Non-pixelated in-materia retinomorphic sensor via photocarrier dynamics for precise spatiotemporal perception

Area of Science:

• Materials Science
• Sensor Technology
• Computational Neuroscience

Background:

• Conventional image sensors struggle with dynamic visual scenes due to inefficient spatiotemporal feature extraction.
• Inter-pixel correlations are missed, causing high data transfer, power consumption, and latency.

Purpose of the Study:

• To develop a non-pixelated sensor for efficient real-time visual information processing.
• To leverage intrinsic material properties for in-sensor spatiotemporal perception.
• To overcome limitations of conventional image sensors in dynamic scene analysis.

Main Methods:

• A non-pixelated in-materia retinomorphic sensor (IMRS) was developed using a graphene/silicon heterostructure.
• Circumferentially arranged electrodes harness the lateral photovoltaic effect for optical pattern conversion.
• Spatiotemporal photocarrier dynamics and lateral inhibition principles are utilized for perception.

Main Results:

• The IMRS achieves over 98% accuracy in human motion recognition.
• Raw visual data is compressed from 10,000 to 48 bytes.
• Postprocessing network parameters are reduced by two orders of magnitude.

Conclusions:

• Spatiotemporal photocarrier dynamics in heterostructures serve as a computational primitive.
• The IMRS enables energy-efficient, ultralow-latency processing of high-dimensional spatiotemporal information.
• This approach offers a paradigm shift from conventional image sensing and processing.