XRL-QNet: an explainable reinforcement learning framework for optimizing and evaluating quantum dots fabrication.

Area of Science:

• Quantum photonics
• Semiconductor device manufacturing
• Artificial intelligence in materials science

Background:

• High-quality, on-demand single-photon sources are crucial for quantum photonics.
• Semiconductor quantum dots (QDs) offer tunable emission spectra but face challenges in scalable, efficient fabrication due to intrinsic randomness.
• Current QD measurement techniques lack real-time control, hindering practical implementation.

Purpose of the Study:

• To develop an automated and scalable method for synthesizing high-performance quantum dots.
• To address the limitations of manual spectrum analysis and lack of real-time control in QD fabrication.
• To integrate machine learning for optimizing fabrication parameters and improving QD quality.

Main Methods:

• Implementation of XRL-QNet, a machine learning platform utilizing Proximal Policy Optimization (PPO).
• Real-time optimization of fabrication parameters (e.g., substrate temperature, material flux, growth time) during molecular beam epitaxy (MBE).
• Utilizing CNN autoencoders for latent spectral feature extraction and neural regression for scoring, enhanced by explainable AI (XAI) techniques (SHAP, LIME).

Main Results:

• XRL-QNet successfully optimizes QD fabrication parameters in real-time.
• The platform extracts spectral features and provides confidence scores for QD quality assessment.
• Explainable AI techniques ensure transparency and interpretability in the QD assessment process.

Conclusions:

• XRL-QNet with PPO optimization enables the synthesis of high-performance QDs tailored for specific quantum applications.
• The closed-loop system integrates fabrication control with emission spectrum analysis for automated, scalable QD production.
• This approach represents an innovative advancement in semiconductor quantum dot manufacturing for quantum technologies.