A superconductor is a material that can conduct electricity without resistance below a superconducting transition temperature, Tc. The highest Tc that has been achieved to date is in the copper oxide system1: 133 kelvin at ambient pressure2 and 164 kelvin at high pressures3. As the nature of superconductivity in these materials is still not fully understood (they are not conventional superconductors), the prospects for achieving still higher transition temperatures by this route are not clear. In contrast, the Bardeen–Cooper–Schrieffer theory of conventional superconductivity gives a guide for achieving high Tc with no theoretical upper bound—all that is needed is a favourable combination of high-frequency phonons, strong electron–phonon coupling, and a high density of states4. These conditions can in principle be fulfilled for metallic hydrogen and covalent compounds dominated by hydrogen5,6, as hydrogen atoms provide the necessary high-frequency phonon modes as well as the strong electron–phonon coupling. Numerous calculations support this idea and have predicted transition temperatures in the range 50–235 kelvin for many hydrides7, but only a moderate Tc of 17 kelvin has been observed experimentally8. Here we investigate sulfur hydride9, where a Tc of 80 kelvin has been predicted10. We find that this system transforms to a metal at a pressure of approximately 90 gigapascals. On cooling, we see signatures of superconductivity: a sharp drop of the resistivity to zero and a decrease of the transition temperature with magnetic field, with magnetic susceptibility measurements confirming a Tc of 203 kelvin. Moreover, a pronounced isotope shift of Tc in sulfur deuteride is suggestive of an electron–phonon mechanism of superconductivity that is consistent with the Bardeen–Cooper–Schrieffer scenario. We argue that the phase responsible for high-Tc superconductivity in this system is likely to be H3S, formed from H2S by decomposition under pressure. These findings raise hope for the prospects for achieving room-temperature superconductivity in other hydrogen-based materials.
A search for high- (room)-temperature conventional superconductivity is likely to be fruitful, as the Bardeen–Cooper–Schrieffer (BCS) theory in the Eliashberg formulation puts no apparent limits on Tc. Materials with light elements are especially favourable as they provide high frequencies in the phonon spectrum. Indeed, many superconductive materials have been found in this way, but only a moderately high Tc = 39 K has been found in this search (in MgB2; ref. 11).
Ashcroft5 turned attention to hydrogen, which has very high vibrational frequencies due to the light hydrogen atom and provides a strong electron–phonon interaction. Further calculations showed that metallic hydrogen should be a superconductor with a very high Tc of about 100–240 K for molecular hydrogen, and of 300–350 K in the atomic phase at 500 GPa (ref. 12). However, superconductivity in pure hydrogen has not yet been found, even though a conductive and probably semimetallic state of hydrogen has been recently produced13. Hydrogen-dominated materials such as covalent hydrides SiH4, SnH4, and so on might also be good candidates for showing high-Tc superconductivity6. Similarly to pure hydrogen, they have high Debye temperatures. Moreover, heavier elements might be beneficial as they contribute to the low frequencies that enhance electron–phonon coupling. Importantly, lower pressures are required to metallize hydrides in comparison to pure hydrogen. Ashcroft’s general idea was supported in numerous calculations7,10 predicting high values of Tc for many hydrides. So far only a low Tc (∼17 K) has been observed experimentally8.
For the present study we selected H2S, because it is relatively easy to handle and is predicted to transform to a metal and a superconductor at a low pressure P ≈ 100 GPa with a high Tc ≈ 80 K (ref. 10). Experimentally, H2S is known as a typical molecular compound with a rich phase diagram14. At about 96 GPa, hydrogen sulphide transforms to a metal15. The transformation is complicated by the partial dissociation of H2S and the appearance of elemental sulfur at P > 27 GPa at room temperature, and at higher pressures at lower temperatures14. Therefore, the metallization of hydrogen sulphide can be explained by elemental sulfur, which is known to become metallic above 95 GPa (ref. 16). No experimental studies of hydrogen sulphide are known above 100 GPa.
