Defect-engineered boron nitride memristors with enhanced uniformity and stability for CNN-based image encryption and...

Hexagonal boron nitride (h-BN) memristors hold great promise for resistive-switching applications but often show poor device-to-device uniformity. In this work, a defect-engineering strategy using oxygen plasma treatment is employed to intentionally introduce oxygen-substitutional doping along with boron and nitrogen vacancies. The treated h-BN memristors show significantly improved switching uniformity, with a coefficient of variation for the SET voltage below 6.12% for all devices, and stable resistive switching performance maintained up to 613 K. These devices can generate true random keys from the intrinsic physical entropy of the low-resistance state (LRS) for image encryption applications. Validation with a convolutional neural network (CNN) shows that encrypted images become unrecognizable (recognition accuracy of ∼10%), while decrypted images regain high recognition accuracy. Oxygen plasma treatment introduces a large number of uniform defects in h-BN, facilitating the stable formation of conductive filaments. This work offers an effective defect-engineering method to boost the performance of h-BN memristors and highlights their potential for high-temperature and hardware security applications.