Carbon Nanotube Synaptic Transistor Network for Pattern Recognition.

Area of Science:

• Neuroscience
• Materials Science
• Computer Engineering

Background:

• Neuromorphic systems inspired by the human brain offer new computing paradigms through massive neural-network parallelism.
• Synaptic devices are crucial building blocks for information storage and processing in these systems, emulating biological synapses.
• Existing two-terminal synaptic devices face challenges due to variability and complex resistance change mechanisms, hindering high-density network implementation.

Purpose of the Study:

• To develop a reliable synaptic device for neuromorphic computing.
• To overcome the limitations of existing two-terminal synaptic devices.
• To demonstrate unsupervised learning capabilities in a synaptic transistor network.

Main Methods:

• Fabrication of a three-terminal synaptic transistor using carbon nanotubes.
• Characterization of the transistor's ability to emulate synaptic functions, including relative timing encoding and weight regulation.
• System-level simulations of a network comprising the developed synaptic transistors and complementary metal-oxide semiconductor (CMOS) circuits.

Main Results:

• The developed carbon nanotube synaptic transistor reliably performs synaptic functions.
• The device effectively encodes relative timing and regulates synaptic weight changes.
• System-level simulations demonstrated the network's capability for unsupervised learning in pattern recognition using a simplified spike-timing-dependent plasticity scheme.

Conclusions:

• A three-terminal carbon nanotube synaptic transistor offers a promising solution for reliable neuromorphic computing.
• This technology can overcome the limitations of previous synaptic devices, enabling high-density networks.
• The developed synaptic transistor network integrated with CMOS circuits can perform complex learning tasks, paving the way for advanced artificial intelligence.