Main

Most of the cuprate superconductors are either uniquely hole-doped (p-type) as in La2−xSrxCuO4, YBa2Cu3Oy and Bi2Sr2CaCu2O8+δ or electron-doped (n-type) as in Ln2−xCexCuO4(Ln=Nd,Pr). The main structural difference is the lack of apical oxygens in the n-type cuprates4,8. Recently a system has been found where the sign of the charge carriers can be flipped without altering the crystal structure9. In Y1−zLaz(Ba1−xLax)2Cu3Oy (YLBLCO), the substitution of La for Ba provides electrons and, for x=0.13 and z=0.62, the charge carriers can be controlled9 from 7% of holes per planar Cu (superconducting below 25 K) to 2% of electrons (non-superconducting) through the zero-doping state, by simply changing the oxygen content, which is done by removing oxygen from the Cu–O chain site; there is no difference in the chemical procedure for p- and n-type doping. Figure 1 shows the crystal structure and the scheme of chemical doping in YLBLCO, together with a generalized phase diagram of the cuprates. Note that in YLBLCO the apical oxygen is intact even in the n-type regime. To the best of our knowledge, YLBLCO is so far the only ambipolar cuprate that provides a unique opportunity to investigate the extremely low-doping region around the Mott insulating boundary spanning both p- and n-type regimes.

Figure 1: Schematics of YLBLCO.
figure 1

a, Crystal structure of YLBLCO and the sites for chemical doping. Note that in actual YLBLCO the Cu–O chains are fragmented and randomly oriented, leading to a macroscopically tetragonal structure. b, Generalized phase diagram of cuprate materials, where AF and SC denote antiferromagnetic and superconducting regions, respectively. Attainable doping ranges are shown for several cuprate materials including YLBLCO.

Full size image

Let us begin by showing how the resistivity changes on ambipolar doping. Figure 2a–c shows the temperature dependence of the c-axis resistivity, ρc, for various oxygen contents in YLBLCO from y=6.22 to 6.80, measured up to 500 K. The highest value of ρc at room temperature is achieved for y=6.32: ρc(300 K)≃105 Ω cm, which is unprecedented for cuprates. This suggests that the real boundary between n- and p-type cuprates, which is defined as the Mott-insulating state, is close to y=6.32. ρc does not change much for y=6.31–6.36 (Fig. 2b), and it starts to decrease for y<6.31 (Fig. 2c) due to the conductance of electron carriers9. A recent X-ray photoemission spectroscopy study10 clearly showed a chemical potential jump at y∼6.32 when y decreases, which confirms the change of carrier type at y∼6.32. Furthermore, a change in the Cu 2p X-ray photoemission spectroscopy peaks has been observed, which indicates that the electron doping indeed results in carriers in the CuO2 planes10.

Figure 2: Temperature dependence of ρc.
figure 2

a–c, ρc(T) of YLBLCO in semi-log plots, where 4W (four-wire) and 2W (two-wire) denote the methods for resistivity measurements. d–f, Temperature derivative of ρc(T) normalized by ρc. The arrows indicate the position of the peak or the dip, at which the Néel transition is supposed to take place.

Full size image

On close examination, it can be seen that ρc(T) exhibits a weak kink at each doping concentration. The anomaly can be seen more clearly, as either a peak or a dip, when the normalized temperature derivative of ρc(T), (dρ/dT)/ρ, is plotted in Fig. 2d–f. The dip, observed in p-type samples with y≥6.32 (Fig. 2d,e), is similar to the dip observed11 in the pristine YBa2Cu3Oy. On the other hand, in n-type samples (y≤6.30) a peak rather than a dip is observed (Fig. 2f). Below, our neutron scattering data will show that the anomaly is associated with magnetic ordering. However, the different kinks suggest an asymmetry between p- and n-type cuprates in the transport properties. Before we discuss the doping dependence of the transport properties, let us discuss how the magnetic ground state evolves when the nature of the charge carriers changes near the Mott critical point, y=6.32.

Elastic neutron scattering measurements were carried out on the same crystals as used in the ρc(T) measurements. Figure 3a–f shows the data taken along the [0.5,0.5,L] direction centred at L=1 and 1.5. For the electron-doped (n-type) sample with an oxygen content of y=6.22, at 7 K a sharp peak exists at the half-integer L (Fig. 3d), and no signal exists at the integer L (Fig. 3a). For the hole-doped (p-type) sample with y=6.69, on the other hand, no peak was observed at the half-integer L (Fig. 3f) and instead a peak appeared at the integer L (Fig. 3c). The magnetic ordering in both compositions occurs at ∼360 K (Fig. 3g,i), which is consistent with the temperature where ρc(T) and dρc/dT show the anomaly. The magnetic Bragg peaks were measured at several different corresponding wave vectors with half-integer Ls for y=6.22 and with integer Ls for y=6.69, and their relative intensities were used to determine their magnetic ground-state structures. For both compositions, the magnetic moments at the Cu sites in the CuO2 planes (called Cu(2)) are antiferromagnetically ordered in the horizontal-plane with an ordered moment 〈M〉 of ∼0.3 μB/Cu(2), represented by red and blue spheres in the insets of Fig. 3a,d. The orientation of the moments, on the other hand, could not be determined within our experimental accuracy (see the Supplementary Information for details). The Cu bilayers are coupled antiferromagnetically in both cases, and when translated by one lattice unit along the c axis the moments change sign for the n-type samples whereas they are the same for the p-type ones. The moments at the Cu sites in the Cu–O-chain layers (called Cu(1)) were small or negligible: 〈M〉Cu(1)∼0.04μB/Cu(1) and 0.00 μB/Cu(1) for y=6.22 and 6.69, respectively.

