Silicon oxycarbides are synthesized from polymer precursors and feature a massively tunable microstructure and composition.  However, the delicate interplay of structure, composition and processing conditions obscures the influence of individual parameters on their properties.
In this work, we fit a machine learning interatomic potential to the system and employ it to investigate processing-structure-property relations. We produce samples based on a variety of precursors, ranging from atoms to polymer-like molecules. For each precursor, the influence of different  processing temperatures is investigated, leading to a plethora of samples with different microstructures and compositions. Finally, we relate elastic properties to various structural features, finding a strong correlation with Si-C and Si-O bonds. Contrary to common assumptions we do not find a dependence on the ‘free carbon’ phase.