In a typical experiment, we performed loading and the initial pressure increase at temperatures of ∼200 K; this is essential for obtaining a good sample (Methods). The Raman spectra of H2S and D2S were measured as the pressure was increased, and were in general agreement with the literature data17,18 (Extended Data Fig. 1). The sample starts to conduct at P ≈ 50 GPa. At this pressure it is a semiconductor, as shown by the temperature dependence of the resistance and pronounced photoconductivity. At 90–100 GPa the resistance drops further, and the temperature dependence becomes metallic. No photoconductive response is observed in this state. It is a poor metal—its resistivity at ∼100 K is ρ ≈ 3 × 10−5 ohm m at 110 GPa and ρ ≈ 3 × 10−7 ohm m at ∼200 GPa.
During the cooling of the metal at pressures of about 100 GPa (Fig. 1a) the resistance abruptly drops by three to four orders of magnitude, indicating a transition to the superconducting state. At the next increase of pressure at low temperatures of T < 100 K, Tc steadily increases with pressure. However, at pressures of >160 GPa, Tc increases sharply (Fig. 1b). As higher temperatures of 150–250 K were involved in this pressure range, we supposed that the increase of Tc and the decrease of sample resistance during warming (Fig. 1a) could indicate a possible kinetic-controlled phase transformation. Therefore in further experiments, after loading and after the initial pressure increase at 200 K, we annealed all samples by heating them to room temperature (or above) at pressures of >∼150 GPa (Fig. 2a, see also Extended Data Fig. 2). This allowed us to obtain stable results, to compare different isotopes, to obtain the dependence of Tc on pressure and magnetic field, and to prove the existence of superconductivity in our samples as follows. (We note that additional information on experimental conditions are given in the appropriate figure legends.)
(1) There is a sharp drop in resistivity with cooling, indicating a phase transformation. The measured minimum resistance is at least as low, ∼10−11 ohm m—about two orders of magnitude less than for pure copper (Fig. 1, Extended Data Fig. 3e) measured at the same temperature19. (2) A strong isotope effect is observed: Tc shifts to lower temperatures for sulfur deuteride, indicating phonon-assisted superconductivity (Fig. 2b, c). The BCS theory gives the dependence of Tc on atomic mass m as Tc ∝ m−α, where α ≈ 0.5. Comparison of Tc values in the pressure range P > 170 GPa (Fig. 2c) gives α ≈ 0.3. (3) Tc shifts to lower temperatures with available magnetic field (B) up to 7 T (Fig. 3). Much higher fields are required to destroy the superconductivity: extrapolation of Tc(B) gives an estimate of a critical magnetic field as high as 70 T (Fig. 3). (4) Finally, in magnetic susceptibility measurements (Fig. 4) a sharp transition from the diamagnetic to the paramagnetic state (Fig. 4a) was observed for zero-field-cooled (ZFC) material. The onset temperature of the superconducting state Tonset = 203(1) K, and the width of the superconducting transition is nearly the same as in electrical measurements (Fig. 4a). Magnetization measurements M(H), where H is magnetic field, at different temperatures (Fig. 4c) revealed a pronounced hysteresis indicating type II superconductivity with the first critical field Hc1 ≈ 30 mT. The magnetization decreases sharply at temperatures above 200 K showing the onset of superconductivity at 203.5 K, in agreement with the susceptibility measurements (Fig. 4a). A list of key properties of the new superconductor is given in Methods.
We have presented purely experimental evidence of superconductivity in sulfur hydride. However the particular compound responsible for the high Tc is not obvious. The superconductivity measured in the low-temperature runs (Fig. 1) possibly relates to H2S, as it is generally consistent with calculations10 for H2S: both the value of Tc ≈ 80 K and its pressure behaviour. However superconductivity with Tc ≈ 200 K (Fig. 2) does not follow from these calculations. We suppose that it relates to the decomposition of H2S, as high temperatures are required to reach the high Tc (Fig. 2b). Precipitation of elemental sulfur on decomposition could be expected (which is well known at low pressures of P < 100 GPa; ref. 14); however the superconducting transition in elemental sulfur occurs at significantly lower temperatures (Fig. 1b). Another expected product of decomposition of H2S is hydrogen. However, the strong characteristic vibrational stretching mode from the H2 molecule was never observed in our Raman spectra (nor was it observed in ref. 14). Therefore we suppose that the dissociation of H2S is different and involves the creation of higher hydrides, such as 3H2S → H6S + 2S or 2H2S → H4S + S. It is natural to expect these reactions, as sulfur can be not only divalent, but also exhibits higher valencies. In fact, calculations10 indirectly support this hypothesis, as the dissociation H2S → H2 + S was shown to be energetically very unfavourable. We found further theoretical support in ref. 20. In that work, the van der Waals compound21 (H2S)2H2 was considered, and it was shown that at pressures above 180 GPa it forms an Im-3m structure with H3S stoichiometry. The predicted Tc ≈ 190 K and its pressure dependences are close to our experimental values (Fig. 2c). Our hypothesis of the transformation of H2S to higher hydrides (in the H3S stoichiometry each S atom is surrounded by 6 hydrogen atoms) is strongly supported by further calculations22,23. All the numerous works based on the Im-3m structure23,24,25,26,27 are consistent in their prediction of Tc >∼200 K, which decreases with pressure. The hydrogen sublattice gives the main contribution to superconductivity20,25,26. Inclusion of zero point vibrations and anharmonicity in the calculations24 corrected the calculated Tc to ∼190 K, and the isotope coefficient from α = 0.5 to α = 0.35—both in agreement with the present work.
