A Near-Infrared Retinomorphic Device with High Dimensionality Reservoir Expression

Area of Science:

• Optoelectronics
• Artificial Intelligence
• Materials Science

Background:

• Reservoir computing (RC) systems offer efficient computation but face limitations in infrared (IR) machine vision.
• Existing physical RC systems struggle with material expression and computational power for IR spectrum analysis.

Purpose of the Study:

• To develop a near-infrared (NIR) retinomorphic device for simultaneous perception and encoding of narrow IR spectral information.
• To enhance the capabilities of physical reservoir computing for IR machine vision applications.

Main Methods:

• Fabrication of a device using core-shell upconversion nanoparticle/poly(3-hexylthiophene) (P3HT) nanocomposite channels.
• Utilizing photon-electron-coupled dynamics, photovoltaic, and photogating effects for NIR absorption and conversion.
• Implementing the device within a reservoir computing framework for image recognition and complex computations.

Main Results:

• The device successfully absorbs and converts NIR light (≈980 nm) to excite photo carriers in P3HT.
• Demonstrated multilevel data storage (≥8 levels), stability (≥2000 s), and durability (≥100 cycles).
• Achieved high recognition accuracies for NIR static (91.13%) and dynamic (90.07%) handwritten digits and solved nonlinear equations with minimal error (NMSE of 1.06 × 10⁻³).

Conclusions:

• The proposed NIR retinomorphic device significantly advances physical reservoir computing for IR machine vision.
• The device offers a promising platform for intelligent perception tasks requiring efficient processing of NIR spectral information.
• This work addresses limitations in IR machine vision by providing a robust and high-performance optoelectronic computing solution.