Bachelor defense by August Lademark Heegaard

Optimizing Atomic Force Microscopy Imaging Protocols Through Engineering Principles

Abstract: Following the principle of "Log it, Visualize it, Utilize it", this study exemplifies data-driven research within the realm of quantum computing. We grow wafers, structurally characterize them, and fabricate devices while logging, visualizing, and utilizing data to optimize material properties. Our focus lies on atomic force microscopy (AFM) image analysis of aluminum thin films on silicon wafers. We ascertain the optimal samples per line (SpL) in AFM scans, finding no over-sampling in scans ranging from 200nm to 10 µm with SpL from 256 to 1008. Our analyses indicate that surface area ratio (SAR) is the most sensitive to SpL changes. But also, that it might only be dependent on the SpL and no other parameters.

Additionally, we assess 12 electron-beam resist recipes for lithography, comparing quantified parameters with a reference sample. Discrepancies found between quantified similarity and qualitative ranks highlight the need for refining these parameters.

A gradient-boosted machine learning algorithm helps assign feature importance using the qualitative ranking. We observe that the mean of a Gaussian fitted to a histogram of the height’s accounts for 51.3% of the total importance. By weighing these importances, we find improved rankings for cleaning methods compared to the reference sample. These findings underline the need for more comprehensive, quantified parameters to enhance data-driven research efficiency.