Electric double-layer transistor using layered iron selenide Mott insulator TlFe1.6Se2

A_1–xFe_2–ySe₂ (A = K, Cs, Rb, Tl) are recently discovered iron-based superconductors with critical temperatures (T_c) ranging up to 32 K. Their parent phases have unique properties compared with other iron-based superconductors; e.g., their crystal structures include ordered Fe vacancies, their normal states are antiferromagnetic (AFM) insulating phases, and they have extremely high Néel transition temperatures. However, control of carrier doping into the parent AFM insulators has been difficult due to their intrinsic phase separation. Here, we fabricated an Fe-vacancy-ordered TlFe_1.6Se₂ insulating epitaxial film with an atomically flat surface and examined its electrostatic carrier doping using an electric double-layer transistor (EDLT) structure with an ionic liquid gate. The positive gate voltage gave a conductance modulation of three orders of magnitude at 25 K, and further induced and manipulated a phase transition; i.e., delocalized carrier generation by electrostatic doping is the origin of the phase transition. This is the first demonstration, to the authors'' knowledge, of an EDLT using a Mott insulator iron selenide channel and opens a way to explore high T_c superconductivity in iron-based layered materials, where carrier doping by conventional chemical means is difficult.Layered copper-based oxides (cuprates) and iron-based superconductors are the most well-known types of superconductor because of their high critical temperatures (T_c) of more than 50 K (1, 2). One common feature of these materials is that superconductivity emerges when the long-range antiferromagnetic (AFM) order in the parent phase is suppressed and varnished by carrier doping. However, the maximum T_c of the cuprates (134 K for HgBa₂Ca₂Cu₃O_8+δ) (3) is much higher than that of the iron-based materials [55 K for SmFeAs(O_1–xF_x)] (4). This difference is related to the different electron correlation interactions of the materials; i.e., the parent phases of the cuprates are AFM Mott insulators, where the electron–electron Coulomb interaction is very strong, whereas the iron-based parent phases are AFM metals with weaker electron correlation. Related to this difference, the Néel transition temperatures (T_N) of the cuprate parent phases (e.g., 420 K for YBa₂Cu₃O₆) (5) are much higher than those of the iron-based phases (e.g., 140 K for SmFeAsO) (6). Based on the high T_c cuprate scenario, it is considered that a key strategy to obtaining higher T_c in the iron-based superconductors is to dope carriers into the parent phase of a Mott insulator with a higher T_N and then disperse the magnetic order by carrier doping.Recently, superconductivity was discovered at 32 K in a layered iron selenide, K_0.8Fe₂Se₂ (7), which has drawn considerable attention, because this is the first material with a parent phase to exhibit an AFM insulating state among the iron-based superconductors (8). The A_1–xFe_2–ySe₂ (A = K, Cs, Rb, Tl) system has the same crystal structure as the 122-type iron-based superconductor BaFe₂As₂ with a tetragonal ThCr₂Si₂-type structure (9) (Fig. 1A), and is composed of alternately stacked A and FeSe layers along the c axis. The Fe atoms in the FeSe layer form a square lattice, where the Se atoms are located at the apical sites of the edge-shared FeSe₄ tetrahedra. However, the ideal chemical formula is A₂Fe₄Se₅ (245 phase), which satisfies charge neutrality conditions with consideration of formal ion charges (i.e., +1 for A, +2 for Fe, and −2 for Se); therefore, the A and Fe sites include vacancies in the ThCr₂Si₂-type structure. It was shown that the Fe vacancies (V_Fe) in the parent 245 phase exhibit an order–disorder transition