Figure 3: Magnetic ground states of YLBLCO studied by neutron scattering.
figure 3

a–f, Neutron scattering intensities obtained by [0.5,0.5,L] scans for y=6.22 (electron doped) (a,d), y=6.32 (nearly zero doping) (b,e) and y=6.69 (hole doped) (c,f) at different temperatures. The insets show schematics of the magnetic structures of YLBLCO for respective regimes. The red and blue spheres at the Cu(2) sites stand for spins with opposite directions (the exact direction of the spin easy axis is unknown), and grey spheres are the Cu(1) atoms. The small (or negligible) moments of the Cu(1) atoms are not plotted. The letters A and B denote the difference (π-phase shift) in spin arrangements in the plane. g–i, Temperature dependencies of the integrated intensity of the magnetic reflections for y=6.22 (g), 6.32 (h) and 6.69 (i). The inset to h magnifies the low-temperature data for y=6.32. The error bars represent one standard deviation.

Full size image

To confirm that the observed magnetic structures are the generic ground states for the n- and p-type YLBLCO and to study how one magnetic structure evolves into another when the oxygen concentration increases, we carried out similar measurements on YLBLCO (y=6.32). The oxygen content of y=6.32 was chosen because this is the closest to the Mott critical point that separates the n- and p-type regions; indeed, the Hall coefficient RH shows a sign change across this composition (see Fig. 4b). As shown in Fig. 3b,e,h, in this case on cooling the integer L peak first develops below ∼370 K but on further cooling it starts to decrease at ∼20 K below which the half-integer L peak appears. A similar change in the magnetic structure on cooling was observed in Nd2CuO4 where the ordering of Nd moments and the Nd–Cu interactions at low temperatures induce a second magnetic transition12. For YLBLCO (y=6.32), the ordered moments of the Cu(2) sites were ∼0.24 and 0.26 μB/Cu(2) for the half-integer L and the integer L phase, respectively, whereas those of the Cu(1) sites were 0.002 and 0.00 μB/Cu(1) for the half-integer L and the integer L phase, respectively (see the Supplementary Information for details.) In the present case, the interaction between the moment on Cu(1) and Cu(2) is likely to be responsible for the spin reorientation similar to the case of Nd2CuO4. Besides, the crossover from the integer L to the half-integer L peak on cooling probably indicates that the sign of the dominant charge carriers in the y=6.32 sample changes at ∼20 K, although RH could not be measured on this sample because of a very high impedance. This also indicates that the observed magnetic structures are indeed the generic ground states for the electron- and hole-doped YLBLCO.

Figure 4: Evolution of key parameters in YLBLCO.
figure 4

a, ρc shows a peak at y=6.32, signifying the zero-doping point. b, The carrier concentration decreases exponentially towards the zero-doping point as y is varied; here, the RH data at 300 K are used for the calculations, except for y=6.36 and 6.49 for which 350 K data are used. Light red and blue backgrounds denote the n- and p-type regimes, respectively. The carrier mobility along the c axis calculated from these data is 10−3–10−2 cm2 V−1 s−1 and is nearly independent of both the concentration and the type of carriers. c, TN as a function of the concentration of both p- and n-type carriers (solid lines are guides to the eye), together with the anticipated phase boundary for the regimes not covered in this study (dashed lines). Note that the horizontal axis is plotted in a logarithmic scale, and the region studied here is highlighted with light colours. TN is extracted from the ρc(T) data shown in Fig. 2d–f. The carrier concentrations for y=6.30, 6.31 and 6.32 are estimated from ρc by extrapolating the relationship between ρc and RH.

