Black Phosphorus/Ferroelectric CuInP 2 S 6 Heterostructures Empowering Nonvolatile Photonic Convolutional Computing

ABSTRACT The growing demands of artificial intelligence require new energy‐efficient and nonvolatile computing paradigms. To meet this challenge, we demonstrate a foundational device platform for optical fiber computing that targets the signal decay and power consumption bottlenecks of conventional systems. Our architecture is enabled by a novel electro‐optic modulator that integrates a 2D van der Waals ferroelectric heterostructure of CuInP 2 S 6 (CIPS) and black phosphorus(BP) with a microfiber knot resonator(MKR). We leverage the robust room‐temperature ferroelectricity of CIPS for 8‐level nonvolatile weight storage and the strong light‐matter interaction of BP for efficient optical modulation. This monolithic all‐fiber design completely eliminates chip‐to‐fiber coupling losses, enabling seamless and highly efficient optical data processing. Furthermore, femtosecond transient reflectance spectroscopy reveals picosecond‐scale carrier dynamics, validating the capability for Gigahertz bandwidth computation. Validation accuracy on MNIST and Fashion‐MNIST datasets yielded high classification accuracies of 96.34% and 92.68%, respectively. This work establishes BP as the promising solution for energy‐efficient fiber photonic computing, empowering AI‐ready neuromorphic systems.