Neuron Circuit Based on a Split-gate Transistor with Nonvolatile Memory for Homeostatic Functions of Biological...

Area of Science:

• Neuromorphic Engineering
• Solid-State Devices
• Artificial Intelligence

Background:

• Biological neurons exhibit homeostatic functionality, crucial for neural network stability.
• Existing artificial neuron circuits face challenges in mimicking this homeostatic behavior.
• Split-gate field-effect transistors (S-G FETs) offer potential for artificial neuron implementation.

Purpose of the Study:

• To propose and evaluate a novel S-G FET with a charge trap layer for neuron circuits.
• To demonstrate the ability to control the threshold voltage (Vth) for neuron firing rate modulation.
• To verify the performance enhancement in a simulated Spiking Neural Network (SNN).

Main Methods:

• Fabrication of an S-G FET incorporating a silicon nitride (Si3N4) charge trap layer.
• Separation of read and Vth control gates using a fin structure to prevent dielectric degradation.
• Development of a pulse modulation circuit for generating Program/Erase pulses.
• Simulation of a 2-layer SNN employing the developed S-G FET for online unsupervised learning and classification.

Main Results:

• The Vth of the S-G FET was successfully modulated by adjusting trapped charges in the Si3N4 layer.
• A high firing rate was reduced by increasing the Vth of the neuron circuit.
• A significant 8% improvement in recognition rate was achieved in the simulated SNN.

Conclusions:

• The proposed S-G FET effectively mimics biological neuron homeostatic functionality.
• The device enables precise control over neuron firing rates, enhancing SNN performance.
• This approach offers a promising pathway for developing more efficient and accurate neuromorphic systems.