Bayesian state estimation unlocks real-time control in thin film synthesis

TL;DR

This paper presents a Bayesian state estimation approach integrated with in situ diagnostics to enable real-time adaptive control of thin film synthesis, demonstrated on transition metal dichalcogenides growth.

Contribution

It introduces a real-time Bayesian estimation method for thin film growth parameters, combining physical modeling with optical diagnostics for autonomous control.

Findings

Robust real-time estimation of growth parameters during early film formation

Successful deployment of the method on an autonomous deposition system

Potential for adaptive control of material synthesis trajectories

Abstract

The rapid validation of newly predicted materials through autonomous synthesis requires real-time adaptive control methods that exploit physics knowledge, a capability that is lacking in most systems. Here, we demonstrate an approach to enable the real-time control of thin film synthesis by combining in situ optical diagnostics with a Bayesian state estimation method. We developed a physical model for film growth and applied the Direct Filter (DF) method for real-time estimation of nucleation and growth rates during pulsed laser deposition (PLD) of transition metal dichalcogenides. We validated the approach on simulated and previously acquired reflectivity data for WSe$_2$ growth and ultimately deployed the algorithm on an autonomous PLD system during growth of 1T$^\prime$-MoTe$_2$ under various synthesis conditions. We found that the DF robustly estimates growth parameters in real-time at early stages of growth, down to 15% percent monolayer area coverage. This approach opens new opportunities for adaptive film growth control based on a fusion of in situ diagnostics, modern data assimilation methods, and physical models which promises to enable control of synthesis trajectories towards desired material states.