CMOS-Compatible LiNbO₃ Domain-Wall Entropy Engine for Ultrafast True Random Number Generation.

Abstract

Random ferroelectric domain nucleation and growth lead to the generation of numerous unpredictable microscopic states that collectively form a natural high-entropy system. Conventional electrical methods can directly measure reversible domain switching currents, offering a viable platform for true random number generation (TRNG). However, TRNG based on random ferroelectric switching events is limited by the noise amplitudes in electrical signals. In this study, TRNG is realized via the stochastic formation of conductive domain walls in single-crystal LiNbO₃ thin films bonded to SiO₂/Si wafers. This approach achieves a noise amplitude and cycling endurance >500 nA and >10¹⁰, respectively. The interfacial-layer-based device exhibits self-reinitialized stochastic sub-10-ns domain switching operations, enabling ultrafast generation of bit outputs and flexible device scaling. The generated random bitstreams, validated via National Institute of Standards and Technology (NIST)tests, exhibit robust resistance against machine-learning-based predictive attacks. This pioneering study establishes ferroelectric conductive domain walls as groundbreaking platforms for CMOS-compatible entropy source extraction, effectively addressing the long-standing challenges in amplifying entropy signals with operational robustness.