Tunable Bipolar Photothermoelectric Response from Mott Activation for In-Sensor Image Preprocessing.

Area of Science:

• Materials Science
• Edge Computing
• Optoelectronics

Background:

• Conventional digital hardware faces limitations due to frequent analog-to-digital conversions and the von Neumann bottleneck.
• There is a need for efficient, large-scale integrated in-sensor devices for high-density, low-power sensory processing at the edge.
• In-sensor image preprocessing is crucial for overcoming current hardware inefficiencies.

Purpose of the Study:

• To introduce an adjustable broadband photodetector for in-sensor image preprocessing.
• To demonstrate a gate-tunable phase-change vanadium dioxide thin-film transistor for enhanced sensory processing.
• To develop wafer-scale integrated systems for advanced edge AI applications.

Main Methods:

• Fabrication of a three-terminal vanadium dioxide/gallium nitride phototransistor utilizing a gate-tunable phase transition.
• Characterization of the device's broadband photoresponse, responsivity modulation, and photocurrent polarity control.
• Development of a wafer-scale bipolar phototransistor array using standard micro-/nano-fabrication techniques.

Main Results:

• The phototransistor exhibits tunable photothermoelectric responsivities and adjustable photocurrent polarities.
• Devices show linear gate dependence for broadband photoresponse and light-intensity dependence for photoresponsivity (positive and negative).
• Achieved ultra-low energy consumption (8 pJ per spike) and demonstrated wafer-scale array stability (>5000 cycles).

Conclusions:

• The developed vanadium dioxide phototransistor offers a significant advancement in low-power, high-density in-sensor processing.
• An integrated in-sensor convolutional network demonstrated successful broadband image classification, medical image denoising, and retinal vessel segmentation.
• This technology paves the way for next-generation smart edge sensors with enhanced capabilities.