at ∼500 K and form a √5 × √5 × 1 supercell (the unit cell formula is A₈Fe₁₆Se₂₀ with four V_Fe) (10), as shown in Fig. 1. Theoretical calculations suggested that the parent 245 phase is a Mott insulator with a Mott gap of ∼100 meV (11, 12). The gap was confirmed experimentally to be ∼430 meV (13). The 245 Mott insulator exhibits an AFM long-range order with T_N as high as 470−560 K, similar to that of the cuprates, along with an ordered magnetic moment of more than 3 Bohr magneton (μ_B) at 10 K (10). Because of this similarity to cuprates (i.e., a Mott insulator with high T_N), it is expected that a much higher T_c would be realized in a V_Fe-ordered AFM insulator of the 245 phase if the AFM order is suppressed by carrier doping.Open in a separate windowFig. 1.Crystal structures of 122-type A_1–xFe_2–ySe₂ (A = K, Cs, Rb, Tl) and 245-type parent-phase A₂Fe₄Se₅ viewed along the [120] (A) and the [001] (B) directions. The spheres represent A (gray), Fe (orange), Se (blue), and Fe vacancy sites (V_Fe, green). The 122-type tetragonal fundamental cell and the 245-type √5 × √5 × 1 supercell are indicated by the red and black lines, respectively.Indeed, chemical electron-doping of the 245 phase has been performed by reducing the V_Fe (2 − y > 1.93), which induced superconductivity at ∼30 K (8). However, it has been reported that these A_1–xFe_2–ySe₂ superconductors intrinsically include phase separation into the superconducting phase, which is believed to exist in a phase without the V_Fe order, and the AFM-insulating phase (V_Fe-ordered 245 phase) (13). Further, there is controversy as whether the superconducting phase is an intercalated phase like Rb_0.3Fe₂Se₂ (14), a V_Fe-free phase like KFe₂Se₂ (13), or this disordered V_Fe phase (15). The coexistence of such multiphases indicates that a well-controlled carrier doping structure of the 245 phase has yet to be realized by chemical-composition doping or substitution, and the carrier doping effects are not yet clear. In contrast, carrier doping by an electrostatic method that uses a field-effect transistor structure is free from this structural alternation and would be suitable for study of phase transitions in V_Fe-ordered 245 Mott insulators that have never become a superconductor.In this study, we focused on the epitaxial films of one iron selenide Mott insulator, TlFe_1.6Se₂, because TlFe_2–ySe₂ is much more stable in air than the other A_1–xFe_2–ySe₂ (A = K, Rb, Cs) (16), and a fully V_Fe-ordered phase with high chemical homogeneity has been obtained in single crystals due to the lower vapor pressure of Tl than those of alkaline metals (17). In addition, the number of V_Fe in TlFe_2–ySe₂ cannot be controlled over a wide range (the maximum 2 − y value is limited to only 1.6) regardless of the starting nominal compositions, and thus superconductivity has not previously been observed in TlFe_1.6Se₂ (16), although a bulk superconductivity was observed in mixed (Tl,K)_1−xFe_2−ySe₂ (18). These features demonstrate that the TlFe_1.6Se₂ AFM insulator is the most ideal target to examine electrostatic carrier doping. We therefore used an electric double-layer transistor (EDLT) structure because the ionic liquid gate works as a nanometer-thick capacitor with a large capacitance and provides an effective way to accumulate a very high carrier density [maximum sheet carrier density of ∼10¹⁵ cm^–2 under small gate voltages (V_G) of around ±3 V]. This high carrier modulation by the EDLT can alter electronic states over a very wide range and convert even a band insulator into a metal, and even further, into a superconductor (19); similarly, a Mott insulator can be converted into a metal (20). We therefore expected that the EDLT structure would also modulate the carrier