SUPERVISORY PROCESS CONTROL

TECHNOLOGY

A Gaussian process (GP) is a supervised learning method used to address regression and probabilistic classification problems. The term “Gaussian” reflects the fact that each GP can be viewed as an infinite-dimensional generalization of multivariate GP distributions. When used for regression, it is referred to as Gaussian process regression (GPR).
GPR has been applied to various real-world problems across fields such as materials science, chemistry, physics, and biology. It is a non-parametric Bayesian approach to inference. Instead of estimating a distribution over the parameters of a parametric function, Gaussian processes allow for direct inference of a distribution over the function of interest.
A Gaussian process defines a prior function that is updated to form a posterior function after observing some values from the prior distribution.

Intelligent automation, driven by machine learning and robotics advances, is poised to transform materials research globally. Researchers can quickly synthesise new thin-film materials with desired properties in PVD tools using closed-loop systems. This involves preparing raw materials and employing machine learning and robotics to synthesise and evaluate materials autonomously. The relevant synthesis conditions and resulting properties data are automatically stored in a database.
The material development and optimisation process can be streamlined into four steps: (1) determining synthesis conditions, (2) performing synthesis, (3) performing evaluation, and (4) adapting the following synthesis conditions. Typically, these tasks are performed manually, and even with automation, the following conditions often require researchers' input, slowing the process. However, advancements in control technology allow the creation of an autonomous closed-loop cycle, reducing the need for human intervention.

The development of new thin-film materials in laboratories involves multiple cycles of conception, synthesis, and characterization conducted manually by researchers. This traditional process is essential for driving innovation and sustainability, but it is often slow, inefficient, and costly.
Plasma PVD tools are versatile and allow for the synthesis of diverse materials and chemistries. However, the complexity of plasma PVD processes and the numerous parameters involved can be overwhelming when trying to understand the underlying mechanisms and the observed effects.
The primary challenge in harnessing AI's potential for accelerating materials research is ensuring a CONSISTENT supply of RELIABLE data at a SUFFICIENTLY large scale.

Incorporation of Artificial Intelligence (AI) into metrological systems have already shown significant improvements in measurement accuracy.
AI can transform our approach to thin film materials research - a rapidly growing field that commonly employs Physical Vapour Deposition (PVD) to synthesize novel materials.
AI provides powerful tools for modelling, evaluation, and control of PVD processes. By leveraging AI, one can now analyze large volumes of data, refine their treatment protocols, and predict treatment outcomes fast and with a level of precision that was previously unattainable.
A modern PLC and robust Inteleg® plasma OES-based process monitoring and control systems ensure consistent delivery of reliable data at a scale sufficient for AI/ML implementations.

The integration of automation, machine learning, robotics, and big data into materials synthesis promises to transform materials research and usher in groundbreaking advancements across various fields.
Throughputs can be over 10 times faster than traditional PVD workflows, leading to accelerated discovery and autonomous optimisation of the vast array of materials synthesised by PVD.