[Paper Review] Autonomous sputter synthesis of thin film nitrides with composition controlled by Bayesian optimization of optical plasma emission

Davi Fébba, Kevin R. Talley+6 more|arXiv (Cornell University)|May 18, 2023

Machine Learning in Materials ScienceMaterials Science3 citations

TL;DR

This paper presents an autonomous sputtering system that uses Bayesian optimization guided by in-situ optical emission spectroscopy (OES) to control the cation composition of ZnₓTi₁₋ₓNy thin films with minimal human intervention. By correlating OES emission lines with ex-situ EDXRF-measured film composition, the system achieves target compositions within 3.5% relative error, even for 15 nm films, demonstrating high-precision, closed-loop synthesis of nitride thin films.

ABSTRACT

Autonomous experimentation has emerged as an efficient approach to accelerate the pace of materials discovery. Although instruments for autonomous synthesis have become popular in molecular and polymer science, solution processing of hybrid materials and nanoparticles, examples of autonomous tools for physical vapor deposition are scarce yet important for the semiconductor industry. Here, we report the design and implementation of an autonomous workflow for sputter deposition of thin films with controlled composition, leveraging a highly automated sputtering reactor custom-controlled by Python, optical emission spectroscopy (OES), and a Bayesian optimization algorithm. We modeled film composition, measured by x-ray fluorescence, as a linear function of emission lines monitored during the co-sputtering from elemental Zn and Ti targets in N$_2$ atmosphere. A Bayesian control algorithm, informed by OES, navigates the space of sputtering power to fabricate films with user-defined composition, by minimizing the absolute error between desired and measured emission signals. We validated our approach by autonomously fabricating Zn$_x$Ti$_{1-x}$N$_y$ films with deviations from the targeted cation composition within relative 3.5 %, even for 15 nm thin films, demonstrating that the proposed approach can reliably synthesize thin films with specific composition and minimal human interference. Moreover, the proposed method can be extended to more difficult synthesis experiments where plasma intensity depends non-linearly on pressure, or the elemental sticking coefficients strongly depend on the substrate temperature.

Motivation & Objective

• To develop a fully autonomous sputtering workflow for inorganic thin film synthesis with minimal human intervention.
• To enable precise control of cation composition in binary nitride thin films (ZnₓTi₁₋ₓNy) during physical vapor deposition.
• To establish a closed-loop feedback system using in-situ optical emission spectroscopy (OES) and Bayesian optimization for real-time composition adjustment.
• To validate the method on sub-100 nm films and demonstrate robustness across varying deposition conditions.
• To identify and mitigate sources of error in OES-based composition prediction, particularly due to chamber pressure and substrate temperature variations.

Proposed method

• A custom high-automation sputtering reactor with four cathodes and substrate heating (up to 1000 °C) was used, controlled via Python scripts and a Graphite Edge controller.
• Optical emission spectroscopy (OES) monitored emission lines from Zn and Ti during co-sputtering in N₂ atmosphere to infer plasma conditions in real time.
• A linear model was trained to map OES signal intensities to actual film composition, measured by ex-situ energy-dispersive X-ray fluorescence (EDXRF).
• Bayesian optimization was employed as the decision-making algorithm, minimizing the absolute error between target and measured OES signals to adjust RF power on each sputtering source.
• The system operated in a closed-loop mode, with real-time data streaming to a data warehouse and continuous feedback to the control algorithm.
• Calibration was performed at a fixed chamber pressure (5 mTorr), and model performance was tested at different total power and gas flow rates.

Experimental results

Research questions

• RQ1 Can in-situ optical emission spectroscopy (OES) be used as a reliable proxy for real-time cation composition during sputter deposition of Zn-Ti nitrides?
• RQ2 To what extent can Bayesian optimization reduce composition deviation from target values in autonomous sputtering of thin films?
• RQ3 How do variations in chamber pressure and total sputtering power affect the accuracy of OES-based composition prediction?
• RQ4 Can the system maintain sub-4% relative error in film composition across different film thicknesses and deposition conditions?
• RQ5 What are the dominant sources of error in OES-based feedback control, and how can they be mitigated for broader applicability?

Key findings

• The system achieved target cation compositions in ZnₓTi₁₋ₓNy films with a relative error of less than 3.5% across all tested compositions (x = 0.65, 0.75, 0.85, 0.95).
• Measured compositions were x = 0.65, 0.77, 0.86, and 0.92, respectively, demonstrating high accuracy and reproducibility even for 15 nm films.
• Variations in total sputtering power and gas flow did not significantly affect prediction accuracy as long as chamber pressure was maintained at the calibration level.
• A 12% relative error in predicted composition occurred when deposition was performed at 15 mTorr instead of the calibration pressure of 5 mTorr, highlighting the critical role of pressure consistency.
• Substrate temperature and pressure variations introduced non-linear effects that could cause degeneracy in OES signals, leading to inaccurate composition predictions if not accounted for.
• The method enables rapid calibration and adaptation to new conditions, supporting future extension to more complex systems with non-linear dependencies.