Hierarchical processing enabled by 2D ferroelectric semiconductor transistor for low-power and high-efficiency AI...

Area of Science:

• Materials Science
• Semiconductor Physics
• Artificial Intelligence Hardware

Background:

• Traditional hardware faces efficiency and power limitations due to separated sensor, memory, and computation units.
• The increasing volume of data and the Internet of Things exacerbate these hardware challenges.
• Novel materials and architectures are needed for integrated processing and memory functionalities.

Purpose of the Study:

• To design and demonstrate an α-phase indium selenide (α-In2Se3) transistor for artificial synapses.
• To enable processing-in-sensor (PIS) and computing-in-memory (CIM) functionalities using a 2D ferroelectric semiconductor channel.
• To evaluate the performance of the α-In2Se3 transistor in both optic-neural and electro-neural synapse applications.

Main Methods:

• Fabrication of a transistor utilizing α-In2Se3 as the 2D ferroelectric semiconductor channel material.
• Characterization of the transistor's performance as an optic-neural synapse, measuring photoresponsivity and detectivity.
• Evaluation of the transistor's performance as an electro-neural synapse, assessing program/erase speed and energy consumption.
• Integration of α-In2Se3 transistors into an AI vision system to demonstrate PIS and CIM functionalities.

Main Results:

• The α-In2Se3 transistor demonstrated high photoresponsivity (2855 A/W) and detectivity (2.91 × 10^14 Jones) as an optic-neural synapse.
• As an electro-neural synapse, it achieved fast program/erase speeds (40 ns/50 µs) and ultralow energy consumption (0.37 aJ/spike).
• An AI vision system utilizing these transistors achieved 92.63% recognition accuracy within 12 epochs, showcasing synergistic PIS and CIM effects.

Conclusions:

• The α-In2Se3 transistor shows significant potential for next-generation vision hardware.
• The developed device enhances processing efficiency and power efficiency in AI applications.
• This work paves the way for advanced integrated sensing and computing systems.