Machine Learning-Based Detection and Classification of Majorana Zero Modes in FeTe_0.55Se_0.45 using millikelvin STM

We have used advanced statistical techniques to analyze putative Majorana zero modes (MZM) in quantum computing applications using millikelvin scanning tunneling microscopy data. We have exploited the high spatial and energy resolution of mK-STM to reveal complex in-gap states of the intrinsic topological superconductor, FeTe_0.55Se_0.45. However, the coexistence of other states near the zero bias, such as defect-induced in-gap states and Caroli-de Gennes-Matricon states, make the detection of definitive MZM challenging. While traditional approaches using rule-based classification through multiple Lorentzian fitting require considerable computational effort and analysis time, we developed a machine learning-based classification model using the scikit-learn package. Based on feature extraction from local density of state (LDOS) peak properties of grid points, we classified the clustered features and built a model to distinguish MZM from other in-gap states in the magnetic field-dependent LDOS data. This approach not only accelerates the multidimensional analysis of local density of states data, but also could be useful to manipulate MZM for quantum computation.