Layer-Confined Excitonic Insulating Phase in Ultrathin Ta2NiSe5 Crystals
So Young Kim,† Youngwook Kim,†,⊥ Chang-Jong Kang,† Eun-Su An,† Hyoung Kug Kim,†,‡ Man Jin Eom,† Minkyung Lee,‡,§ Chibeom Park,‡,# Tae-Hwan Kim,†,‡ Hee Cheul Choi,‡,§ Byung Il Min,† and Jun Sung Kim*,† †
Department of Physics and §Department of Chemistry, Pohang University of Science and Technology, Pohang 37673, Korea Center for Artificial Low Dimensional Electronic Systems, Institute for Basic Science (IBS), Pohang 37673, Korea
‡
S Supporting Information *
ABSTRACT: Atomically thin nanosheets, as recently realized using van der Waals layered materials, offer a versatile platform for studying the stability and tunability of the correlated electron phases in the reduced dimension. Here, we investigate a thickness-dependent excitonic insulating (EI) phase on a layered ternary chalcogenide Ta2NiSe5. Using Raman spectroscopy, scanning tunneling spectroscopy, and in-plane transport measurements, we found no significant changes in crystalline and electronic structures as well as disorder strength in ultrathin Ta2NiSe5 crystals with a thickness down to five layers. The transition temperature, Tc, of ultrathin Ta2NiSe5 is reduced from its bulk value by ΔTc/Tcbulk ≈ −9%, which strongly contrasts the case of 1T-TiSe2, another excitonic insulator candidate, showing an increase of Tc by ΔTc/Tcbulk ≈ +30%. This difference is attributed to the dominance of interband Coulomb interaction over electron−phonon interaction and its zeroordering wave vector due to the direct band gap structure of Ta2NiSe5. The out-of-plane correlating length of the EI phase is estimated to have monolayer thickness, suggesting that the EI phase in Ta2NiSe5 is highly layer-confined and in the strong coupling limit. KEYWORDS: excitonic insulator, van der Waals materials, ternary chalcogenides, ultrathin crystals, interband Coulomb interaction, direct band gap semiconductor condensation (BEC).12 Although several exotic properties, such as superfluidity13,14 and the BCS−BEC crossover,12 were predicted a half century ago, experimental realization of the EI phase has been elusive. This difficulty occurs mainly because the excitonic instability coexists or completes with instabilities of other ground states like charge density wave (CDW) or staggered orbital orders in real solids.15−17 How the EI phase is stabilized in the complex interplay of Coulomb interaction, electron−phonon (e−ph) coupling, and band hybridization has remained a challenging question. Among a handful of material candidates, two vdW materials, 1T-TiSe2 and Ta2NiSe5, are the most promising systems to host the EI phase. 1T-TiSe2 has a semimetal or indirect gap structure and shows a CDW phase with a 2 × 2 × 2 modulation below Tc ≈ 190 K.18 Its underlying mechanism is still controversial, but the EI instability is considered to be crucial for CDW
T
wo-dimensional (2D) layered materials, atomically thin layers stacked via van der Waals (vdW) interactions, have recently been used in the investigation of unusual physical properties at the 2D limit.1−3 As the vdW layered materials are thinned down toward a few layers, electronic or phonon structures can be significantly modified, as evidenced by recent studies on various 2D transition metal dichalcogenides.4−7 In vdW materials hosting a many-body correlated phase, such changes in the reduced dimension drastically affect the phase stability and the nature of the corresponding order. In this respect, recent development of the so-called 2D materials allows us to study how various ordered phases are stabilized and modified in the reduced dimension. An excitonic insulating (EI) phase is one of the intriguing correlated phases in which electron−hole bound pairs, that is, excitons, are condensed in semimetals or semiconductors with a low carrier density.8−11 A weakly screened interband Coulomb interaction that exceeds a small band overlap or gap induces spontaneous formation of excitons, in close analogy with Bardeen−Cooper−Schrieffer (BCS) theory or Bose−Einstein © 2016 American Chemical Society
Received: July 19, 2016 Accepted: August 15, 2016 Published: August 15, 2016 8888
DOI: 10.1021/acsnano.6b04796 ACS Nano 2016, 10, 8888−8894
Article
www.acsnano.org
Article
ACS Nano formation.19−22 Ta2NiSe5 is another strong candidate having a direct gap structure.23−25 A characteristic flattening of the valence band top, observed in angle-resolved photoelectron spectroscopy (ARPES) below Tc ≈ 325 K, is taken as a strong signature for the EI phase.23,24 How the EI instability differs in these compounds with contrasting electronic structures is one of the key questions to clarify the nature of the EI phase. Considering that two compounds have common vdW structure, one effective approach is to investigate their properties in ultrathin crystals. In fact, Tc of 1T-TiSe2 is strongly enhanced by reducing its thickness.26 Further study on ultrathin Ta2NiSe5, in comparison with 1T-TiSe2, thus can offer important clues to clarify how the EI phase is stabilized in the reduced dimension. In this work, we report that the EI phase of Ta2NiSe5 remains almost unchanged even in ultrathin crystals. Based on a systematic study of the thickness-dependent Raman-active phonon modes, scanning tunneling spectroscopy (STS), and transport properties, we found no significant change of crystalline and electronic structures as well as disorder strength in ultrathin Ta2NiSe5 down to five layers. Tc in ultrathin Ta2NiSe5 is found to be ∼9% lower than its bulk value. This response is clearly distinct from 1T-TiSe2, in which Tc is ∼30% higher in an ultrathin crystal than in the bulk.26 This difference is attributed to the dominance of Coulomb interaction over e− ph coupling in Ta2NiSe5, mainly due to its direct band gap structure and the relevant ordering wave vector of q = 0, in strong contrast to 1T-TiSe2. Furthermore, the coherence length of the EI phase in Ta2NiSe5 is estimated to have a monolayer thickness, suggesting that the EI phase is highly layer-confined and in the strongly coupled BEC limit.