The highest Tc of 203 K that we report here has been achieved most probably in H3S having the Im-3m structure. It is a good metal; interestingly, there is also strong covalent bonding between H and S atoms in this compound20. This is in agreement with the general assumption (see for instance ref. 28) that a metal with high Tc should have strong covalent bonding (as is realized in MgB2; ref. 29) together with high-frequency modes in the phonon spectrum. This particular combination of bonding type and phonon spectrum would probably provide a good criterion when searching for the materials with high Tc at ambient pressure that are required for applications. There are many hydrogen-containing materials with strong covalent bonding (such as organics) but typically they are insulators. In principle, they could be tuned to a metallic state by doping or gating. Modern methods of structure prediction could facilitate exploration for the desired materials.
For electrical measurements we used diamond anvil cells (DACs) with anvils of the following shape: tip diameter of 200–300 μm bevelled at 7–8° to a culet of 40–80 μm. An insulating gasket is required to separate the metallic gasket from the electrodes. It was prepared in the following way (Extended Data Fig. 3). First a metallic gasket of T301 stainless steel (or Re) 250 μm thick was indented with about 17–20 GPa pressure. Then the bottom of the imprint of diameter ∼200 μm was drilled out, and a powder insulating material was put in the imprint and pressed between the anvils to form a layer. The insulating layer was made of either Teflon, NaCl or CaSO4 as these materials do not react with H2S. The layer was pressed to obtain a thickness in the centre of ∼3–5 μm to provide stable clamping. A larger thickness leads to instability in the sample—it shifts or escapes under pressure—while with a thinner gasket it is difficult to reach high pressures. A hole of diameter ∼10–30 μm was then drilled in the insulating layer. Four Ti electrodes were sputtered on the diamond anvil. The electrodes were capped with Au to prevent oxidation of the Ti. (To check a possible contribution of the diamond surface to the conductivity, we prepared a different configuration of electrodes for a once-only experiment: two electrodes were sputtered on one anvil and another two on another anvil, similar to ref. 13). After preparation of the electrodes the gasket was put back on the anvil and the DAC was assembled so that the separation between the anvils was about 20–100 μm (measured by interference fringes). The DAC was placed into a cryostat and cooled down to ∼200 K (within the temperature range of liquid H2S) and then H2S gas was put through a capillary into a rim around the diamond anvil where it liquefied (Extended Data Fig. 4). H2S of 99.5% and D2S of 97% purity were been used. The filling was monitored visually (Extended Data Figs 4, 5) and the sample was identified by measuring Raman spectra. Then liquid H2S was clamped in the gasket hole by pushing the piston of the DAC with the aid of screws outside the cryostat. The thickness of the sample can be estimated to be few micrometres, as measured from interference spectra through the clamped transparent sample. The thickness might be ∼1 μm if the sample expanded over the culet (Fig. 4). After the clamping, the DAC was heated to ∼220 K to evaporate the rest of the H2S, and then the pressure was further increased at this temperature. The pressure remained stable during the cooling within ±5 GPa. The pressure was determined by a diamond edge scale at room temperature and low temperatures32. For optical measurements a Raman spectrometer was equipped with a nitrogen-cooled CCD and notch filters. The 632.8 nm line of a He–Ne laser was used to excite the Raman spectra and to determine pressure.
