of neurons, saving the hardware overhead of neuron circuits by getting rid of external reset circuits. Moreover, the AFeFET neuron exhibits other comprehensive merits, such as low energy consumption (37 fJ/spike), excellent endurance (>1012), high uniformity and high stability. Using such an AFeFET neuron, we further construct a two-layer fully ferroelectric (784×400×10) SNN combining established FeFET synapse, achieving 96.8% recognition accuracy on MNIST datasets. This work opens the way to emulate spiking neurons with anti-ferroelectric materials and provides a more competitive approach to build high efficient neuromorphic hardware systems.