As artificial vision systems evolve, bridging the gap between sensing and processing remains a key challenge. Now, researchers from the University of Science and Technology of China, led by Prof. Yong Yan and Prof. Haiding Sun, have developed a reconfigurable bioinspired vision sensor using GaN/AlN quantum-disk-in-nanowires (QD-NWs) that emulates the human retina’s dual-cell system—delivering in-sensor computing for high-accuracy human action recognition (HAR).
Why This Bioinspired Sensor Matters
- Dual-Mode Operation: Mimics Parvo cells (slow, high-contrast vision) and Magno cells (fast, motion-sensitive vision) via voltage-tunable persistent photocurrent (PPC).
- In-Sensor Computing: Combines image enhancement and reservoir computing in a single device, reducing latency and power consumption.
- High Accuracy: Boosts HAR accuracy from 51.4% to 81.4% through synergistic integration of both photoresponse modes.
Innovative Design and Features
- Quantum-Confined Stark Effect (QCSE): Enables bias-tunable control over carrier recombination, switching between long-term and short-term PPC.
- Nanowire Architecture: Ultrathin GaN/AlN QD-NWs grown on Si substrates offer CMOS compatibility, strain relaxation, and strong optoelectronic tunability.
- Reservoir Computing System: Uses short-term PPC for temporal feature extraction and long-term PPC for image denoising and enhancement.
Applications and Performance
- Image Enhancement: Under negative bias, the sensor enhances image contrast in real time—improving SNR from 1/0.3 to 1/0.15 without external processing.
- Human Action Recognition: Under positive bias, the sensor acts as a hardware-based reservoir, classifying 10 human actions from the Weizmann dataset with >95% accuracy.
- Robustness: Maintains >90% recognition accuracy even under 50% device noise, outperforming software-only approaches.
Conclusion and Outlook
This work introduces a compact, intelligent vision sensor that unites biological inspiration with semiconductor engineering, enabling real-time, low-power, high-accuracy visual perception. The QD-NW platform opens new pathways for neuromorphic vision systems, edge AI, and smart surveillance applications.
Stay tuned for more breakthroughs from Prof. Yong Yan and Prof. Haiding Sun’s team at USTC!