A magnetic field normally destroys superconductivity. A team from MIT and the University of Basel describes the opposite: in rhombohedrally stacked graphene, three superconducting states survive 8.5 teslas applied in the plane — dozens of times the theoretical ceiling that should have killed them — and one of them simply does not exist at zero field. The material is neither exotic nor hard to obtain: it is pure carbon, stacked in the right order.

Source: nature.com

In plain terms

A superconductor carries current without resistance because its electrons pair up two by two. A magnet usually breaks these pairs: the two electrons of a pair have opposite spins — tiny magnetized needles pointing in opposite directions — and a magnetic field wants to align them in the same direction, which dissolves the pair. There is even a calculable ceiling, called the Pauli limit, beyond which an ordinary superconductor cannot survive.

Here the researchers stack four or five sheets of graphene — carbon one atom thick — offsetting each sheet always in the same direction, which is called rhombohedral stacking. In it they find four different superconductors depending on the voltages applied. The three that react to the magnet take 8.5 teslas, that is dozens of times their theoretical ceiling: their pairs therefore do not behave like ordinary pairs in the face of the field — either their spins are not opposite, or something shields them from the field. The authors examine this second hypothesis, the one that had served until now for other materials, and rule it out. One of these states appears only from about 8 teslas onward: without a magnet, it does not exist. Another is strengthened when a minuscule field, a few thousandths of a tesla, is applied perpendicular to the plane. Finally, on a second sample — a four-sheet one this time — by laying a thin layer of another crystal on top, the team makes five additional regions appear that also behave like superconductors, without degrading the cleanliness of the material.

All this happens at extreme temperatures, a few hundredths of a degree above absolute zero: this is not a usable superconductor, but a remarkably clean test bed for understanding how electrons can pair up otherwise than in the textbooks.

Discovery

Technical explanation

1. Why rhombohedral stacking, and not just any graphene — Ordinary graphite stacks its planes in the alternating ABAB fashion known as Bernal stacking. Rhombohedral stacking, denoted ABC, offsets each plane in the same direction, so that the structure only repeats after three layers. This geometry changes the dispersion of the low-energy electrons: instead of the linear cone of monolayer graphene, the energy varies as $E \propto k^{N}$E∝kN for $N$N layers. For five layers, the band is therefore extraordinarily flat. Now, a flat band means a large density of states at a given energy: the kinetic energy, which ordinarily disperses the electrons, becomes small compared with their mutual repulsion. It is then the interactions that command, and the system spontaneously reorganizes into phases where the spin and valley degrees of freedom cease to be equivalent. The authors add a second knob: the graphite gates placed on either side impose a perpendicular electric field, called the displacement field, which continuously deforms the band. Carrier density and displacement field thus form the control plane in which the superconductors appear as so many distinct islands.

2. What fermiology establishes — the identity of the parent state — Before qualifying a superconductivity, one must know which metal it comes out of. The method used is classic and conclusive: the longitudinal resistance is measured as a function of the perpendicular field; the Landau levels sweep the Fermi surface and imprint oscillations periodic in $1/B_\perp$1/B⊥​, known as Shubnikov–de Haas oscillations. A Fourier transform converts these oscillations into frequencies, and each frequency is proportional to the area of a Fermi pocket. By normalizing to the carrier density, the authors obtain reduced frequencies whose differences are diagnostic. For the parent state of SC1, the relation $f_1 - f_2 = 1/4$f1​−f2​=1/4 marks a "full-metal" with an annular Fermi surface. For SC2 it is $f_1 - 2f_3 \approx 1/2$f1​−2f3​≈1/2; for SC3, $f_1 - f_2 \approx 1/2$f1​−f2​≈1/2 and $f_1 - f_3 \approx 1/2$f1​−f3​≈1/2: in both cases a "half-metal" — a single combination of spin and valley conducts — with a Fermi surface that is likewise annular. What this demonstrates: the isospin polarization is not postulated from the superconducting behavior, it is measured independently on the normal state. Two complementary checks, which must not be confused with one another, come to refine it: the absence of an anomalous Hall effect indicates zero valley polarization, and the absence of magnetic hysteresis suggests a vanishing orbital ferromagnetism. (The writer's reading:) the polarization brought to light is therefore indeed of spin, and not of valley or of orbit.

