Reconfigurable Cascaded Thermal Neuristors for Neuromorphic Computing.

Area of Science:

• Materials Science
• Neuroscience
• Computer Engineering

Background:

• Complementary metal-oxide semiconductor (CMOS) technology dominates neural network hardware.
• Alternative approaches are needed for more energy-efficient and scalable computing.

Purpose of the Study:

• To explore thermal neuristors as a novel hardware implementation for neural networks.
• To demonstrate reconfigurable dynamics and information processing using thermal interactions.

Main Methods:

• Utilized the insulator-to-metal transition (IMT) in vanadium dioxide (VO2) to create thermal neuristors.
• Demonstrated device functionality through experimental characterization.
• Developed a theoretical model to elucidate the underlying physical mechanisms.

Main Results:

• Achieved diverse electrical dynamics mimicking biological neurons using thermal neuristors.
• Demonstrated inhibitory functionality within a single oxide device.
• Realized cascaded information flow exclusively through thermal interactions.

Conclusions:

• Established thermal neuristors as a viable component for brain-inspired computing.
• Laid the foundation for scalable and energy-efficient thermal neural networks.