Real space first-principles derived semiempirical pseudopotentials applied to tunneling magnetoresistance

TL;DR

This paper introduces a real space DFT-based semi-empirical pseudopotential method applied to iron and magnesium oxide, demonstrating its transferability to tunnel junctions and emphasizing the importance of interface-specific parameters for nanoscale conduction modeling.

Contribution

The paper develops a real space semi-empirical pseudopotential approach from first principles and introduces ghost pseudopotentials for accurate interface modeling in tunnel junctions.

Findings

Bulk SEP and LSDA band structures agree within 0.1 eV.

SEP transferability is characterized in real space and orbital space.

Interface-specific ghost pseudopotentials improve tunneling predictions.

Abstract

In this letter we present a real space density functional theory (DFT) localized basis set semi-empirical pseudopotential (SEP) approach. The method is applied to iron and magnesium oxide, where bulk SEP and local spin density approximation (LSDA) band structure calculations are shown to agree within approximately 0.1 eV. Subsequently we investigate the qualitative transferability of bulk derived SEPs to Fe/MgO/Fe tunnel junctions. We find that the SEP method is particularly well suited to address the tight binding transferability problem because the transferability error at the interface can be characterized not only in orbital space (via the interface local density of states) but also in real space (via the system potential). To achieve a quantitative parameterization, we introduce the notion of ghost semi-empirical pseudopotentials extracted from the first-principles calculated Fe/MgO bonding interface. Such interface corrections are shown to be particularly necessary for barrier widths in the range of 1 nm, where interface states on opposite sides of the barrier couple effectively and play a important role in the transmission characteristics. In general the results underscore the need for separate tight binding interface and bulk parameter sets when modeling conduction through thin heterojunctions on the nanoscale.