Figure 1. (a) Crystal structure of Ta2NiSe5 viewed along the [100] (left) and [010] (right) directions. (b) Schematic representations of lattice distortion of the Ta and Ni chains below Tc. (c) Temperature-dependent resistivity of a bulk Ta2NiSe5 crystal. The inset shows its temperature derivative curve near Tc, indicated by the arrow. (d) Atomic force microscopy image of a typical ultrathin Ta2NiSe5 crystal. The corresponding optical image and the height profile along the horizontal line are shown in the inset.
RESULTS Ta2NiSe5 has a vdW structure consisting of 2D layers of a Ta double chain and a Ni single chain alternately aligning along the c-axis and running along the a-axis. The Ta−Ni layers are loosely bonded by vdW interaction along the b-axis (Figure 1a). For a bulk sample, the EI transition is accompanied by an orthorhombic-to-monoclinic transition27 with tilting of the Ta double chains against a Ni single chain (Figure 1b). This change induces a resistivity anomaly at Tc ≈ 325 K.27 The resistivity and its temperature derivative curves (Figure 1c) of our Ta2NiSe5 crystals show a clear kink, which confirms the presence of the EI transition. Figure 1d shows the optical and the atomic force microscopy (AFM) images for ultrathin Ta2NiSe5 crystals with a thickness down to four layers. The step height of individual layers, taken from AFM (Figure 1d), is d = 6.7(3) Å, which corresponds to the interlayer spacing, that is, half of the b-axis lattice constant (b = 12.829 Å).28 The optical contrast and the AFM thickness did not change noticeably even after 1 week in ambient conditions, as shown in the Supporting Information (SI). These results confirm that ultrathin and air-stable crystals can be successfully isolated from bulk Ta2NiSe5, which clearly manifests its vdW interlayer coupling. Raman spectra of ultrathin Ta2NiSe5 crystals with a thickness down to five layers are shown in Figure 2a. The crystals, thinner than four layers, yielded no measurable Raman signal. In the low-T monoclinic and the high-T orthorhombic phases, density functional theory (DFT) calculations predict 22 and 24 modes below 300 cm−1. Among these modes, some are close in frequency and are observed as a single peak in experiments due to a typical line broadening of ∼7 cm−1. Two strong Raman
Figure 2. (a) Raman spectra of ultrathin Ta2NiSe5 crystals at different thicknesses. The vertical black (red) bars at the bottom indicate the calculated frequencies of Raman-active phonon modes for the monoclinic (orthorhombic) phase below (above) Tc. The inset shows a typical Ta2NiSe5 crystal used for Raman spectroscopy with positions of different thicknesses where the laser was aimed. (b−e) Frequencies of Raman modes as a function of the number of layers. The insets show a schematic illustration of atomic displacement for each Raman-active phonon mode.
