Differential Image Sensor With Decoupled Static and Dynamic Outputs

Area of Science:

• Computer Vision
• Sensor Technology
• Energy-Efficient Systems

Background:

• Machine vision systems require significant energy for full-motion processing.
• Hierarchical architectures offer energy-performance tradeoffs via standby and on-demand activation.
• Decoupling static and dynamic visual information is key for on-demand sensing.

Purpose of the Study:

• To develop a power-efficient and scalable differential image sensor (DIS).
• To enable on-demand activation of vision functions for reduced energy consumption.
• To facilitate advanced motion processing in AIoT applications.

Main Methods:

• Utilized differential photodiodes with decoupled differential and tunneling modes.
• Implemented on-demand activation for static (brightness) and dynamic (motion) data.
• Integrated with conventional Read-Out Integrated Circuits (ROICs) for scalability (e.g., 640 × 512).

Main Results:

• Demonstrated a differential image sensor (DIS) with decoupled static and dynamic modes.
• Achieved on-demand activation for hierarchical motion processing, including sparse motion detection, optical flow, and depth analysis.
• Showcased potential for up-scaled integration and power efficiency.

Conclusions:

• The DIS provides a power-efficient and scalable strategy for machine vision.
• Enables optimized energy-performance tradeoffs for AIoT applications.
• Facilitates advanced, low-power motion analysis in vision systems.