3. The Pauli limit, and what its violation allows one to conclude — In a conventional superconductor, Cooper pairs associate two electrons of opposite spin. An in-plane magnetic field acts on these spins through the Zeeman effect; when the energy gained by aligning the spins exceeds the condensation energy, the pair breaks. The corresponding field is the Clogston–Chandrasekhar ceiling, whose usual BCS form is written:
   $B_{\mathrm{P}} \approx 1{,}84 \, T_{\mathrm{c}}$BP​≈1,84Tc​
   with $B_{\mathrm{P}}$BP​ in teslas and $T_{\mathrm{c}}$Tc​ in kelvin — this is the textbook BCS form. The violation ratio is defined as the ratio of the measured critical field to this ceiling:
   $\mathrm{PVR} = \frac{B_{\mathrm{c},\parallel}^{\text{mesuré}}}{B_{\mathrm{P}}}$PVR=BP​Bc,∥mesureˊ​​
   The authors write that the three superconductivities boosted or induced by the field in the pentalayer are "all robust up to an in-plane field of 8.5 T, exceeding the Pauli limit by several tens of times". The ratio reaches more than 34 for SC3, whose limit they estimate at about 0.24 T, and about 68 for SC4 at the highest available field, 8.835 T. This last figure can be reconstructed exactly from the formula: $1{,}84 \times 0{,}070 = 0{,}129$1,84×0,070=0,129 T, and $8{,}835 / 0{,}129 \approx 68$8,835/0,129≈68. That of SC3, on the other hand, would imply a temperature of about 130 mK, which the article does not specify and which does not match the BKT temperatures cited above. What this excess proves, and what it does not prove: it is incompatible with singlet pairing subject to the sole Zeeman effect, and points toward triplet pairing. It should be noted that the authors examine and then explicitly rule out the usual escape route, that of the so-called Ising superconductors, in which a strong spin–orbit coupling locks the spins out of the plane: those superconductors, they write, are destroyed by a high in-plane field before being completely spin-polarized, whereas SC2 to SC4 withstand it and arise from half-metals — they are therefore spin-polarized in a different way from an Ising superconductor. The fact remains that the argument, even after this exclusion, remains an argument from exclusion and not a direct identification of the order parameter.

4. The mechanism proposed for SC3: tilting the spins to unbalance the valleys — This is the most specific passage in the article, and the most delicate. Three energies vie for precedence there, of very different orders of magnitude: Hund's coupling, which aligns the spins, of the order of 2 meV for $10^{12}$1012 carriers per square centimeter; the intrinsic spin–orbit coupling of graphene $\lambda$λ, about 50 μeV; and the Zeeman energy of the applied field, which at 1.8 mT amounts to only 0.2 μeV. Naively, a field ten thousand times weaker than the spin–orbit coupling should be able to do nothing. The authors' reasoning gets around the objection: at zero field the spins are not arbitrary, they are tilted almost entirely into the plane, with a canting angle close to 90°. In this configuration, the out-of-plane component is almost zero but very easy to rotate — it is a soft direction of the energy landscape. A minuscule perpendicular field is therefore enough to create a nonzero $\langle \sigma_z \rangle$⟨σz​⟩ component. Now, the spin–orbit coupling converts this component into an energy shift of opposite sign in the two valleys, of the order of $\pm\lambda\langle \sigma_z \rangle/2$±λ⟨σz​⟩/2. One valley gains density of states, the other loses; intra-valley pairing is strengthened in the one that has gained. Measured consequence: the BKT transition temperature of SC3 rises from 47 to 76 mK. Symmetrically, an in-plane field, even of a few milliteslas, brings the spins back toward their easy plane, cancels the imbalance and suppresses the boost — a nontrivial prediction, and one consistent with the observation. The authors support the picture with a free-energy density calculation including Hund, spin–orbit and Zeeman, which predicts a saturation of the out-of-plane component at around 1.1 mT, to be compared with the experimental optimal field of 1.4 to 1.8 mT. The agreement is good; the authors themselves indicate that full validation of this mechanism calls for further experiments.