density sufficiently to induce a phase transition such as superconductivity in the iron selenide Mott insulator without chemical doping or structural alternation. Here, we used a single phase (i.e., homogeneous in structure, chemical composition, and vacancy distribution) and V_Fe-ordered TlFe_1.6Se₂ insulating epitaxial film grown by pulsed laser deposition (PLD) with an atomically flat surface as the transport channel layer of the EDLT. Large field-effect current modulation was demonstrated in the EDLT, particularly at low temperatures. The electric field clearly decreased the activation energy (E_a) and also induced a phase transition.Fig. 2A shows the out-of-plane X-ray diffraction (XRD) pattern of a TlFe_1.6Se₂ thin film grown at the optimum temperature of 600 °C. Only the sharp peaks of the 00l diffractions of the TlFe_1.6Se₂ phase were observed, along with those of the CaF₂ substrate, indicating that the film grows along the [00l] direction. Although the in-plane lattice parameter of CaF₂ (a/√2 = 0.386 nm) is almost the same as that of (La,Sr)(Al,Ta)O₃ (LSAT) (a/2 = 0.387 nm), the full width at half maximum (FWHM) values of the 004 rocking curve (Δω) of the film are much smaller when grown on the CaF₂ substrate (0.08°) than that grown on the LSAT substrate (0.8°) (Fig. 2B). This suggests that an interface reaction occurs on the oxide LSAT substrate, whereas the fluoride CaF₂ substrate is more suitable for TlFe_1.6Se₂; similar results are reported also for iron chalcogenide FeSe_0.5Te_0.5 epitaxial films (21). To confirm the epitaxial relationship between the TlFe_1.6Se₂ film and the CaF₂ substrate, asymmetric diffractions were measured. Fig. 2C shows the results of ϕ scans of the 123 diffraction of the TlFe_1.6Se₂ film and the 202 diffraction of the CaF₂ substrate. Both peaks appear every 90° and exhibit fourfold symmetry, substantiating the heteroepitaxial nature of the TlFe_1.6Se₂ film growth on the CaF₂ substrate. Each peak (FWHM value = 0.2°) of the TlFe_1.6Se₂ film is rotated by 45° with respect to the peaks of the CaF₂ substrate, showing that the TlFe_1.6Se₂ film grows on the CaF₂ substrate with epitaxial relationships of [001] TlFe_1.6Se₂//[001] CaF₂ (out of plane) and [310] TlFe_1.6Se₂//[100] CaF₂ (in plane). These epitaxial relationships are a natural consequence of the smallest in-plane lattice mismatching [Δ(d_Tl–Tl − d_Ca–Ca)/d_Ca–Ca × 100 = 0.8%] as shown in Fig. 2D. Fig. 2E shows the surface morphology of the TlFe_1.6Se₂ epitaxial film on the CaF₂ substrate. A flat surface with a step-and-terrace structure (root-mean-square roughness of 1.4 nm) was observed, indicating the layer-by-layer growth of TlFe_1.6Se₂ epitaxial films under optimized conditions. This result is also consistent with the observation of Pendellösung interference fringes (Fig. 2A, Inset). The step height (Fig. 2F) observed by atomic force microscopy is ∼0.7 nm, which agrees well with the distance between the nearest-neighbor FeSe–FeSe layers [corresponding to a half unit of the c-axis length (1.397 nm) of the TlFe_1.6Se₂ unit cell indicated in Fig. 1A]. These results guarantee that the TlFe_1.6Se₂ epitaxial film is of sufficiently high quality to be used for the EDLT transport channel.Open in a separate windowFig. 2.