Spiking neurons from tunable Gaussian heterojunction transistors

Area of Science:

• Materials Science
• Neuroscience
• Computer Engineering

Background:

• Spiking neural networks (SNNs) offer energy-efficient neuromorphic computing by leveraging spatiotemporal processing and spiking sparsity.
• Conventional silicon-based SNN circuits face limitations in integration density due to complex layouts and multiple transistors.

Purpose of the Study:

• To demonstrate a simplified implementation of spiking neurons using novel transistor designs.
• To explore the potential of electrostatic control in heterojunction transistors for neuromorphic applications.

Main Methods:

• Fabrication of dual-gated Gaussian heterojunction transistors using van der Waals heterostructures.
• Integration of chemical vapor deposited monolayer molybdenum disulfide (MoS2) with solution-processed semiconducting single-walled carbon nanotubes (s-SWCNTs).
• Emulation of biological neuron ion channel dynamics within the transistor circuits.

Main Results:

• Achieved unprecedented electrostatic control over dual-gated Gaussian heterojunction transistors.
• Demonstrated circuits capable of various biological spiking responses: phasic spiking, delayed spiking, and tonic bursting.
• Successfully emulated spike-generating ion channels using mixed-dimensional van der Waals heterojunctions.

Conclusions:

• The developed transistors offer a simplified and highly integrable approach to implementing spiking neurons for neuromorphic computing.
• The tunable Gaussian response of these devices has broad implications beyond AI, including telecommunications, computer vision, and natural language processing.