DeePTB-NEGF: transport from learned Hamiltonians

A quantum transport calculation usually begins with a Hamiltonian. Constructing it directly from DFT can be reliable, but it becomes expensive when the system is large or when we want to compare many different structures. DeePTB can predict a tight-binding Hamiltonian from an atomic structure, which raises a natural question: can we use that Hamiltonian directly in a quantum transport solver?

This collaborative work tests that idea. We connect a DeePTB-predicted Hamiltonian to DPNEGF and study three two-dimensional materials: graphene, hexagonal boron nitride and MoS₂. The conference paper compares the resulting band structures and electronic transmission spectra with conventional DFT-NEGF calculations. It reports good agreement and lower computational cost for the tested systems.

There are really two separate things to validate. First, the learned Hamiltonian has to be accurate. Second, the transport solver has to treat contacts, device geometry and boundary conditions correctly. Similar-looking band structures do not automatically guarantee accurate transmission. That is why testing the entire route from the learned model to a transport observable matters more than reporting a Hamiltonian error alone.

The work was published in the IEEE MoSICom 2025 proceedings, with a public arXiv version. It provides evidence for this approach on representative materials, not a guarantee of the same accuracy for arbitrary defects, interfaces or devices. IEEE paper · arXiv version.