Design and prediction of PIT devices through deep learning

Area of Science:

• Materials Science
• Nanotechnology
• Optoelectronics

Background:

• Graphene's unique properties make it suitable for mid-infrared (MIR) applications.
• Deep learning accelerates the design of high-performance MIR devices.

Purpose of the Study:

• To propose a graphene nano-ring-symmetric sector-shaped disk array for sensing.
• To investigate the impact of structural parameters and Fermi energy on device performance.
• To develop a deep learning model for predicting device parameters and performance.

Main Methods:

• Fabrication of a graphene nano-ring-symmetric sector-shaped disk array based on the Plasmon Induced Transparency (PIT) principle.
• Analysis of structural parameters and Fermi energy effects on sensing performance.
• Design and implementation of a six-layer deep learning network for predictive modeling.

Main Results:

• The proposed sensor achieved a Figure of Merit (FOM) of 28.7 and a sensitivity of 574 cm⁻¹/RIU.
• The deep learning model demonstrated high accuracy, with Mean Absolute Percentage Error (MAPE) of 0.5 for parameter prediction and Mean Square Error (MSE) of 1.2 for curve prediction.
• The deep neural network is simple yet effective for device data prediction.

Conclusions:

• The symmetrical sector disk array structure shows promise for sensing applications.
• The developed deep learning network provides accurate predictions, facilitating sensor design and optimization.
• This work lays a foundation for accelerated design and optimization of future graphene-based sensors.