Analog Signal Summation for Reinforcement Learning via Simultaneous Light-Voltage Modulation in a Synaptic Device

Area of Science:

• Materials Science
• Computer Engineering
• Artificial Intelligence Hardware

Background:

• Artificial intelligence (AI) software advances outpace hardware, hindered by the von Neumann architecture.
• Neuromorphic devices offer biomimetic computation but struggle with computational load reduction.

Purpose of the Study:

• To propose a light-voltage dual-modulating synaptic transistor for device-level computing.
• To significantly lower computational load in AI hardware.

Main Methods:

• Fabrication of a hybrid indium-gallium-zinc-oxide and InAs quantum dot synaptic transistor.
• Leveraging dual memory effects (light and voltage-induced) within a single device.
• Demonstration using a Dueling Deep Q-Network for traffic signal optimization.

Main Results:

• The device enables dual-modulation (light and voltage) for distinct memory effects.
• Traffic signal optimization achieved computation performance comparable to ideal software.
• Significant reduction in computational load demonstrated through device-level computing.

Conclusions:

• The developed synaptic transistor offers a promising solution for energy-efficient AI computing.
• Highlights the potential for high computational density in future AI hardware.
• Device-level computing via dual-modulation transistors can overcome current hardware limitations.