(A) Out-of-plane XRD pattern of a TlFe_1.6Se₂ film grown on a CaF₂ (001) substrate. (A, Inset) The magnified pattern around the 006 diffraction. The vertical lines indicate the positions of the Pendellösung interference fringes. (B) Rocking curves of the 004 diffractions of the TlFe_1.6Se₂ films on CaF₂ and LSAT substrates. (C) ϕ scans of the 123 diffraction of the TlFe_1.6Se₂ film and the 202 diffraction of the CaF₂ substrate. (D) In-plane atomic configuration of the TlFe_1.6Se₂ epitaxial film on the CaF₂ substrate. (E) Topographic atomic force microscopy image of the surface of the TlFe_1.6Se₂ epitaxial film on the CaF₂ substrate. (F) Height profile across the line A–B shown in E.The atomic structure and V_Fe ordering in the TlFe_1.6Se₂ epitaxial film were examined by high-angle annular dark field scanning transmission electron microscopy (HAADF-STEM) and selected area electron diffraction (SAED). Fig. 3 A and B show the plan-view HAADF-STEM images of the TlFe_1.6Se₂ epitaxial film. V_Fe are detected as the dark regions due to the enhanced atomic number contrast of HAADF, showing the long-range periodic V_Fe ordering. In addition, superlattice diffractions due to the V_Fe ordering, similar to that of TlFe_1.6Se₂ single crystal (17), were observed in the SAED pattern Fig. 3C as indicated by q₁ and q₂. These results substantiate that the present sample is of a highly V_Fe-ordered phase. Fig. 3D demonstrates the arrangement of V_Fe more clearly by superimposing yellow lines on the HAADF-STEM image of Fig. 3A. Fully ordered V_Fe are dominant in almost the whole region, while small phase separation to disordered-V_Fe regions ≤5 nm in size (the unmarked regions in Fig. 3D) were also observed by keeping the perfect coherency of the fundamental crystal structure.Open in a separate windowFig. 3.(A) [001] plan-view HAADF-STEM image of TlFe_1.6Se₂ epitaxial film. (B) Magnified HAADF-STEM image of the yellow square region in A. Vertical yellow arrows indicate the V_Fe sites with dark contrast. (B, Inset) The crystal structure of TlFe_1.6Se₂, where only Tl and Fe sites are shown because the positions of Se and Tl sites overlap over them (Fig. 1B). The square shows the superlattice unit cell, where V_Fe are shown (green circles). (C) The SAED pattern with electron beam along [001]. Two superlattice reciprocal vectors due to V_Fe ordering are indexed by q₁ and q₂. (D) Small green circles indicate all detected V_Fe, and yellow lines indicate V_Fe arrangement.A 20 nm-thick TlFe_1.6Se₂ epitaxial film was used as the EDLT transport channel. Fig. 4 shows a schematic illustration of the EDLT, in which a six-terminal Hall bar channel and Au pad electrodes were formed using shadow masks. After pouring the ionic liquid, N,N-diethyl-N-methyl-N-(2-methoxyethyl)-ammonium bis-(trifluoromethylsulfonyl) imide (DEME-TFSI), into a silica glass cup, a Pt coil electrode was inserted into the ionic liquid to act as a gate electrode. The transfer curves [V_G dependence of drain current (I_D)] at a drain voltage (V_D) of +0.3 V and output curves (I_D vs. V_D under various V_G) of the EDLT were then measured.Open in a separate windowFig. 4.Schematic of the EDLT using TlFe_1.6Se₂ epitaxial film with a six-terminal Hall bar structure on a CaF₂ substrate. V_G was applied via a Pt counter electrode through the ionic liquid, DEME-TFSI, contained in a silica glass cup. Electrical contacts were formed using Au wires and In/Au pads.Fig. 5A shows the cyclic transfer characteristics (I_D vs. V_G) of the TlFe_1.6Se₂ EDLT at T = 280 K. A positive V_G of up to +4 V was applied to the Pt coil gate electrode, which accumulates electrons at the interface. When V_G = +1.7 V was applied, I_D began to increase. The maximum I_D in the transfer curve reached 12 μA at V_G = +4 V, along with a small on:off ratio of 1.5. The gate leakage current (I_G) (Fig. 5A, Lower) also increased at V_G up to +4.0 V but was clearly smaller than I_D in the whole V_G region. After applying V_G = +4 V, I_D recovered to the initial values of 8 μA when V_G was decreased to 0 V. The large hysteresis loop is observed due to the slow response of ion displacement in the ion liquid. Probably due to the same reason, some parallel shift remains in the second I_D−V_G loop; however, the shape