Owl-vision-inspired near sensor computing

Area of Science:

• Neuromorphic Engineering
• Materials Science
• Artificial Intelligence

Background:

• Passive target detection is vital for machine vision in low-light conditions, enabling applications like surveillance and guidance.
• Existing technologies struggle with sensitivity and adaptation in photon-starved environments.
• Mimicking biological vision systems, like that of owls, offers a promising approach for enhanced performance.

Purpose of the Study:

• To develop a bimodal synaptic transistor inspired by owl vision.
• To achieve optoelectronic decoupling for parallel photonic perception and electrical plasticity emulation.
• To enable energy-efficient, low-light image processing for neuromorphic vision sensors.

Main Methods:

• Fabrication of a bimodal synaptic transistor with an optoelectronic decoupling mechanism.
• Characterization of the device's performance in perceiving low light intensities.
• Demonstration of synaptic weight modulation for artificial neural network applications.
• Testing adaptive contrast enhancement for image processing.

Main Results:

• The device achieved a high active adaptation index of approximately 331.
• It successfully detected light intensities as low as 0.146 nW cm^-2.
• Cyclically stable synaptic weight modulation (enhancement and inhibition) was demonstrated.
• Feasible weight deployment across three artificial neural levels was verified over a specific light intensity range.

Conclusions:

• The developed owl-vision-inspired device provides a hardware foundation for advanced neuromorphic vision sensors.
• It enables efficient and effective low-light image processing.
• This technology has significant potential for applications requiring high sensitivity and adaptation in challenging visual environments.