Emerging edge states on the surface of the epitaxial semimetal CuMnAs thin film

Antiferromagnetic (AFM) tetragonal CuMnAs thin films possess many potential advantages in spintronic applications.^1–6 The AFM spin states in CuMnAs can be efficiently and reversibly controlled with an applied electrical current using current-induced spin–orbit torques.² On the fundamental side, CuMnAs is predicted to be an exotic magnetic Dirac semimetal, where the Dirac crossing in the band structure can be manipulated using electrically controlled magnetic order.^7–9 As a result, CuMnAs is a particularly interesting candidate material for studying the relationship between Dirac fermions and magnetism and for exploring topological metal-insulator transitions (MIT) driven by a Néel vector.⁷ However, little is known about the surface morphology, intrinsic defects, and electronic properties of CuMnAs, all of which play an important role in its magnetic properties and device performance and are prerequisites for the further understanding of its topological physics. In this regard, scanning tunneling microscopy (STM), an essential technique for probing the local electronic and magnetic structures of materials especially on the surface, is highly desirable for revealing direct correlations between the surface morphology, microscopic defects, and electronic and magnetic behaviors.^10–13 

Here, we report STM studies of a tetragonal CuMnAs thin film grown epitaxially on a GaP(001) substrate.² The STM topography reveals the As-terminated surface. The local density of states (LDOS) confirms the semi-metallic behavior of CuMnAs. Interestingly, the clean step edge displays a zigzag periodic structure with a doubling of the in-plane lattice constant, which induces localized states below the Fermi level. Density functional theory (DFT) calculations reproduce the zigzag step structure and show that the presence of As vacancies is responsible for the emerging edge states at the surface step in the CuMnAs film. This work brings to light the atomic and local electronic structures of CuMnAs surfaces and step edges, inspiring future studies on the topological properties of this material using such angle-resolved photoemission spectroscopy (ARPES)^14,15 and the local magnetic structure using spin-polarized STM.¹⁶ 

A tetragonal CuMnAs thin film was grown epitaxially on a GaP(001) substrate and then capped with an As amorphous layer before transferring into our STM system (details are given in the supplementary material). The crystal structure of tetragonal CuMnAs is illustrated in Figs. 1(a) and 1(b). The lattice parameters are c = 0.632 nm and a = b = 0.382 nm. A typical large-scale STM image of CuMnAs after removing the As capping layer is shown in Fig. 1(c). Notably, there is still some residual contamination on the surface as indicated by the bright spots in Fig. 1(c). Nevertheless, the inset in Fig. 1(c) shows an atomically resolved STM image of the CuMnAs surface, revealing a square lattice in the a and b directions with a unit cell of 0.385 ± 0.005 nm, which is half of the diagonal of the GaP unit cell (GaP lattice constant of 0.545 nm) and corresponds to the As-As distance in the (ab) plane of CuMnAs epitaxially grown on a GaP(001) substrate.¹⁷ Step edges are also observed along the a or the b direction, with small deviations due to the thermal drift of the large-scale STM image acquired at room temperature. The lower inset in Fig. 1(c) shows a line profile across several steps taken along the fast scan direction (black line) to avoid the effect from the thermal drift. As expected, step heights appear with an integer number of unit cell size along the c direction, e.g., 0.64 nm (one unit cell) or 1.28 nm (double of a unit cell). Additionally, step heights of 0.30 ± 0.01 nm and 0.33 ± 0.01 nm are also observed, which are less than a unit cell size but correspond well to the spacing between adjacent As layers as shown in Fig. 1(b). Note that the spacings between Mn–Mn layers are 0.215 nm and 0.417 nm, while the spacing of Cu–Cu layers is 0.62 nm. In addition, STM spectra on these different terraces are the same (to be discussed in below), which suggests that they have the same chemical termination. As a result, we determine the top surface of the CuMnAs thin film to be terminated by As atoms.

Figure 2(a) shows an STM image with a clean step edge (height of 0.3 nm) on the surface. The atomic resolution of As atoms is visible both on the CuMnAs surface and at step edges as displayed in the zoomed-in image of Fig. 2(b). Along the step edge, an interesting 1D zigzag feature (blue dashed line) is observed with a periodicity of 0.77 ± 0.01 nm, which doubles the As-atomic row distance of 0.385 ± 0.005 nm. The trench-like feature on the right in Fig. 2(a) is an artifact induced by a change in the tip-sample interaction when scanning across the step edge. Based on the observed images, a model structure of this step edge is proposed in Fig. 2(c). The dI/dV spectra are recorded along the line marked in Fig. 2(b) and shown in Fig. 2(d) near a step edge (black dashed arrow). The dI/dV spectra show that finite but strong suppression of the DOS is observed at the Fermi level, consistent with the predicted semi-metallic behavior in the CuMnAs samples.⁷ Upon approaching the step edge, there is a clear enhancement of the DOS below the Fermi level down to −0.8 eV, indicating the emergence of new states at step edges.