and the hysteresis width are very similar to those of the first loop, guaranteeing the observed results are reversible and reproducible. These results demonstrate the electrostatic nature of carrier accumulation. In the output characteristics (Fig. 5B), the conductance dI_D/dV_D increased with increasing V_G at ≥ +2.0 V. Two output characteristics, which were measured before and after the transfer curve measurements in Fig. 5A, remain unchanged, which further guarantees the reversibility of the EDLT characteristics. However, the I_D modulation is small at 280 K because of the high conductance at V_G = 0, which originates from the highly naturally doped carriers in the TlFe_1.6Se₂ film, as reported for a TlFe_1.6Se₂ bulk crystal in which the carrier density was estimated to be ∼5 × 10²¹ cm^–3 at T = 150 K (16). Using the reported gate capacitance value of ∼10 μF/cm² (22), the maximum accumulated carrier density is estimated to be 2.5 × 10¹⁴ cm^–2 at V_G = 4 V, and the field-effect mobility in the linear region of the output characteristics is estimated to be 0.18 cm²/(V·s) at V_G = +4 V. To estimate the carrier density induced in the TlFe_1.6Se₂ EDLT, we performed Hall effect measurements by applying magnetic fields of up to 9 T at temperatures between 300 and 25 K, but the Hall voltages (V_xy) obtained were below the detection limit of our measurement system. This suggests that the Hall mobility is smaller than 0.02 cm²/(V·s), which is roughly consistent with the small field-effect mobility above. Fig. 5C plots the V_G dependence of the sheet conductance (G_s) at T = 300–25 K. The V_G dependences of G_s were reversible also against repeated variation of measurement temperature (compare the open symbols and the filled symbols in Fig. 6A). With decreasing T, G_s at V_G = 0 V steeply decreased from 2.2 × 10⁻⁵ to 1.5 × 10⁻⁸ S because of the decrease in carrier density. It should be noted that large G_s modulation with gains of three orders of magnitude was demonstrated at T = 25 K.Open in a separate windowFig. 5.(A) Transfer characteristics (I_D–V_G) at V_D = +0.3 V and T = 280 K cyclically measured for two loops. Arrows indicate the V_G sweep directions, and triangles show the positions where I_D begins to increase. I_G vs. V_G is also shown at the bottom. (B) Output characteristics (I_D–V_D) at V_G = + 0−4 V and T = 280 K, measured before (Upper) and after (Lower) the transfer characteristics measurement in A. (C) V_G dependence of G_s measured with decreasing T (open symbols) and increasing T (filled symbols) over the 25–300 K range. The solid lines are visual to show the change in G_s clearly.Open in a separate windowFig. 6.(A) T dependences of R_s for the TlFe_1.6Se₂ EDLT measured with increasing T (open symbols) and decreasing T (filled symbols) at V_G = 0→+2.0→+4.0→0 V. The arrows indicate the positions of resistance humps. The reported ρ–T curves of (Tl,K) Fe_2–ySe₂ bulk materials (18) are shown for comparison (Inset). A resistance hump appears at 2 − y ≥ 1.68, and superconductivity emerges at 2 − y ≥ 1.76. (B) The E_a estimated from A in the high T region as a function of V_G.Fig. 6A shows the T dependences of the sheet resistance R_s (R_s–T) for the TlFe_1.6Se₂ EDLT at V_G = 0, +2, and +4 V. The R_s–T characteristics from 300 to 30 K at V_G = 0 V indicate simple thermally activated behavior, given by R_s = R_s0 exp(E_a/k_BT) (where R_s0 is a constant and k_B is the Boltzmann constant). This trend is similar to the R_s–T behavior of insulating TlFe_2–ySe₂ single crystals with 2 − y < 1.5 (18), but the resistivity anomaly due to a magnetic phase transition of spin reorientation at 100 K observed in the TlFe_1.6Se₂ single crystal (16, 17) was not detected. The E_a value of the TlFe_1.6Se₂ EDLT at