Gate-Tunable GeSe/MoSe₂ Heterojunctions for In-Sensor Image Processing.

Area of Science:

• Materials Science
• Optoelectronics
• Device Physics

Background:

• Traditional vision systems face latency and energy issues due to separate sensing and processing units.
• In-sensor computing offers a solution by processing visual data directly in the analog domain.

Purpose of the Study:

• To develop a high-performance, gate-tunable van der Waals heterostructure for simultaneous light sensing and in-sensor computing.
• To address the challenges of integrating nonlinear device physics with linear neural network algorithms for efficient computation.

Main Methods:

• Fabrication of a GeSe/MoSe2 van der Waals heterostructure with engineered Type-II band alignment.
• Development of a physics-based quantitative weight mapping strategy using the zero-gate voltage state as a baseline.
• Demonstration of in-sensor convolution operations for edge detection and image sharpening.

Main Results:

• The GeSe/MoSe2 heterostructure exhibited efficient charge separation, suppressed dark current, and effective gate tunability.
• The proposed weight mapping strategy enabled linear and robust modulation of convolution weights.
• Low-power in-sensor convolution operations were achieved with high fidelity for image processing tasks.

Conclusions:

• GeSe-based heterostructures show significant potential for advanced optoelectronic applications and in-sensor computing.
• The developed methodology provides a robust approach for bridging device physics and neural network algorithms.
• This work paves the way for more efficient and integrated vision systems.