Ultralow‐Power Optoelectronic Organic‐Inorganic Heterostructure Synapses for Monolithic Sensing‐Memory‐Computing Integrated Autonomous Driving Vision

ABSTRACT Autonomous driving requires high‐precision compact integrated visual processing systems, while conventional vision architectures suffer from bulky discrete configuration, single‐function operation, and separated sensing, memory and computing units. To address these bottlenecks for low‐power, highly‐integrated edge vision hardware, a multifunctional artificial synaptic array based on a layered organic–inorganic heterostructure is developed, achieving monolithic tri‐modal sensing‐memory‐computing integration and electro‐optic dual responsiveness. The device mechanism relies on humidity‐modulated hydronium ion migration in the poly (amic acid) layer and charge trapping in HfO 2 , which regulates dynamic electron‑hydronium charge‐storage competition to realize reversible bidirectional non‐volatile memory. The devices deliver a carrier mobility of 11.89 cm 2 V −1 s −1 , over 10 4 s retention (extrapolated 10‐year lifetime) and stable 5000‐cycle endurance. They effectively emulate biological synaptic plasticity with an ultralow energy consumption of 2.86 aJ per spike. Two complementary neuromorphic networks are built for autonomous driving urban scene perception, achieving semantic segmentation (mean Intersection over Union (mIoU) > 82%) and monocular depth estimation (Absolute Relative (Abs Rel) error < 0.08). This work provides a high‐integration, ultralow‐power hardware solution for intelligent autonomous visual perception, overcoming traditional bionic optoelectronic device limitations and paving a path for next‐generation multimodal neuromorphic electronics.