High-Fidelity and Low-Latency Machine Vision via Manipulation of Contact Potential Behaviors at Perovskite Hetero-Interfaces.

Abstract

Artificial intelligence (AI)-driven machine vision is reshaping the industrial perception paradigm, highlighting the critical demands of high-performance vision hardware. Among various emerging vision techniques, perovskite photodetectors have demonstrated exceptional promise but faced a long-term trade-off between external quantum efficiency (EQE) and response time, posing a significant challenge for high-fidelity and low-latency vision applications. Here, the transport-limited mechanism is confirmed to fundamentally dominates response time, and high-bandwidth and high-sensitivity perovskite photodiodes are demonstrated by manipulating contact potential behaviors through polymer-polarized passive dipoles. Specifically, the synergistic incorporation of dipole-rich molecular modifiers with polar functional groups and electronegative atoms effectively rearranges charge distribution, leading to an increased potential difference and homogeneous potential profiles at perovskite hetero-interfaces. Accordingly, reduced conduction and valence band offsets facilitate carrier transport and suppress non-radiative recombination. The device simultaneously delivers an impressive bandwidth of 25.7 MHz and a high EQE exceeding 88%, enabling a high-fidelity and low-latency imaging of fast rotating motion (≈18 750 rpm) with negligible ghosting. This work provides an effective strategy to approach the transport limit of vision hardware, laying a solid foundation for next-generation AI scenarios.