To better understand the atomic structure near the step edge and emerging edge states, we compare the experimental results with DFT calculations. In particular, we simulated an atomic model of the step edge as illustrated in Fig. 2(c). Further details can be found in the supplementary material. First to reproduce the previous literature,¹⁸ we examined the thermodynamic stability of their bulk counterparts by computing the corresponding formation energies employing a 4 × 4 × 2 CuMnAs supercell with a single Mn or As vacancy. Consistent with Ref. 18, we take the chemical potentials of Mn/As corresponding to Mn/As rich conditions (see the supplementary material). The formation energies of As and Mn vacancies in the bulk are presented in Table I. Similar to Ref. 18, we find that the formation energies of As and Mn vacancies in the bulk are positive and negative, respectively. This implies that in the bulk, the formation of a Mn vacancy in Mn rich conditions is energetically favorable, whereas As vacancies in As rich conditions are not likely to form.

Next, we consider two types of edge defects that may give rise to the 1 × 2 zigzag reconstruction observed in STM: A periodic As vacancy on the upper terrace and a Mn vacancy in the lower MnAs layer [see Figs. S3(b) and S3(c)]. The CuMnAs sample has been annealed in an ultra-high vacuum (UHV) system before the STM measurements in order to remove the As capping layer. To mimic these conditions, we take the chemical potential of As and Mn, which corresponds to a pressure P = 10⁻¹⁶atm (∼10⁻¹³Torr) and a temperature T = 550 K (see the supplementary material). Interestingly, we find that in this case, the energetic favorability of the Mn and As vacancies at a step edge is reversed (Table I). Along the step edge, Mn vacancies are energetically unfavorable, while As vacancies have negative formation energies. Therefore, we conclude that periodic As vacancies are the more likely explanation for the experimentally observed 1 × 2 reconstruction along the step edge.

Using the energetically favorable As vacancy step edge model, the STM image of the surface is simulated from our DFT charge density profiles. Figure 3(a) presents STM simulations for the step edge with periodic As vacancies obtained with the P4VASP software,¹⁹ using an isosurface value of 141.2 Å⁻³. The images show a zig-zag edge structure along the edge of As vacancies, which resemble well the STM image shown in Figs. 2(a) and 2(b). Furthermore, we plot the DOS calculated in vacuum, 2–5 Å above As at the zigzag step edge and As on the surface of the terrace [see Fig. 3(b)]. Irrespective of the height, we find a significant increase in the DOS below the Fermi level for As at the step edge, which corresponds very well with our STS data in Fig. 2(d).

Moreover, the contributions to the edge state at –0.5 eV are estimated through analyzing the spin polarized projected-DOS (PDOS), shown in Fig. S4. Specifically, we present the PDOS for As and Mn ions near the step edge in the upper MnAs layer. We find that the edge state is predominantly composed of As-p and Mn-d states present near the step edge. Interestingly, while As-p states in the bulk and inside the slab have small a magnetic moment of ∼0.013 μB, As-p states on the surface are significantly spin polarized, with a net magnetic moment of ∼0.128 μB. The highly asymmetric nature of the position of the As-ion—arising not only from the step edge but also from the adjacent As-vacancy—could be responsible for the tenfold increase in the value of local magnetic moment for the As ion at the step edge compared with the bulk As. Given that CuMnAs can be a Dirac semimetal,⁷ these step edge states may have a topologically nontrivial nature.²⁰ However, further studies are needed to evaluate this behavior.

In summary, we report a detailed study on a tetragonal CuMnAs thin film grown on a GaP substrate using STM and ab initio calculations. The surface is found to be As-terminated. The clean surface has an electronic density of states resembling those of a semimetal. The step edges exhibit a zigzag 1 × 2 reconstruction with edge states below the Fermi level. DFT calculations capture the observed emergent edge states and indicate that the zigzag reconstruction is due to the presence of As vacancies on the step edge.

See the supplementary material for methods and additional figures of the atomic model used for DFT calculations and spin polarized PDOS near the step edge.

This research was conducted at the Center for Nanophase Materials Sciences, which is a DOE Office of Science User Facility. Part of the theory effort (T.B., K.P., and V.R.C.) was supported by the U.S. Department of Energy, Office of Science, Basic Energy Sciences, Materials Sciences and Engineering Division. We acknowledge the computational resources provided by the National Energy Research Scientific Computing Center (NERSC) and Compute and Data Environment for Science (CADES) at the Oak Ridge National Laboratory, which are supported by the Office of Science of the U.S. Department of Energy under Contract Nos. DE-AC02-05CH11231 and DE-AC05-00OR22725, respectively.