Artificial Synapse Based on Black Phosphorus/SnS₂ Heterostructure Transistor for Neuromorphic Computing with High...

Area of Science:

• Materials Science
• Nanotechnology
• Computer Science

Background:

• Conventional von Neumann architecture faces limitations in neuromorphic computing.
• Achieving high training and learning accuracy in hardware neural networks is challenging due to nonlinear synaptic weight updates.
• Two-dimensional (2D) materials offer tunable properties for advanced artificial synapse development.

Purpose of the Study:

• To demonstrate an artificial synaptic device based on a black phosphorus (BP)/SnS2 van der Waals heterostructure.
• To achieve electrically modulated nonlinearity in synaptic weight updates.
• To integrate the device into an artificial neural network for high-accuracy neuromorphic applications.

Main Methods:

• Fabrication of a BP/SnS2 van der Waals heterostructure-based artificial synaptic device.
• Emulation of various synaptic functionalities and analysis of nonlinear weight update mechanisms using density functional theory (DFT).
• Integration of the artificial synapses into a neural network for handwritten digit recognition on the MNIST dataset.

Main Results:

• The device exhibits a large memory window and emulates essential synaptic functionalities.
• Electrical modulation of nonlinearity in synaptic weight updates was successfully achieved and analyzed.
• A BP/SnS2-based artificial neural network achieved 96.3% recognition accuracy on the MNIST dataset.

Conclusions:

• The developed BP/SnS2 artificial synaptic device shows promise for high-accuracy neuromorphic computing.
• Band structure engineering of 2D materials provides an effective pathway for advanced artificial synapses.
• This work contributes to overcoming limitations in current neuromorphic hardware through novel material-based solutions.