High-speed and multi-level AlScN ferroelectric memristor on CMOS-compatible tungsten for electronic synapses

Abstract

The AlN doped with Sc has a high residual polarization value, and good ferroelectricity has become a research hotspot. However, traditional growth of AlScN typically requires substrates such as sapphire, GaN, or SiC, which have complex fabrication processes and hinder their integration and application with mainstream complementary metal oxide semiconductor (CMOS) technologies. Here, we demonstrated ferroelectric Al0.8Sc0.2N grown on CMOS-compatible tungsten metal and further exhibited its memristor characteristics suitable for artificial electronic synapses. This memristor features a programmable multi-level configuration, excellent retention (>104 s) and nanosecond (∼36 ns) ultrafast opening speed, which can enable high-speed and multi-level storage and computation. In addition, biological synaptic plasticity, including spike timing dependent plasticity and the human brain's learning, forgetting, and relearning behavior, was simulated by electrical pulse modulation. Finally, we built an artificial neural network to recognize handwritten digits in the Modified National Institute of Standards and Technology database, achieving 93.9% recognition accuracy. This study provides a feasible approach for integrating group III-nitride (AlScN) ferroelectric materials with mature CMOS technology and paves the way for the application of AlScN ferroelectric memristors in the next generation of artificial electronic synapse devices.