The Digitisation of the Sputter Deposition Process of Transparent Conductive Oxides by Implementing Artificial Intelligence

Plasma-based sputtering is extensively employed to fabricating thin film Transparent Conductive Oxides (TCOs), a category of semiconducting material used for a wide variety application from flat panel display to energy harvesting devices. Methods of evaluating the plasma i.e. glow discharge has been greatly studied requiring complex theoretical physics which is not viable for applied/materials scientist who frequently use this method of deposition at an operational level. The aim of the project was to explore new methods of characterizing the plasma sputtering process to evaluate the possibility of simplifying the monitoring and assessment of the sputtering process. The first method involved monitoring the RF-based plasma sputtering process through optical spectroscopy and characterizing the discharge based on its specific colour. The 2nd method involved implementing Artificial intelligence/Machine learning and feeding the emission spectrum of the plasma extracted from an array of depositions to a deep learning model to evaluate the accuracy of predicting not only the properties of the deposited TCO film but also the growth process conditions. Implementation of such methods pave the way for the design of a digital shadow for plasma-based deposition in the material engineering industry. Spectral data from the plasma was obtained by placing an in-vacuum collimator optic probe (Plasus) which featured a unique honeycomb structure capturing photons whilst simultaneously trapping sputtering particles and preventing gradual coating of the collimator’s quartz window. The spectrometer was programmed to calculate the area under the peak of the spectral range based on predesignated segments of the spectrum. In addition to this, the light collected from the plasma was also guided to a 2nd spectrometer (Jeti) that calculates the chromaticity index of the light. The colour of the plasma source was deduced based on conventional chromaticity index analysis and it was compared to the direct spectral data plots of the emission peaks to investigate the possibility of characterizing the plasma based on its specific colour. This technique was demonstrated to be a viable potential for evaluating the plasma in terms of providing information regarding the stability of the plasma, chamber pressure and plasma power. A linear relationship between the colour functions and the plasma power was observed, while the stability of the sputtering plasma can be assessed based on the plasma colour functions. The colour functions also follow a unique pattern when the working gas pressure is increased. The spectral properties and colour functions of a radio frequency (RF)-based sputtering plasma source was also monitored during consecutive sputter deposition of Indium doped zinc oxide (IZO) thin films under argon and argon/hydrogen mix. The effect of target exposure to the hydrogen gas on charge density/mobility and spectral transmittance of the deposited films was investigated. Consecutive exposure to the hydrogen gas during the deposition process progressively affects the properties of thin films with a certain degree of continuous improvement in electrical conductivity while demonstrating that reverting to only argon from argon/hydrogen mix follows a complex pathway. Preparation of highly conductive zinc oxide thin films without indium presence was exhibited eliminating the need for the expensive indium addition. The complexity of the reactive sputtering of highly conductive zinc oxide thin films in the presence of hydrogen at room temperature was investigated. A hypothesis was put forward regarding importance ii of precise geometric positioning of the substrate with respect to the magnetron to achieve maximum conductivity. The electrical properties of the deposited zinc oxide thins films based on their position on the substrate holder relative to the magnetron were examined. Machine Learning/Deep learning models were incorporated to examine the accuracy of predicting a single feature (sheet resistance) of thin films of indium-doped zinc oxide deposited via plasma sputter deposition by feeding the spectral data of the plasma to the deep learning models. It was shown that Artificial Neural networks could be implemented as a model that could predict the sheet resistance of the thin films as they were deposited, taking in only the spectral emission of the plasma as an input. The spectral emission data from the plasma glow of various sputtering targets containing indium oxide, zinc oxide, and tin oxide were obtained. These spectral data were then converted into twodimensional arrays by implementing a basic array-reshaping technique and a more complex procedure utilizing an unsupervised deep-learning technique, known as the self-organizing-maps method. The twodimensional images obtained from each single-emission spectrum of the plasma mimic an image that can then be used to train a convolutional neural network model capable of predicting certain plasma features, such as impurity levels in the sputtering target, working gas composition, plasma power, and chamber pressure during the machine operation. It was demonstrated that that the single-array-to-2D-array conversion technique, coupled with deep-learning techniques and computer vision, can achieve high predictive accuracy and can, therefore, be fundamental to the construction of a sputtering system’s digital twin.