Infrared In-Sensor Computing with ReSe₂ Photoelectric Memristor Arrays

Area of Science:

• Materials Science
• Electrical Engineering
• Computer Science

Background:

• Infrared (IR) vision is crucial for sensing in challenging conditions where visible light fails.
• Current IR systems suffer from high data transfer overhead, latency, and energy use due to separated sensing, memory, and computation.
• There is a need for integrated, efficient IR sensing and processing solutions.

Purpose of the Study:

• To develop an infrared in-sensor computing platform.
• To integrate IR sensing, nonvolatile memory, and computation into a single system.
• To demonstrate a practical pathway for compact, low-latency, and energy-efficient intelligent vision hardware.

Main Methods:

• Developed a ferroelectric photoelectric memristor array using a ReSe2 channel and in-plane ferroelectric polarization.
• Engineered devices with highly linear, nonvolatile, and multilevel modulation of self-powered photoresponsivity.
• Utilized a 5x9 photoelectric memristor crossbar array for direct IR image sensing and processing.

Main Results:

• Achieved stable analog photoelectric weight programming with near-infrared sensitivity.
• Demonstrated proof-of-concept IR image recognition with in situ sensing and classification.
• Attained high recognition accuracy and robust classification margins, even with noise, without external computation.

Conclusions:

• The developed platform integrates IR sensing, memory, and computation efficiently.
• This work presents a viable approach for next-generation intelligent vision systems.
• The in-sensor computing platform offers significant advantages in latency and energy consumption for IR applications.