Machine Learning-enhanced Efficient Spectroscopic Ellipsometry Modeling
Over the recent years, there has been an extensive adoption of Machine Learning (ML) in a plethora of real-world applications, ranging from computer vision to data mining and drug discovery. In this paper, we utilize ML to facilitate efficient film fabrication, specifically Atomic Layer Deposition (ALD). In order to make advances in ALD process development, which is utilized to generate thin films, and its subsequent accelerated adoption in industry, it is imperative to understand the underlying atomistic processes. Towards this end, in situ techniques for monitoring film growth, such as Spectroscopic Ellipsometry (SE), have been proposed. However, in situ SE is associated with complex hardware and, hence, is resource intensive. To address these challenges, we propose an ML-based approach to expedite film thickness estimation. The proposed approach has tremendous implications of faster data acquisition, reduced hardware complexity and easier integration of spectroscopic ellipsometry for in situ monitoring of film thickness deposition. Our experimental results involving SE of TiO2 demonstrate that the proposed ML-based approach furnishes promising thickness prediction accuracy results of 88.76% within +/-1.5 nm and 85.14% within +/-0.5 nm intervals. Furthermore, we furnish accuracy results up to 98% at lower thicknesses, which is a significant improvement over existing SE-based analysis, thereby making our solution a viable option for thickness estimation of ultrathin films.