Reconfigurable Neuron and Synapse Operations in a Steep-Switching Nonvolatile Transistor

Area of Science:

• Materials Science
• Neuroscience
• Computer Engineering

Background:

• Neuromorphic systems aim to mimic biological neural networks but face integration challenges due to device dissimilarities.
• Current architectures are often complex and inefficient, hindering scalability.

Purpose of the Study:

• To develop a single device capable of both neuron and synapse emulation for improved neuromorphic system integration.
• To overcome the limitations of current heterogeneous device approaches in neuromorphic computing.

Main Methods:

• A steep-switching nonvolatile field-effect transistor based on a CuInP2S6/h-BN/WSe2 heterostructure was fabricated.
• Electrostatically modulating the channel's carrier density and Fermi level enabled reconfigurable neuron and synapse modes.
• Ferroelectric-gating effects were utilized to enhance channel chemical potential and reduce operating bias.

Main Results:

• The device successfully demonstrated leaky-integrate-and-fire (LiF) neuron operation by controlling the Fermi level.
• Synaptic mode was achieved by shifting the Fermi level towards the valence band, enabling weight-modulated functionality.
• Device-to-system simulations showed 95.83% accuracy in human face recognition using a single integrated neuron-synapse system.

Conclusions:

• The proposed heterostructure transistor offers a promising solution for cointegrated and scalable neuromorphic computing.
• This technology facilitates efficient emulation of neural functions, paving the way for advanced AI hardware.