3D Neuromorphic Hardware with Single Thin-Film Transistor Synapses Over Single Thin-Body Transistor Neurons by...

Area of Science:

• Neuromorphic Engineering
• Artificial Intelligence
• Materials Science

Background:

• Spiking neural networks (SNNs) offer energy efficiency for AI but are limited by complex CMOS circuits.
• Existing neuromorphic hardware faces challenges in scalability and power consumption.

Purpose of the Study:

• To demonstrate a novel neuromorphic module with vertically integrated synapses and neurons.
• To improve the energy efficiency and scalability of neuromorphic hardware.

Main Methods:

• Monolithic vertical integration of a single thin-film transistor (1TFT) synapse and a single transistor (1T) neuron.
• Utilizing Excimer Laser Annealing (ELA) and Rapid Thermal Annealing (RTA) for dopant activation.
• Employing internal electro-thermal annealing (ETA) to enhance synapse endurance.

Main Results:

• Achieved high classification accuracy (≈92.3%) for American Sign Language (ASL) using the fabricated neuromorphic module.
• Demonstrated robust performance of the 1TFT-synapse even after extensive update pulses (204,800).
• Successfully implemented neuromorphic vision sensing capabilities.

Conclusions:

• The demonstrated neuromorphic module offers a scalable and energy-efficient solution for AI.
• Vertical integration of 1TFT-synapses and 1T-neurons presents a promising architecture for future neuromorphic systems.
• The device shows significant potential for real-world applications like neuromorphic vision sensing.