1. The clean limit, condition for the existence of the phenomenon — A field-boosted superconductor is, as the authors recall from the abstract onward, more vulnerable to impurities than its conventional counterpart. It still has to be proved that the sample is effectively clean. The demonstration is quantitative and rests on the comparison of two lengths. The coherence length, the characteristic size of a Cooper pair, is extracted from the perpendicular critical field:
   $\xi = \sqrt{\frac{h/2e}{2\pi B_{\mathrm{c},\perp}}}$ξ=2πBc,⊥​h/2e​​
   The mean free path, the distance an electron covers before scattering, is obtained from the sheet resistance of the normal state and from the Fermi wave vector $k_{\mathrm{F}} \approx \sqrt{\pi |n|}$kF​≈π∣n∣​. The values measured on SC1 give a coherence of about 200 nm for a mean free path of 1.6 μm in the R5G, and 300 nm for 1 μm in the R4G. The ratio $\xi/\ell$ξ/ℓ therefore remains clearly below 1: the pairs are smaller than the distance between two collisions, which is the definition of the clean limit. The authors emphasize the contrast with the superconductors of twisted graphene and of the dichalcogenides, where the normal-state resistance is higher and where the ratio exceeds 1. It is this material quality, and not a measurement trick, that makes the fragile states observable.

2. Making new superconductors by simple contact — The abstract devotes its last sentence of results to a distinct strand. By laying a layer of tungsten diselenide (WSe₂) on a tetralayer device, the team increases the spin–orbit coupling by proximity effect. One must be precise about the status of the figures: the two orders of magnitude at play — about 50 μeV for bare graphene, of the order of the meV under a dichalcogenide, depending on the relative angle between the two crystals — are values that the authors take from the literature to feed their free-energy model; no spin–orbit coupling is measured on this device. The experimental result, for its part, is not a mere modulation of the existing states: five additional regions appear, while SC2 is suppressed. The authors designate them as superconducting on the strength of a body of evidence — disappearance of the longitudinal resistance, nonlinearity of $\mathrm{d}V_{xx}/\mathrm{d}I$dVxx​/dI, suppression by a perpendicular field — without applying to them the full characterization (fermiology of the parent state, critical fields, violation ratio, $\xi/\ell$ξ/ℓ ratio) that SC1 to SC4 enjoy in the pentalayer. They are labeled SC3 to SC7 according to the numbering specific to the tetralayer, which must not be confused with the SC3 and SC4 of the pentalayer mentioned above. The crucial point is one of method: the sheet resistance of the normal state remains about 25 Ω per square at the hole density where SC1 emerges, whether measured on the bare tetralayer, on the region far from the WSe₂ or on the region in contact with it. Proximity therefore adds spin–orbit coupling without adding disorder — which, in a system that has just been shown to owe its richness solely to its cleanliness, is the condition for the engineering to be usable. The authors see in this an engineering route, for which they point to one precise application: combining superconductivity and quantum anomalous Hall states within a single stack, which would avoid having to make them coexist starting from different materials.

3. What reproduction across several devices adds — and what it does not add — Precision matters in a field where the correlated states of graphene have sometimes resisted reproduction: the main results are recovered on two R4G devices and five R5G devices, in two independent cryostats belonging to two laboratories — a Bluefors LD250 at MIT, a Leiden MNK126-700 in Basel. This rules out the artifact of a particular cryogenic setup, of a given electrical filtering or of a single favored sample. On the other hand, the two laboratories co-sign the article and share the fabrication method; this is therefore not yet a replication by an outside team, the only kind able to test independence from the stacking protocol itself.

Why It Worked

The result rests on the conjunction of three things that the article makes explicit. First the platform: rhombohedral stacking produces a gate-tunable flat band, hence a system where correlation dominates and where the material can be walked continuously from one phase to another without changing it. Then cleanliness: with $\xi/\ell \ll 1$ξ/ℓ≪1, against a ratio greater than 1 in twisted graphene and the dichalcogenides, the system remains in a regime where fragile superconductivities survive instead of being drowned by disorder. Finally experimental finesse: detecting a boost at less than 2 mT presupposes field control and filtering to match, and working at a base temperature of 7 mK gives the margin needed to resolve transitions at a few tens of millikelvin.

The relevant comparison is not with just any superconductor, but with the immediate state of the art, which the authors themselves cite: Bernal bilayer graphene already showed a field boost, but only in the plane. What the pentalayer adds is twofold — an out-of-plane field boost, and the obtaining of these states at much lower gate fields. Quantitatively, a conventional singlet superconductor with a critical temperature of 70 mK should disappear at around 0.13 T; SC4 is still there at 8.835 T, and does not even appear until above about 8 T.

