Leveling Up Upconverting Nanoparticles with Machine Learning

Area of Science:

• Materials Science
• Nanotechnology
• Photonic Materials

Background:

• Upconverting nanoparticles (UCNPs) convert low-energy light to higher-energy photons, enabling diverse applications.
• Current UCNPs suffer from low brightness and limited spectral tunability, hindering widespread adoption.
• Optimizing UCNP composition and structure is challenging due to complex photophysics and vast parameter spaces.

Purpose of the Study:

• To explore the use of artificial intelligence (AI) and machine learning (ML) to enhance UCNP properties.
• To demonstrate AI/ML-driven design of novel UCNP compositions and heterostructures.
• To accelerate the discovery of advanced upconverting nanomaterials.

Main Methods:

• Utilized automated experimental workflows, physical modeling, and robotic synthesis.
• Integrated ML for data analysis, including image processing and luminescence curve analysis.
• Employed closed-loop active learning with Bayesian optimization and kinetic Monte Carlo simulations.
• Leveraged differentiable deep learning surrogate models (heterogeneous graph neural networks) for inverse design of UCNP heterostructures.

Main Results:

• Achieved a 110-fold enhancement in UCNP emission using active learning.
• Identified optimal UCNP core-shell heterostructures through AI-driven design.
• Predicted UCNP compositions with 6.5-fold greater emission intensity than the brightest training set UCNP using heterogeneous graph neural networks.
• Demonstrated AI's capability to navigate complex compositional spaces and reveal optical phenomena.

Conclusions:

• AI and ML are powerful tools for overcoming UCNP limitations in brightness and spectral tunability.
• AI/ML approaches significantly accelerate the discovery and optimization of UCNPs.
• Future UCNP research will likely involve autonomous self-driving laboratories and large language models for accelerated discovery and fundamental understanding.