Tuning atomic scale magnetism with artificial nanostructures

In a world where the demand for quantum technology is rapidly increasing, scanning tunneling microscopy (STM) remains one of the few experimental techniques capable of not only imaging and measuring atomic scale systems, but assembling artificial nanostructures and lattices at atomic precision. The emergent properties of increasingly complex quantum systems can be designed and characterised by assembling structures from individual atoms and molecules. Some of the most interesting building blocks for such lattices have magnetic properties: by coupling spin systems into lattices, a rich tapestry of physics becomes accessible for experimentation and applications. Despite the promising theoretical predictions, the interplay of artificial nanostructures and atomic-scale magnets remains relatively unexplored. This thesis discusses recent experimental efforts to understand magnetic impurities coupled to a conduction bath, how machine learning techniques can be utilized in atom manipulation, and finally the behaviour of magnetic impurities inside artificial nanostructures. A magnetic impurity coupled to a conduction bath gives rise to the Kondo effect, whereby the magnetic moment of the impurity is screened by conduction electrons. This many-body effect results in a resonance with an intrinsic temperature dependence. We experimentally verify a new model for this temperature dependence, and demonstrate the importance of various broadening factors in the analysis of the spectral features. Our work provides a widely applicable model for verifying the Kondo nature of a resonance at the Fermi level, and how to accurately determine the energy scale determining the low-temperature dynamics of such systems, i.e. the Kondo temperature. We then proceed to explore how deep reinforcement learning (DRL) methods can be applied to lateral atom manipulation. A DRL algorithm is designed and trained to find suitable manipulation parameters for moving Ag and Co atoms on a Ag(111) surface. The trained model is capable of adjusting to changing conditions, and combined with path planning algorithms forms the basis for an autonomous nanostructure assembly system. Finally, we combine Kondo systems and atom manipulations by studying magnetic impurities inside quantum corrals, closed structures built from individual atoms. By confining the surface state electrons of the underlying Ag(111) substrate, we tune the conduction bath environment of Co atoms and H2-Pc molecules and observe changes in their low-energy excitations. The presented results pave the way for further studies combining magnetic impurities and artificial lattices built atom by atom.