The gap between the announcement and the proof deserves to be stated plainly. What the study demonstrates: the existence, in one and the same material and across several devices, of distinct superconductivities of which three respond to the magnetic field in a way incompatible with singlet pairing subject to the Zeeman effect alone, all of it in a clean limit characterized quantitatively; and, without degrading that cleanliness, the appearance by proximity of five additional regions that transport designates as superconducting. What it does not demonstrate: the symmetry of the order parameter. No measurement sensitive to the phase of the order parameter — the Fraunhofer patterns mentioned below probe the phase coherence of transport, not the symmetry of the pairing —, no gap spectroscopy comes here to establish it directly; the triplet remains an inference drawn from the exclusion of the other scenarios, including the Ising one that the authors explicitly rule out. The authors are moreover measured about their own explanation of the boost of SC3, which they present as calling for additional experimental validation.

Two reservations deserve to be reported just as the authors state them. SC4 is fragile, which complicates its detailed analysis and leads them to retain a conservative estimate of 70 mK for its transition temperature — the figure is a methodological floor, not a fine measurement. And some states retain a nonzero residual resistance relative to the normal-state resistance, that is to say the drop is not complete — the authors attribute this to the presence of non-superconducting islands in a micrometer-sized device, a consequence of the sample's inhomogeneity, including when density and displacement field are tuned to a superconducting phase. Conversely, a point that is often poorly reported must be set right: the Fraunhofer interference patterns, the signature of a coherent superconducting transport, are indeed observed in most of the states, including SC1 to SC4 in the pentalayer. This is a positive and relatively rare result for crystalline rhombohedral devices, in contrast with moiré devices; where they are missing, the authors attribute it to an out-of-plane critical field that is too low, which destroys the superconductivity before the interference pattern has time to form.

A final word on the temperature scale, to avoid any misunderstanding: at a few tens of millikelvin, one is five to six orders of magnitude below the superconductors used in medical imaging or in accelerator magnets. The scope of the work is conceptual — a clean and tunable test bed for unconventional superconductivity — and not applied.

Causal Chain

Isolation of graphene by exfoliation and discovery of its Dirac carriers (2004) → the finding that a flat band exalts electronic correlations → superconductivity discovered in twisted bilayer graphene at the magic angle (2018), but in a disorder regime where $\xi/\ell > 1$ξ/ℓ>1 → search for a route without twisting: ABC rhombohedral stacking, whose dispersion $E \propto k^{N}$E∝kN flattens the band without a moiré → observation, in the Bernal bilayer, of a field-boosted superconductivity, but only in the plane and at high gate field → mastery of encapsulation in boron nitride and of graphite gates, which yields high mobilities and a tunable displacement field → attainment of a clean limit, $\xi/\ell \ll 1$ξ/ℓ≪1, where fragile superconductivities become observable → mapping of the density–displacement field plane revealing four superconducting islands in the bare pentalayer, whose parent states are identified independently by Shubnikov–de Haas oscillations → measurement of their response to the field: the three sensitive states survive 8.5 T in the plane, exceeding the Clogston–Chandrasekhar limit by a factor of up to 68, and SC4 exists only at high field → observation of a boost of SC3 under one to two perpendicular milliteslas, attributed to a tilting of the spins converted by the spin–orbit coupling into an imbalance between valleys → addition of a WSe₂ layer that strengthens the spin–orbit coupling and makes five additional regions appear, designated as superconducting, without adding disorder → reproduction on seven devices and two cryostats from two laboratories → today: a tunable platform for testing unconventional pairings, whose order-parameter symmetry remains to be determined.

Anecdote

The ceiling that these states cross bears two names because it was found twice, a fortnight apart. On 1 September 1962, in Applied Physics Letters, B. S. Chandrasekhar published an upper bound on the critical field of a superconductor; on 15 September, in Physical Review Letters, Albert Clogston reached the same result by a neighboring route, comparing the condensation energy with the one the metal would gain by polarizing its spins. The community settled the matter by joining the two names. Sixty-four years later, a stack of five sheets of carbon exceeds this bound by a factor of about 68 — not because the 1962 calculation was wrong, but because its starting assumption, pairs of antiparallel spins sensitive to the Zeeman effect alone, does not apply here.

Legacy and Current Data

What can be asserted on the basis of the article comes down to laboratory figures, and one must resist the temptation to extrapolate. Four superconducting states characterized in the bare pentalayer and five additional regions designated as superconducting by proximity, transition temperatures of a few tens of millikelvin, a robustness of the three sensitive states up to 8.5 T in the plane, Pauli-limit violation ratios greater than 34 and close to 68, a coherence length of 200 to 300 nm for a mean free path of 1 to 1.6 μm, seven devices and two cryostats. No application is claimed by the authors, and none would be reasonable at these temperatures.

