Data-Enhanced Deep Greedy Optimization Algorithm for the On-Demand Inverse Design of TMDC-Cavity Heterojunctions

Area of Science:

• Materials Science
• Optoelectronics
• Computational Physics

Background:

• Strong coupling in heterojunctions is key for novel optoelectronic devices.
• Inverse design of complex nano-devices requires efficient optimization methods.
• Traditional methods for designing transition metal dichalcogenides (TMDC)-photonic cavities are time-consuming.

Purpose of the Study:

• To propose a novel data-enhanced deep greedy optimization (DEDGO) algorithm for efficient inverse design.
• To enable on-demand design of TMDC-photonic cavity heterojunctions operating in the strong coupling regime.
• To explore new physics associated with strong coupling effects in tailored heterostructures.

Main Methods:

• Developed a DEDGO algorithm integrating reinforcement learning decision theory and deep greedy optimization.
• Utilized rigorous coupled wave analysis (RCWA) to generate a training dataset.
• Employed a data enhancement method to reduce reliance on large datasets.

Main Results:

• The DEDGO algorithm demonstrated significantly higher accuracy and effectiveness than random forest and deep neural networks.
• Achieved efficient and on-demand inverse design of TMDC-photonic cavity heterojunctions.
• Showcased the algorithm's universality, interpretability, and performance in exploring underlying physics.

Conclusions:

• The DEDGO algorithm offers a powerful and efficient alternative to conventional scanning optimization methods.
• This research lays the foundation for the on-demand inverse design of low-dimensional material-based nano-devices.
• The DEDGO algorithm facilitates the exploration of fundamental physics in complex material systems.