Fully hafnium-based ferroelectric/antiferroelectric devices for hybrid SNN neuromorphic computing

This is the authors' abstract. We don't add key points for this paper.

Neuromorphic computing based on ferroelectric devices has shown great potential in artificial intelligence tasks. However, the challenge of heterogeneous integration between synaptic and neuronal devices has imposed additional limitations on hybrid neural network hardware. In this work, ferroelectric and antiferroelectric devices based on element doping engineering of HfZrOx films are employed to emulate synaptic and neuron functionality for hybrid spiking neural network (SNN) neuromorphic computing. By constructing a functional circuit based on an Hf0.5Zr0.5O2 ferroelectric device, neural characteristics were emulated for SNN computing. Furthermore, the integrate-and-fire dynamics is implemented based on spontaneous depolarization characteristic of a full hafnium-based hybrid ferroelectric neural network, greatly simplifying conventional neuron circuits. The novel neural network architecture achieves a classification accuracy of 92.8% on the Dynamic Vision Sensor dataset, establishing a crucial foundation for the development of high-efficiency neuromorphic computing systems. This work demonstrates considerable promise for realizing hybrid neuromorphic hardware using CMOS-compatible full hafnium-based ferroelectric devices.