Reconfigurable, non-volatile neuromorphic photovoltaics

Area of Science:

• Materials Science
• Neuromorphic Engineering
• Optoelectronics

Background:

• Current neural network image sensors mimic the human retina but lack tunable responsivity and non-volatile data storage.
• These limitations hinder advancements in the compactness and functionality of artificial neural networks for visual perception.

Purpose of the Study:

• To demonstrate a reconfigurable and non-volatile neuromorphic device capable of simultaneous image sensing and processing.
• To achieve highly tunable responsivity and non-volatile storage in a single artificial neural network component.

Main Methods:

• Developed a two-dimensional semiconducting metal sulfide-based metal-semiconductor-metal (MSM) device.
• Utilized pulse-tunable sulfur vacancies at the metal-semiconductor junctions to modulate device properties.
• Investigated bias-induced motion of sulfur vacancies to dynamically alter Schottky barriers and responsivity.

Main Results:

• Demonstrated a photovoltaic detector with reconfigurable and non-volatile characteristics.
• Achieved tunable short-circuit photocurrent polarities and magnitudes by modulating sulfur vacancy concentrations.
• Successfully designed and implemented a convolutional neuromorphic network for image processing and object detection using the developed device.

Conclusions:

• The novel neuromorphic device offers a significant advancement in artificial visual perception hardware.
• Neuromorphic photodetectors are key components for future visual perception systems, enabling compact and functional designs.