Biomimetic ferroelectric-semiconductor transistor enables neuronal multisensory integration.

Area of Science:

• Materials Science
• Neuroscience
• Artificial Intelligence

Background:

• Human brains seamlessly integrate multisensory stimuli for enhanced perception using principles like superadditivity and temporal congruency.
• Replicating multisensory integration in AI is challenging due to inefficient algorithms and lack of hardware mechanisms.

Purpose of the Study:

• To demonstrate biomimetic audiovisual integration at the device level using Bi2O2Se ferroelectric-semiconductor field-effect transistors (FeS-FETs).
• To develop a physics-aware framework for neuromorphic multisensory intelligences.

Main Methods:

• Utilized multiphysics coupling in FeS-FETs for audiovisual integration.
• Configured FeS-FETs into memristor-chip-based spiking neural networks.
• Explored multi-physical computing to mirror biological multisensory hierarchies.

Main Results:

• Achieved a superadditive integration factor of 2800%, dynamical reweighting based on inverse effectiveness, and temporal congruency exceeding 10^3 s.
• The neuromorphic system demonstrated sensory synaptic plasticity, population-coded spiking, and Bayesian-optimal fusion.
• Attained 98.2% recognition accuracy for fuzzy objects, outperforming conventional fusion algorithms.

Conclusions:

• Established a physics-aware framework for neuromorphic multisensory intelligences by bridging physical dynamics with neurobiological principles.
• Demonstrated the potential of FeS-FETs for creating advanced, self-adaptive edge computing systems capable of complex sensory fusion.