A Monolithic Ferroelectric-Ionic Duality for Stochastic-Neuromorphic Core Integration

Abstract

Learning under uncertainty has become increasingly critical in data-intensive artificial intelligence applications, requiring computing systems that unify deterministic and probabilistic functions. Hardware that combines stable memory with tunable stochasticity is essential for such systems, but achieving this integration within a miniaturized architecture remains significantly challenging due to intrinsic conflicts between precise memory retention and controllable stochastic variability. Here we demonstrate a hafnia-based ferroelectric-ionic duality that integrates stochastic encoding and synaptic memory within a single device. By deliberately engineering the ferroelectric interface to repurpose oxygen vacancies—traditionally regarded as defects that degrade reliability in hafnia ferroelectrics—we exploit these vacancies as functional ionic components that dynamically modulate device behavior. The resulting ferroelectric-ionic dual-mode switching introduces voltage-tunable stochasticity and enhances synaptic behavior within a single device architecture. Crucially, this ferroelectric-ionic duality exhibits full complementary metal-oxide-semiconductor (CMOS) compatibility and scalability to very-large scale integration (VLSI) system, enabled by wafer-scale atomic layer deposition-based growth of hafnia. These results establish a novel device paradigm that unifies memory, randomness, and learning capabilities within a single ferroelectric platform.