Event-Driven Sentinel-Expert Neuromorphic Vision Enabled by Polarization-Reconfigurable Organic Ferroelectric...

Area of Science:

• Materials Science
• Neuroscience
• Computer Engineering

Background:

• Conventional always-on machine vision systems face energy limitations due to continuous processing of redundant data.
• The human retina's dual-pathway strategy offers an efficient model for visual information processing.

Purpose of the Study:

• To develop a bioinspired, energy-efficient vision system for the Internet of Things and edge intelligence.
• To leverage polarization-reconfigurable organic ferroelectric phototransistors for adaptable visual processing.

Main Methods:

• A novel "Sentinel-Expert" synergetic vision system was designed using P(VDF-TrFE) ferroelectric phototransistors.
• The system utilizes ferroelectric polarization states to switch between Sentinel (motion detection) and Expert (recognition) modes.
• Physical reservoir computing and synaptic weight updates were implemented for system functionality.

Main Results:

• The Sentinel mode achieved 98.46% recognition accuracy, mimicking magnocellular pathway dynamics for motion detection.
• The Expert mode achieved 97.45% recognition accuracy, simulating parvocellular pathway for static recognition.
• An event-driven strategy reduced computational cost by ~50x at a 1% duty cycle compared to always-on systems.

Conclusions:

• The proposed bioinspired system offers a significant advancement in energy-efficient machine vision for edge intelligence.
• Reconfigurable ferroelectric phototransistors enable adaptable, dual-mode visual processing, reducing computational load.
• This technology holds promise for low-power, high-performance sensing in IoT devices.