V_G = 0 obtained from the Arrhenius fitting is 20 meV (Fig. 6B). This value is smaller than the 57.7 meV of the TlFe_1.47Se₂ single crystal (18) but almost double that of TlFe_1.6Se₂ single-crystals (11 meV) (16). These observations indicate the naturally doped carrier density should be smaller than that of the TlFe_1.6Se₂ single crystals. On the other hand, E_a is far smaller than the calculated (∼100 meV) (11, 12) and experimentally measured Mott gaps (∼430 meV) (13), which suggests that the TlFe_1.6Se₂ film is doped with carriers.The R_s–T behavior largely varied with V_G, particularly in the low T region, which is seen also in Fig. 5C. The R_s–T curves were reversible in the cooling and heating cycles. The E_a value estimated in the high T region decreased from ∼20 meV at V_G = 0−2 V to 8.9 meV at V_G = +4 V (Fig. 6B); i.e., R_s at V_G = +4 V is almost independent of T, indicating that a highly accumulated channel was formed by the application of V_G. The R_s–T curves at V_G ≥ 2 V do not show a simple thermally activated behavior and exhibit humps at T = 55 K for V_G = +2 V and at 40 K for V_G = +4 V (indicated by the vertical arrows in Fig. 6A). That is, when V_G was increased from 0 to +2 V, the resistance hump appeared at T_hump = 55 K, and R_s increased steeply again at T ≤ 31 K. At V_G = +4 V, T_hump shifted to a lower T (40 K), and R_s leveled off in the further lower T region. It would be possible to consider that the resistivity humps are attributed to a precursory phenomenon of a metal–insulator (MI) transition because E_a decreased sharply as confirmed in Fig. 6B. As for the A_1–xFe_2–ySe₂ superconductor single crystals, they also exhibit resistance humps and cross-overs from an insulating state to a metallic state at T_hump, and finally to a superconducting state (18). In the case of (Tl,K) Fe_2–ySe₂ in the literature (Fig. 6A, Inset) (18), the resistance hump appears at 2 − y = 1.68, and the superconductivity appears at 2 − y ≥ 1.76. This value is 16% larger than that of the TlFe_1.6Se₂ film in this work. This preceding work suggests that the T_hump observed in this study can also be related to the MI transition and superconductivity; however, in this study we could not observe superconductivity up to the maximum V_G of +4 V.A similar phenomenon, which is attributed to a magnetic phase transition, has been observed also in fully V_Fe-ordered (100 K) (17) and multiphase (100–150 K) (23) TlFe_1.6Se₂ single crystals. On the other hand, the resistance humps are attributed to the formation of an orbital-selective Mott phase (OSMP) for the V_Fe-poor A_1−xFe_2–ySe₂ bulk crystals, where one of the Fe 3d orbitals, d_xy, remains localized and the other four orbitals are delocalized (24, 25). The Mott insulator phase dominates the resistance above T_hump, whereas the OSMP prevails below it; finally, superconductivity appears below T_hump of the OSMP transition at higher carrier doping levels. Due to these similarities, we consider that the resistance humps observed in this study are more likely assigned to the same origin of magnetic phase transition or OSMP.In summary, electrostatic carrier doping of the Mott insulator iron selenide V_Fe-ordered TlFe_1.6Se₂ by the EDLT structure was demonstrated using a single-phase epitaxial film with an atomically flat surface grown on a CaF₂ substrate. The EDLT structure, based on an ionic liquid gate, successfully controlled the conductance and induced the phase transition assignable to a magnetic phase transition or OSMP. This demonstration of carrier doping of the Mott insulator iron selenide by the electrostatic method offers a way to extend the exploration of high T_c superconductors even to insulating materials, in which chemical doping methods do not work.