An artificial neural network chip based on two-dimensional semiconductor

Area of Science:

• Materials Science
• Electrical Engineering
• Computer Science

Background:

• Research on 2D semiconductors is advancing from fundamental studies to functional circuits.
• Molybdenum disulfide (MoS₂) is a promising 2D material for electronic applications.

Purpose of the Study:

• To demonstrate the first functional artificial neural network (ANN) chip based on MoS₂.
• To integrate essential ANN components like multiply-and-accumulate (MAC), memory, and activation functions onto a single chip.

Main Methods:

• Fabrication of 818 field-effect transistors (FETs) using wafer-scale, high-homogeneity MoS₂ films.
• Implementation of a gate-last process for top-gate structured FETs.
• Design and optimization of analog ANN circuits using a 62-level SPICE model.

Main Results:

• Successful fabrication of a functional MoS₂ ANN chip.
• Demonstration of a tactile digit sensing recognition application with over 97% accuracy after training.
• Validation of wafer-scale integration potential for 2D semiconductors.

Conclusions:

• The MoS₂ ANN chip showcases the potential of 2D materials in integrated circuits.
• This work lays the groundwork for future applications of 2D semiconductors in AI computation.