Neuromorphic Transistors Integrating Photo-Sensor, Optical Memory and Visual Synapses for Artificial Vision...

Area of Science:

• Materials Science
• Neuromorphic Engineering
• Artificial Intelligence

Background:

• Commercial artificial vision systems (AVS) face challenges with separate sensing, storage, and computing units, leading to increased volume, complexity, and energy loss.
• Current AVS architectures limit integration and scalability due to performance gaps between distinct modules.

Purpose of the Study:

• To develop a single neuromorphic transistor capable of integrating photo-sensing, optical memory, and visual synapse functionalities.
• To overcome the limitations of conventional AVS by creating a multi-mode device for enhanced efficiency and reduced complexity.

Main Methods:

• Development of a novel neuromorphic transistor utilizing a gate-tunable out-of-plane electric field for multi-mode operation.
• Characterization of the device's photo-sensing capabilities, non-volatile optical memory, and visual synapse functions under varying gate voltages.
• Integration of the transistor's synaptic plasticity with an artificial neural network (ANN) for image recognition tasks.

Main Results:

• The device demonstrated high-sensitivity photo-response (responsivity ≈6.515 kA W^-1, detectivity ≈3.92 × 10¹⁴ Jones) and non-volatile multi-level optical memory (>4 bits, endurance >10,000 s, ratio 10⁶).
• In visual synapse mode, the transistor exhibited neuromorphic computing capabilities, enabling complex biological learning and synaptic plasticity.
• The integrated system achieved precise image recognition and classification with up to 95.26% accuracy using an ANN.

Conclusions:

• A multi-mode transistor successfully integrates key artificial vision system components, addressing challenges in all-in-one integration and manufacturing.
• This novel device offers a pathway towards more efficient, compact, and powerful neuromorphic computing and artificial vision applications.