Amphibian-inspired neuromorphic dynamic vision systems based on ferroelectric field-effect transistor

Area of Science:

• Bio-inspired engineering
• Neuromorphic computing
• Advanced sensor technology

Background:

• Current dynamic vision sensors offer high-speed imaging but lack color sensitivity and efficient data transfer.
• Amphibian retinas process visual information hierarchically, enabling efficient and robust perception.
• Ferroelectric field-effect transistors (FeFETs) show potential for novel electronic applications.

Purpose of the Study:

• To develop an amphibian-inspired dynamic vision system (ADVS) using ferroelectric transistors.
• To emulate the spectral perception, spatial preprocessing, and neural encoding functions of amphibian retinas.
• To overcome the limitations of existing dynamic vision sensors.

Main Methods:

• Fabrication of an ADVS based on ferroelectric field-effect transistors.
• Characterization of broadband photosensitivity and bidirectional photoresponses of the transistors.
• Implementation of device arrays for center-surround receptive-field processing.
• Integration with a bioinspired hierarchical preprocessing framework and an event-driven convolutional neural network.

Main Results:

• Ferroelectric transistors demonstrated broadband photosensitivity (365-637 nm) and bidirectional responses for multichannel spectral recognition.
• Device arrays replicated center-surround receptive-field processing, enhancing contrast and reducing noise under low light.
• The system achieved microsecond-scale event-driven spiking responses due to steep switching characteristics (SSmin = 53.8 mV dec−1).
• The ADVS achieved 96.5% accuracy in dynamic facial expression recognition and real-time multi-agent trajectory prediction with <5% error.

Conclusions:

• The amphibian-inspired dynamic vision system effectively emulates retinal functions using ferroelectric transistors.
• This bio-inspired approach significantly enhances dynamic vision capabilities, including spectral perception and spatial processing.
• The ADVS offers a promising platform for advanced, high-performance neuromorphic vision applications.