Full size image

What causes the magnetic ground state to change abruptly when the sign of the charge carrier changes? We can rule out any crystal structural origin because no abrupt change in the crystal structure including the lattice constants was found at the n-to-p-type crossover. Experimentally, it seems that the magnetic moments at the Cu(1) sites in the Cu–O-chain layers play a crucial role in selecting the particular magnetic ground state. For y=6.22 (n-type), the small but non-zero moments at the Cu(1) planes would favour the observed Qm=(0.5,0.5,0.5) magnetic structure along the c axis if they are coupled antiferromagnetically with the neighbouring Cu(2) layers. For y=6.69 (p-type), on the other hand, the non-magnetic Cu(1) layers would not have such an effect, leading to the observed Qm=(0.5,0.5,1) magnetic structure13. Note that the Cu(1) ions become magnetically ordered when excess electrons are doped rather than when excess holes are doped, even though the electron doping would decrease the number of magnetic Cu2+ ions at the Cu(1) site14. The observed magnetic ground states for the p- and n-type regimes may provide a possible explanation for why dρc/dT at TN shows a dip (increase in ρc below TN) for the p-type and a ‘peak’ for the n-type (decrease in ρc below TN) (Fig. 2d–f): in the p-type magnetic structure, when a carrier hops from one bilayer to another bilayer it sees opposite Cu spins (see the inset of Fig. 3a), whereas in the n-type magnetic structure it sees parallel Cu spins (inset of Fig. 3d). Thus, a sort of ‘spin-valve effect’ may yield15 a higher hopping probability for the n-type than for the p-type, which is consistent with our transport data.

In Fig. 4a, ρc is plotted as a function of the oxygen content at three different temperatures. It clearly shows that YLBLCO realizes the Mott-insulating state at y≃6.32. When electrons are added or subtracted from this Mott state, ρc decreases but with a faster rate for n-type cuprates and a slower rate for p-type cuprates. The carrier concentration nc calculated from the Hall coefficient RH is plotted in Fig. 4b as a function of y, which confirms both the sign change of the main carriers near y=6.32 and the different rate of charge transfer for the two regimes. It is interesting that this doping asymmetry seems to mimic the spectral-weight-transfer asymmetry expected for cuprates: because the cuprates are charge-transfer-type Mott insulators16, the valence and conduction bands are primarily oxygen and copper bands, and their spectral weights are 2N and N, respectively, if the system has N cells (this is because the original spectral weight 2N of copper is split into the upper and lower Hubbard bands as a result of the on-site repulsion U); when an electron is added to the insulating state, in the localized limit it creates one doubly occupied copper site on which the two electrons have the same energy, so the new filled state has the spectral weight of two, whereas that of both the upper and lower Hubbard bands becomes N−1; on the other hand, doping a hole simply reduces the spectral weight of the oxygen band to 2N−1. This means that the rate of the spectral-weight transfer on doping is twice as quick for electron doping, which is an inherent asymmetry16 in the localized limit. The observed doping-rate asymmetry (Fig. 4a,b) may be related to the fundamental spectral-weight-transfer asymmetry in Mott-insulating cuprates.

Figure 4c shows the nc-versus-TN phase diagram of YLBLCO. In the n-type regime, the half-integer L magnetic state is stable over a very wide range of nc spanning five orders of magnitude, with the maximum TN at nc∼10−5 per planar Cu atom. When nc approaches the zero-doping Mott-insulator point, TN decreases. When the charge carrier becomes p-type, TN starts increasing again as nc increases up to ∼10−5 per planar Cu atom beyond which TN decreases. The unexpected dip in TN at the zero-doping point cannot be explained by a simple argument. Furthermore, the magnetic structure changes between p- and n-types in response to as small as 0.1 ppm of carriers, which is surprising. The ‘competition’ between the two magnetic structures, which can be inferred in the y=6.32 sample in its temperature dependence, may be responsible for the observed intriguing phenomena near the zero-doping state. Clearly, the physics of very dilute carriers in Mott insulators is more intricate than expected, and its understanding may help disentangle the complex physics17 of the cuprates.

Methods

Single crystals of YLBLCO are grown by a flux method using 99.9%-pure Y2O3 crucibles. The chemical composition of the grown crystals is analysed by inductively coupled plasma atomic emission spectroscopy, which gives z=0.62 and x=0.13. The carrier concentration in YLBLCO can be controlled from ∼2% of n-type carriers to ∼7% of p-type ones through the Mott insulator by changing the oxygen content y from 6.21 to 6.95 (ref. 9). Reducing y to below 6.21 results in decomposition of the crystal. ρc is measured by the four-wire method for resistance below ∼100 kΩ, and by the two-wire method for higher resistance up to ∼1011 Ω, where an electrometer is used with the guarding technique. All of the measurements of ρc are carried out on the same two crystals while varying the oxygen content; removable electrodes are used and the crystals are reannealed after each measurement. The error in the absolute value of the resistivity in the two crystals is within 20% for the four-wire method, and 50% for the two-wire method. The Hall coefficient is measured by sweeping the magnetic field applied along the c axis up to 7 T, and the error can be up to 20% for high-resistance compositions. The elastic neutron scattering measurements were carried out on the cold neutron triple-axis spectrometer SPINS at the National Institute of Standards and Technology. The neutron energy was fixed at 3.7 meV. Contamination from higher-order beams was effectively eliminated using Be and BeO filters before and after the sample, which was 0.5×0.5×0.2 mm3 in size. For each sample, elastic scans were carried out at several different magnetic Bragg reflections. The integrated intensities were normalized in an absolute unit using the integrated intensity of the nuclear (003) Bragg reflection, and used for determining the magnetic moments of the Cu(1) and Cu(2) ions. See the Supplementary Information for details.