Multi-Modal Low-Power Adaptive Braille Recognition System Based on Hf0.3Zr0.7O2 Memristor.

Tactile-assisted perception systems hold great promise for robotics and human-machine interaction, yet current assistive systems for visually impaired individuals suffer from redundant data transmission, complex fusion algorithms, and high-power consumption. These limitations can introduce millisecond‑scale processing latencies and restrict hardware‑level dynamic adaptability, posing substantial challenges for complex scenarios such as temperature‑induced tactile blurring. Here, we propose a multimodal adaptive perception system based on a Hf0.3Zr0.7O2 (HZO) epitaxial thin-film ferroelectric memristor. The device features excellent ferroelectric properties with a remanent polarization of 2Pr ≈ 36.4 µC/cm2, stable 16-state multilevel resistance, and robust synaptic plasticity for biomimetic emulation of biological learning rules. We developed adaptive sensing units integrating light, temperature, and pressure, which autonomously regulate response in dynamic environments. By encoding Braille digits into pulse sequences and adopting reservoir computing, we achieved a 91.65% maximum recognition accuracy for Braille digits 0-9. The system completes a single analog computation in ∼48 µs with only 8.51 nJ energy consumption, three orders of magnitude lower than conventional electronics. This work provides an efficient hardware strategy for neuromorphic perception and visually impaired assistive devices in complex environments.