Majorana zero modes (MZMs), emerging as exotic quasiparticles that carry non-Abelian statistics, hold great promise for achieving fault-tolerant topological quantum computation. A key signature of the presence of MZMs is the zero-bias peaks (ZBPs) from tunneling differential conductance. However, the identification of MZMs from ZBPs has faced tremendous challenges, due to the presence of topological trivial states that generate spurious ZBP signals. In this work, we introduce a machine-learning framework that can discern MZM from other signals using ZBP data. Quantum transport simulation from tight-binding models is used to generate the training data, while persistent cohomology analysis confirms the feasibility of classification via machine learning. In particular, even with added data noise, XGBoost classifier reaches 85% accuracy for 1D tunneling conductance data and 94% for 2D data incorporating Zeeman splitting. Tests on prior ZBP experiments show that some data are more likely to originate from MZM than others. Our model offers a quantitative approach to assess MZMs using ZBP data. Furthermore, our results shed light on the use of machine learning on exotic quantum systems with experimental-computational integration.
@article{arxiv.2310.18439,
title = {Machine learning detecting Majorana Zero Mode from Zero Bias Peak measurements},
author = {Mouyang Cheng and Ryotaro Okabe and Abhijatmedhi Chotrattanapituk and Mingda Li},
journal= {arXiv preprint arXiv:2310.18439},
year = {2023}
}