Sumner B. Harris¹, Arpan Biswas², Daniel T. Yimam¹, Ruth Fajardo³, Feng Bao³, Christopher Rouleau¹, Alexander Puretzky¹, Kai Xiao¹, Rama Vasudevan^1
1Oak Ridge National Laboratory, Oak Ridge, TN
²UT-Oak Ridge Innovation Institute, Oak Ridge, TN
³Florida State University, Tallahassee, FL
Advancing thin film fabrication systems towards autonomous machines that integrate the synthesis process controls with artificial intelligence (AI) and in situ diagnostics promises to enable the exploration of large parameter spaces for thin film optimization at rates beyond what is possible with human operators alone. In this talk, I will discuss the key challenges for enabling AI-driven pulsed laser deposition (PLD) platforms and the solutions we are developing at the Center for Nanophase Materials Sciences (CNMS). I will describe PLD systems with two different approaches for AI-driven PLD: a cluster system approach with robotic, in vacuo sample transfer to characterization stations, and a stand-alone approach with rotary sample exchange and real-time optical diagnostics. I will then discuss the first demonstrated autonomous PLD synthesis experiment, in which we optimized the crystallinity of ultrathin WSe2 films using in situ Raman spectroscopy as feedback to Gaussian process regression with Bayesian optimization algorithms. This effort demonstrated at least a 10x increase in throughput over tradition PLD workflows and autonomously discovered the growth windows with sparse sampling of the parameter space. Next, I will show that deep learning with intensified-CCD image sequences of the plasma plume generated during PLD can be used for anomaly detection and the prediction of thin film growth kinetics. The predictive capabilities of plume dynamics on the kinetics of film growth or other film properties prior to deposition provides a means for rapid pre-screening of growth conditions for the non-expert, which promises to accelerate materials optimization with PLD. Finally, I will discuss on-line Bayesian state estimation methods for model-based real-time control over synthesis using optical reflectivity.