Development of Photonic In-Sensor Computing Based on a Mid-Infrared Silicon Waveguide Platform.

Area of Science:

• Photonics and Neuromorphic Engineering
• Mid-Infrared Optoelectronics
• Graphene-based Photodetectors

Background:

• Neuromorphic in-sensor computing offers energy-efficient solutions for smart sensors and on-chip data processing.
• Existing optoelectronic chips primarily focus on free-space configurations for neuromorphic vision.
• There is a need for waveguide-based in-sensor computing capable of handling diverse data modalities.

Purpose of the Study:

• To develop an on-chip waveguide-based in-sensor processing unit for mid-infrared applications.
• To demonstrate the integration of a responsivity-tunable graphene photodetector onto a silicon waveguide.
• To showcase the device's capability in performing neural network tasks using dynamic bias tuning for weighting operations.

Main Methods:

• Integration of a responsivity-tunable graphene photodetector with a silicon waveguide.
• Utilizing dynamic bias tuning of the photodetector for weighting operations, achieving up to 4-bit precision.
• Demonstration of three distinct neural network tasks: image preprocessing, gesture recognition, and gas mixture classification.

Main Results:

• Successful realization of a waveguide-based in-sensor processing unit operating in the mid-infrared range (3.65–3.8 μm).
• Achieved 4-bit weighting precision through photodetector bias tuning.
• Validated the device's performance on image preprocessing, gesture recognition, and spectroscopic data classification.

Conclusions:

• A tunable graphene photodetector integrated onto a silicon waveguide provides a viable weighting solution for photonic in-sensor computing.
• The developed approach extends neuromorphic in-sensor computing to the mid-infrared wavelength range.
• This technology holds potential for large-scale neuromorphic in-sensor computing in photonic integrated circuits for edge applications.