Bioinspired Tripartite Synaptor with Conjoined Twin Transistors for Homeostasis in Neuromorphic Hardware

Area of Science:

• Neuroscience and Computer Engineering
• Development of novel electronic components for artificial intelligence

Background:

• Conventional von Neumann architecture faces limitations, driving research into neuromorphic computing.
• Biological tripartite synapses, involving astrocytes, are crucial for robust and adaptive neural systems.
• Mimicking these biological properties is key for advancing artificial neural systems.

Purpose of the Study:

• To design and fabricate an artificial tripartite synaptic transistor (synaptor).
• To integrate both synaptic and homeostatic functionalities into a single, CMOS-compatible device.
• To enhance the performance of neural networks for complex tasks.

Main Methods:

• Utilized a split-gate silicon-oxide-nitride-oxide-silicon (SONOS) structure for the tripartite synaptor.
• Employed two independently addressable gates: one for synaptic weight updates (charge trapping/detrapping) and another for homeostasis (dynamic conductance regulation).
• Applied the tripartite synaptor to neural networks for image classification.

Main Results:

• Demonstrated long-term retention and distinct synaptic plasticity (potentiation and depression).
• Achieved dynamic conductance regulation for homeostasis via the secondary gate.
• Showcased higher classification accuracy for grayscale digits and RGB images compared to bipartite synapse systems.

Conclusions:

• The artificial tripartite synaptor successfully mimics biological tripartite synapses.
• The device offers a scalable hardware platform for reliable neuromorphic computing with integrated homeostasis.
• This approach enhances neural network performance and reliability.