peaks at 96 and 122 cm−1 are present only in the monoclinic phase (T < Tc) and thus can identify the EI phase in ultrathin crystals. In all samples, Raman peaks, taken at room temperature, agree well with the calculation for the monoclinic phase (Figure 2a). Particularly, Raman peaks at 96 and 122 cm−1, a fingerprint of the EI phase, remain unchanged. The Raman peaks at 8889
DOI: 10.1021/acsnano.6b04796 ACS Nano 2016, 10, 8888−8894
Article
ACS Nano frequencies of 175 and 289 cm−1, having vibrational displacements mostly parallel and orthogonal to the layer, also do not show any systematic change in peak frequency and shape as the thickness decreases (Figure 2b−e). These observations suggest that the low-T monoclinic structure, which occurs below Tc, remains unchanged at room temperature even in a five-layer sample. In-plane resistivity measurements confirm the presence of the stable EI phase in ultrathin Ta2NiSe5 crystals. The resistivity was measured along the a-axis, that is, the Ta (or Ni) quasi-1D chain direction, in the samples down to eight layers. All samples exhibit semiconducting dependence, and the sheet resistance per monolayer, R□, remains almost the same (Figure 3a). The
the Arrhenius plot of the conductance G = R□−1 gave E0 = Eg/2 ≈ 0.14 eV (Figure 3c), which is consistent with the ARPES results.23,24 Employing this approach, we obtained the activation energy, E0, for each ultrathin crystal, as shown in Figure 3h. No systematic thickness dependence of E0 was observed, indicating that reducing thickness has a negligible effect on the renormalized band gap. This independence of the gap is further confirmed by STS on ultrathin Ta2NiSe5 crystals with different thicknesses, prepared on top of a graphene/SiC substrate. As shown in Figure 3e, the scanning tunneling microscopy (STM) image clearly resolves the atomic quasi-1D chain structures on the Ta2NiSe5 surface along the a-axis. This indicates that the crystal surface was well preserved during mechanical exfoliation of the sample and its transportation to the ultrahigh vacuum chamber, consistent with the observed stability in ambient conditions (see SI). The STS results, taken at T ≈ 295 K, are shown in Figure 3f for several ultrathin crystals with different thicknesses. The normalized differential conductance (dI/dV) curves for all of the samples have a common dip feature around zero bias, corresponding to the conduction band minimum. Another dip feature near bias voltage Vb ≈ −0.3 V corresponds to the valence band maxima. The small spectral weight at Vb ≈ −0.17 V may be due to the in-gap states from surface defects (see SI). All of these common features in the dI/dV curves do not vary for the different samples. The renormalized gap Eg ≈ 2Δ is determined by the energy difference where the exploration lines from the valence and conduction band slopes cross the minimum density of states near the Fermi levels (see SI). We observed that the estimated Δ from the normalized dI/dV curves are in good agreement with that obtained from the activation energy E0 (Figure 3h), showing an almost negligible dependence on thickness. The disorder strength also remains almost the same for ultrathin Ta2NiSe5 crystals. At low temperatures, charge conduction occurs by hopping between disorder-induced localized states near the Fermi level and is therefore a sensitive probe to measure the disorder strength. When the electron interaction is important, as in the case of Ta2NiSe5, a soft Coulomb gap is introduced in the density of states near the Fermi level, and the conduction is best described by the EfrosShklovskii variable-range hopping (ES-VRH) model,29,30 as expressed by
Figure 3. (a) Temperature dependence of the sheet resistance per monolayer (R□) for ultrathin Ta2NiSe5 crystals with different thickness, indicated by the same color code as in the inset. The inset shows the thickness dependence of R□ at 350 K (filled circles) and 60 K (open circles). (b) Temperature derivative curves of R□(T) near Tc, indicated by the arrows. The thickness is given with numbers in a unit of layer. (c) Arrhenius plot of the conductance G = R□−1 at T > Tc. The solid lines are the linear fit. (d) Variablerange hopping (VRH) plot of G(T) at low temperatures. Solid lines are the fit to the Efros−Shklovskii (ES)-VRH model. The inset shows optical image of a typical device. (e) Scanning tunneling microscopy (STM) image for the surface of the Ta2NiSe5 thin crystal with a thickness of 45 layers. The atomic chain direction (aaxis) is indicated by the arrow. The inset shows the corresponding large-scale STM image of the crystal on top of graphene substrate. (f) Normalized dI/dV curves for ultrathin crystals with different thicknesses. The valence band maxima and the conduction band minima are indicated by the arrows. (g) Thickness dependence of Tc. The gray line is a guide for the eye. (h) Thickness dependence of TES, E0, and Δ, estimated from (d), (c), and (f), respectively. The bulk value of Δ (open square), obtained from ARPES in ref 23, is also plotted for comparison.
⎛ T ⎞1/2 G ∝ exp⎜ − ES ⎟ ⎝ T ⎠
For bulk or relatively thick samples, for example, with ∼425 layers, deviation from the ES-VRH model is observed at low temperatures (see SI). However, samples with a thickness of
