Ferroelectric-Configured In-Sensor Dynamic Computing with 2D Perovskites for Dim Object Recognition

Area of Science:

• Materials Science
• Optoelectronics
• Artificial Intelligence

Background:

• Machine vision systems struggle with feature extraction from dim objects in complex environments.
• Existing optoelectronic methods often lack the dynamic adaptability required for challenging visual tasks.

Purpose of the Study:

• To develop a reconfigurable photovoltaic device array for in-sensor dynamic computing.
• To enhance the robustness and accuracy of dim object detection in machine vision.

Main Methods:

• Direct synthesis of 2D perovskite ferroelectric nanoplates.
• Fabrication of graphene/ferroelectric/graphene heterostructures for reconfigurable photovoltaics.
• Design of a convolution kernel optoelectronic sensor array leveraging ferroelectric polarization.

Main Results:

• Demonstrated switchable ferroelectric polarization for nonvolatile photovoltaic modulation.
• Achieved selective amplification of local image pixel differences for edge feature extraction.
• Significantly improved dim object detection accuracy and robustness when integrated with a convolutional neural network.

Conclusions:

• The ferroelectric-configured photovoltaic device array offers a promising platform for advanced machine vision.
• In-sensor dynamic computing enhances machine vision capabilities for low-contrast and dim object recognition.
• This approach overcomes limitations of traditional static optoelectronic processing.