Bio-Inspired Multimodal Hardware Front-End Enabled by 2D Floating-Gate Memory for UAV Perception

Reliable environmental perception for small autonomous unmanned aerial vehicles (UAVs) remains challenging under rapid ego-motion, visual blind regions, and aerodynamic disturbances. Inspired by birds’ efficient sensing-to-computing pathways, we design a multimodal joint-modulation hardware system in which a 2D floating-gate (FG) memory serves as the computing core, integrating visual, inertial, and wind-field cues to enable fast and stable tracking and obstacle avoidance in dynamic environments. We develop a MoS₂/h-BN/graphene FG device that provides stable multilevel conductance states, an on/off ratio above 10⁸, sub-10 µs switching, long retention, and high device uniformity. A 4 × 4 FG-memory array robustly encodes temporal visual variations for real-time target tracking, while a single FG device acts as an airflow neuron that rapidly detects UAV-induced airflow in visual blind regions. An inertial-information-driven adaptive threshold modulation scheme further stabilizes both pathways under rapid ego-motion, enabling bird-like tracking and avoidance. Experiments show that visual processing latency is ∼7 ms, the average tracking center offset rate is 11.5%, background drift suppression exceeds 80%, and airflow disturbances trigger avoidance within 2 ms. These results demonstrate that the proposed system significantly improves signal-processing speed and robustness, enhancing UAV applicability in unstructured environments.