We will discuss a recently built state-of-the-art ultra-high-vacuum (UHV) test system with sensitive detection and quantification of hydrogen uptake, solubility and diffusion in various materials. The test system’s manifold is equipped with capacitance diaphragm gauges (CDG), residual gas analyzer, spinning rotor gauge, calibrated volumes, turbo pumps, scroll pumps and a leak detector. The manifold is surrounded by a bake-out oven to keep impurity levels sufficiently low. Two sample chambers are enclosed by high-temperature furnaces. The design was developed by Pacific Northwest National Laboratory (PNNL) in collaboration with Vacuum Technology Inc (VTI). This automated UHV system can be used for several studies of hydrogen-metal interactions including absorption/desorption kinetics, thermodynamics, isotherms, plateau pressures, isotope studies, gaseous impurity identification and permeation rate.

Vacuum deposition processes are being equipped with an ever-expanding array of sensors to gain more control over the process conditions. Unfortunately, this often presents the operator with too much data to be able to draw clear insights into the performance of the process. Machine learning algorithms are a powerful tool for analyzing large and complex sets of data and have been at the forefront of a revolution artificial intelligence. These techniques are ideally suited for analyzing problems encountered in vacuum processes, which are often expressed as “classification problems”, i.e., identifying if a leak is present in the system or not. In particular, they can be applied to the automated analysis of vacuum processes by remote plasma optical emission spectroscopy (RPOES).

Remote plasma OES provides critical information on the state or condition of a process via measurement of residual gas partial pressure present in the chamber. RPOES is now a popular method of residual gas analysis (RGA), as it is industrially robust compared quadrupole mass spectrometry methods and operates from 0.5 mbar to 10^-7 mbar. Whilst RGA information is important, expert knowledge is often required to be able to interpret the data, and in some cases, the spectra are too complex to extract key information using the human eye alone. This paper will present the application of a machine learning A.I. to the automated analysis of magnetron and remote plasma OES data. Examples include leak detection, organic contamination detection and the identification of organic molecules from cracking patterns.