Ultrathin TiO2-interfaced hafnia ferroelectric transistor for large-scale neuromorphic computing

Abstract

The growing demand for large-scale neuromorphic computing necessitates the development of innovative memory devices capable of supporting high-density synaptic arrays with frequent, low-power weight updates. Among the promising candidates, hafnia-based ferroelectric field-effect transistors (FeFETs) have emerged due to their low-power switching and CMOS compatibility. However, conventional hafnia FeFETs are limited by their poor endurance and switching dynamics-both of which are attributed to the degradation mechanisms arising from the ferroelectric/dielectric interface-impeding the realization of large-scale neuromorphic computing. Herein, we propose a synergistic ferroelectric polarization-interface dipole modulation (IDM) switching in hafnium-zirconium oxide (HZO) FeFETs to improve switching dynamics and endurance. Integration of an ultrathin (< 0.5 nm) TiO2 layer into the gate stack has three critical functions: (i) reducing the oxygen vacancies in HZO; (ii) mitigating trapping at the ferroelectric/dielectric interface; and (iii) improving the switching dynamics through the polarization coupling effect via IDM. Consequently, this synergistic improvement significantly enhances the FeFET performance with 10(6)-fold endurance enhancement. Moreover, by demonstrating large-scale neuromorphic integration that meets the update demands required for CIFAR-100 dataset, our work underscores the transformative potential of this approach for realizing reliable and energy-efficient systems capable of real-time learning.