Taylor Stock1
1University College London
Scanning Tunneling Microscopy Hydrogen Resist Lithography (STM-HRL) is a fabrication technique capable of producing electronic devices in silicon with true atomic precision. In this technique, the probe of an STM is used to pattern a single atomic layer of hydrogen on an atomically perfect silicon surface. This forms a chemically resistant mask for the atomically precise positioning of substitutional dopant atoms, delivered via gas phase molecular precursors. STM-HRL is the most precise semiconductor device fabrication technique available, and in principle, could be used to fabricate a solid-state universal quantum computer. The potential applications of STM-HRL thus hold great promise to deliver transformative technologies in the future, however the pathway to realization is challenging. Currently, device fabrication relies on skilled STM operators who handcraft individual devices, one at a time, in a labor-intensive process. By introducing machine learning assistance or replacement of the STM operator for tasks including complex real-time image analysis and dynamic instrument and process optimization, it should be possible to significantly increase the complexity, yield, and throughput of atomically precise device fabrication. In this work, we introduce the full process of STM-HRL fabrication and explore opportunities for the introduction of machine learning assistance. In particular, we explore machine learning image recognition as a means to identify and map distributions of dopant atoms and atomic-scale defects and to align these features within STM-HRL fabricated device structures.