The interest lies elsewhere, and it is methodological: having a system in which one moves from one superconductor to another by turning two knobs — carrier density and displacement field — on a material composed of a single chemical element, with no twist angle to control and no stoichiometry to adjust; and being able to make others appear by simply laying a crystal on top. This is what the authors put forward when they speak of a clean and tunable platform for unconventional superconductivity and topological quantum states.

What remains to be established is precise: the order-parameter symmetry of each of the states, the experimental validation of the spin-tilting mechanism advanced for SC3, the full characterization of the five states obtained by proximity, and a replication by a team that did not take part in this work.

The researcher's eye — open questions

The experiments below are extensions formulated by the writer; they do not figure among the results of the study.

• Settle the symmetry of the order parameter. The exceeding of the Pauli limit excludes the singlet subject to the Zeeman effect alone, and the authors moreover rule out the Ising scenario; the fact remains that nothing directly proves the triplet. A phase-sensitive measurement, by SQUID-type interferometry on a closed junction, or a tunneling spectroscopy resolving the structure of the gap, would say whether nodes exist and what their symmetry is. This is the missing decisive experiment.
• Measure the spin–orbit coupling instead of assuming it. The model rests on two orders of magnitude borrowed from the literature, about 50 μeV for bare graphene and the meV under a dichalcogenide; neither is measured on the devices of the study. Determining them in situ, then sweeping the intermediate values by varying the nature of the dichalcogenide, its angle or the thickness of a spacer, would test the mechanism proposed for SC3, which predicts a regular shift of the optimal perpendicular field with $\lambda$λ. A monotonic dependence would support it; its absence would refute it.
• Look for time-reversal symmetry breaking. The abstract places these states within the framework of superconductors in which time-reversal symmetry is broken independently of gauge symmetry. A local magnetometer, or the search for a polar Kerr effect, would directly test this breaking instead of inferring it from the response to the field.
• Test the inhomogeneity explanation directly. The authors attribute the residual resistance to non-superconducting islands scattered through the device. A local mapping of the critical current would verify this reading instead of inferring it from a transport average, and would say whether the islands are indeed randomly distributed or correlated with a fabrication defect.
• Have it replicated outside the consortium. The seven devices share a fabrication method and two co-signing laboratories. A reproduction by an independent team, with its own stacking protocol, is what is missing for the result to stop depending on local know-how.

Sources

Primary source consulted in its full text in open access; metadata and DOIs of the background references verified during the fact-checking audit, on 7 August 2026.

• Seo J., Cotten A. A., Ye S. et al., "Family of magnetic field-boosted superconductors in rhombohedral graphene", Nature, vol. 656, pp. 60–66 — published online 29 June 2026, print issue of 6 August 2026 (no. 8126). Peer-reviewed article, open access. DOI: 10.1038/s41586-026-10815-x

Background references

These references support the reminders of general mechanism and of historical context; they do not come from the study under review.

• Chandrasekhar B. S., "A note on the maximum critical field of high-field superconductors", Applied Physics Letters, 1962, 1(1), 7–8 — first statement of the paramagnetic bound. DOI: 10.1063/1.1777362
• Clogston A. M., "Upper limit for the critical field in hard superconductors", Physical Review Letters, 1962, 9(6), 266–267 — independent derivation of the same bound, two weeks later. DOI: 10.1103/PhysRevLett.9.266
• Cao Y., Fatemi V., Fang S., Watanabe K., Taniguchi T., Kaxiras E., Jarillo-Herrero P., "Unconventional superconductivity in magic-angle graphene superlattices", Nature, 2018, 556, 43–50 — precedent of superconductivity in twisted graphene. DOI: 10.1038/nature26160
• Novoselov K. S. et al., "Electric field effect in atomically thin carbon films", Science, 2004, 306(5696), 666–669 — isolation of graphene, starting point of the causal chain. DOI: 10.1126/science.1102896

---

Transparency: the order-parameter symmetry of these superconductors is not determined by this study; the triplet character is an inference drawn from the exceeding of the Pauli limit and from the exclusion, by the authors, of the Ising scenario — not a direct measurement. The five states obtained by proximity with WSe₂ are designated as superconducting on the strength of a body of transport evidence, without the detailed characterization applied to the four states of the pentalayer. The spin-tilting mechanism proposed for SC3 is presented by the authors themselves as calling for additional experimental validation. The transition temperatures, of a few tens of millikelvin, rule out any prospect of application.