Bidirectional Photoadaptive Organic Heterojunction Synaptic Transistors for Accurate Image Recognition in Machine...

Area of Science:

• Neuromorphic Engineering
• Materials Science
• Computer Vision

Background:

• Machine vision systems require effective photoadaptation in photosensors.
• Existing synaptic phototransistors often lack bidirectional photoresponses, limiting performance in complex lighting.
• Biological vision inspires neuromorphic solutions for enhanced image acquisition.

Purpose of the Study:

• To develop bidirectional photoadaptable organic heterojunction synapse phototransistors.
• To improve image contrast and feature extraction under adverse lighting conditions.
• To enhance image recognition accuracy in neuromorphic systems.

Main Methods:

• Fabrication of n-n and n-p heterojunction synapse phototransistors.
• Mimicking bidirectional plasticity transformation of biomimetic neuromorphic synapses.
• Integrating device characteristics into convolutional neural networks (CNNs) with optimized algorithms.

Main Results:

• n-p heterojunctions demonstrated a bidirectional photoresponse with a prominent inhibitory postsynaptic current.
• The developed devices facilitated image contrast enhancement and feature extraction.
• Integration with CNNs increased image recognition accuracy to 97.4% within ten cycles, enhancing details and edges of low-contrast images.

Conclusions:

• Bidirectional photoadaptable organic heterojunction synapse phototransistors offer a novel strategy for high-performance neuromorphic visual systems.
• These devices show promise for in-sensor computing applications, particularly under adverse lighting.
• The study highlights the potential of mimicking biological vision for advanced machine perception.