AI-driven autonomous workflow for PLD growth with in situ RHEED

by Pietro Mani (PSI)

Wednesday 8 Jul 2026, 15:00 → 16:00 Europe/Zurich

OVGA/200

While high-throughput computational simulations can rapidly predict novel materials, the low efficiency and poor reproducibility of experimental synthesis create a persistent "synthesizability gap." Pulsed laser deposition (PLD) offers the versatility needed to finely tune properties and structure of a new material, but its highly complex parameter space and not-reproducible manual operations hinder efficient experimental growths. In this talk, we demonstrate an autonomous, closed-loop workflow that optimizes the layer-by-layer growth of Ag(Nb,Ta)O₃ (ANT) films. By processing in situ reflection high-energy electron diffraction (RHEED) data frame-by-frame, our algorithm transforms qualitative experimental feedback into a reproducible, quantitative score. This metric drives a Gaussian process regression and Bayesian optimization loop to iteratively guide the selection of laser fluence and O₂ background pressure for following depositions. This computationally lightweight, hardware-agnostic approach promises to minimize experimental iterations while yielding higher-quality films than traditional manual optimization. Ultimately, this workflow establishes a scalable framework to accelerate the experimental realization of novel materials in PLD labs.