Reconfigurable Vertical Schottky Photodiodes Based on Ferroelectric 2D Semiconductors for Perceptual-Precision, Context-Aware Neuromorphic Vision.

Abstract

To realize in-memory sensing and computing platforms, it is essential to integrate sensing, computation, and memory functionalities within a single device, enabling energy- and time-efficient vision systems with high perceptual precision. However, achieving such multi-functional processing capability within a compact device structure remains a major challenge. Here, a reconfigurable vertical photodiode based on α-In₂Se₃ is presented, a ferroelectric 2D semiconductor. Gradual and reversible modulation of built-in electric fields at the top and bottom Schottky junctions is achieved through partial out-of-plane polarization switching of α-In₂Se₃, enabling multi-level, non-volatile, and polarity-tunable photoresponsivity. This allows analog programmability with a high degree of freedom in processing within a two-terminal metal-ferroelectric semiconductor-metal (MFsM) structure, effectively resolving trade-off between structural compactness and computing versatility. Leveraging these intrinsic characteristics, we demonstrate a versatile sensor-level perceptual processing framework using a reconfigurable photodiode crossbar array. By exploiting the high-density spatial integration capability, the system adaptively configures its spatial support to prioritize either suppressing environmental noise for robust feature extraction or preserving fine-grained details for precise classification, depending on the task requirements. These results lay the foundation for highly scalable and energy-efficient neuromorphic vision system with high perceptual precision.