The low temperature loading seems to be required to prepare samples with high Tc. If H2S was loaded at room temperature in the gas loader, for example, only sulfur was detected in Raman and X-ray scattering. Apparently in this route the sample decomposes before reaching the required high-pressure phase of H3S. We did not explore all (P,T) paths to reach the state with high Tc. We found however that superconductivity is not observed in sample loaded at ∼200 K but heated to room temperature at low pressure <∼100 GPa.
The resistance and Raman spectra were measured during the pressurizing using the four-probe van der Pauw method (Extended Data Fig. 3) with a current of 10–10,000 μA. The temperature was reliably determined by using a slow warming rate (∼1 K min−1) and allowing the DAC to equilibrate with attached thermometer. The determined Tc was well reproduced in measurements with the PPMS6000 (Physical Property Measurement System from Quantum Design) and other set-ups. Tc was determined as the point of steepest change of resistance from the normal state (Fig. 2b).
The influence of the magnetic field on superconducting transitions has been measured with a non-magnetic DAC (diameter 25 mm) in a PPMS6000 in a 4–300 K temperature range and fields up to 7 T.
Magnetic susceptibility measurements were performed in an MPMS (Magnetic Property Measurement System) from Quantum Design. For these measurements a miniature non-magnetic cell made of Cu:Ti alloy working up to 200 GPa was designed (Fig. 4b). Samples of diameter ∼50–100 μm and a thickness of a few micrometres were prepared to provide a sufficient signal. Magnetic susceptibility measurements using a high-pressure cell were performed using a background subtraction feature of the MPMS software of the SQUID magnetometer (Extended Data Fig. 6).
We present here some important key features of our new high-Tc sulfur hydride superconductor:
(1) The new superconductor is of type II. This fact is clearly supported by (i) a difference in temperature-dependent ZFC and FC magnetization (Fig. 4a), which is due to the Meissner effect (ZFC) and magnetic flux capture when the sample is cooled down from its normal state (FC); and (ii) the magnetic hysteresis curves (Fig. 4c, d). The magnetic hysteresis curves also have all the features of typical type II superconductors with a mixed state between Hc1 and Hc2.
(2) A typical value of the coherence length ξGL in the framework of the Ginzburg–Landau theory can be estimated on the basis of the measured upper critical fields from conductivity measurements (Fig. 3c). Using the experimental estimation 60 T < Hc2 < 80 T and the relation we find limits for the coherence length: 2.3 nm > ξGL > 2.0 nm. We note that this relatively short coherence length is of the same order as, for instance, the values for superconducting YBa2Cu3O7 (1.3 nm) and Nb3Sn (3.5 nm).
(3) The London penetration depth λL can be estimated from the known relation of the lower critical field Hc1 to the upper critical field Hc2 for a type II superconductor in the limit κ >> 1 of the Ginzburg–Landau parameter . Considering the experimental value of the first critical field of 3 × 10−2 T (Fig. 4c) and the above-mentioned relation 60 T < Hc2 < 80 T, we can obtain the following estimate for the London penetration depth: λL ≈ 125 nm.
(4) According to Bean’s model, the magnetic critical current density of the superconductor can be estimated from the distance between the direct and the returning branches of the magnetic hysteresis loop at a given magnetic field (Fig. 4c). Provided grain radii are about 0.1 μm, the intra-grain critical current Jc is about 107 A cm−2.
Support provided by the European Research Council under the 2010 Advanced Grant 267777 is acknowledged. We appreciate help provided in MPI Chemie by U. Pöschl. We thank P. Alireza and G. Lonzarich for help with samples of CuTi; J. Kamarad, S. Toser and C. Q. Jin for sharing their experience on SQUID measurements; K. Shimizu and his group for cooperation; P. Chu and his group for many discussions and collaboration, and L. Pietronero, M. Calandra and T. Timusk for discussions. V.K. and S.I.S. acknowledge the DFG (Priority Program No. 1458) for support. M.I.E. thanks H. Musshof and R. Wittkowski for precision machining of the DACs.
Extended data figures
About this article
Journal of Superconductivity and Novel Magnetism (2019)