This paper was converted on www.awesomepapers.org from LaTeX by an anonymous user.
Want to know more? Visit the Converter page.

Evidence for spin-polarized bound states in
semiconductor-superconductor-ferromagnetic insulator islands

S. Vaitiekėnas Center for Quantum Devices, Niels Bohr Institute, University of Copenhagen, 2100 Copenhagen, Denmark    R. Seoane Souto Center for Quantum Devices, Niels Bohr Institute, University of Copenhagen, 2100 Copenhagen, Denmark Division of Solid State Physics and NanoLund, Lund University, 22100 Lund, Sweden    Y. Liu Center for Quantum Devices, Niels Bohr Institute, University of Copenhagen, 2100 Copenhagen, Denmark    P. Krogstrup Center for Quantum Devices, Niels Bohr Institute, University of Copenhagen, 2100 Copenhagen, Denmark    K. Flensberg Center for Quantum Devices, Niels Bohr Institute, University of Copenhagen, 2100 Copenhagen, Denmark    M. Leijnse Center for Quantum Devices, Niels Bohr Institute, University of Copenhagen, 2100 Copenhagen, Denmark Division of Solid State Physics and NanoLund, Lund University, 22100 Lund, Sweden    C. M. Marcus Center for Quantum Devices, Niels Bohr Institute, University of Copenhagen, 2100 Copenhagen, Denmark
(September 1, 2025)
Abstract

We report Coulomb blockade transport studies of semiconducting InAs nanowires grown with epitaxial superconducting Al and ferromagnetic insulator EuS on overlapping facets. Comparing experiment to a theoretical model, we associate cotunneling features in even-odd bias spectra with spin-polarized Andreev levels. Results are consistent with zero-field spin splitting exceeding the induced superconducting gap. Energies of subgap states are tunable on either side of zero via electrostatic gates.

In hybrid quantum devices with both ferromagnetic and superconducting components, competition to align electron spins or pair them into singlets can result in complex ground states and corresponding electrical properties Meservey and Tedrow (1994); Buzdin (2005); Sau et al. (2010); Eschrig (2011); Li et al. (2014); Linder and Robinson (2015); Strambini et al. (2017); Bergeret et al. (2018); Manna et al. (2020). Recently, coexistence of proximity-induced superconductivity and ferromagnetism have been demonstrated in hybrid semiconducting nanowires Vaitiekėnas et al. (2021). Coulomb-blockade spectroscopy of superconducting quantum dots provides a window into subgap spectra  Higginbotham et al. (2015) and their spin structure Prada et al. (2020).

Multiple Andreev scatterings at superconducting boundaries of a small normal conductor give rise to Andreev bound states (ABSs) Prada et al. (2020). The states can carry supercurrent through the normal region and appear in tunneling spectroscopy as discrete levels below the superconducting gap Pillet et al. (2010); Nichele et al. (2020). Coulomb effects modify transport via ABSs Grove-Rasmussen et al. (2009); Chang et al. (2013), for instance resulting in supercurrent reversal van Dam et al. (2006); Jørgensen et al. (2007). When magnetic fields Lee et al. (2014); Shen et al. (2018) or magnetic materials Heinrich et al. (2018) are involved, spin-degenerate ABSs split and becomes spin selective, as seen in tunneling spectroscopy Whiticar et al. (2021) and circuit quantum electrodynamics measurements Hays et al. (2020). The spin-active interface between a superconductor and, for example, a ferromagnetic insulator Tokuyasu et al. (1988) can also lead to spin-split ABSs Hübler et al. (2012) or, in some cases, triplet superconductivity Diesch et al. (2018).

Refer to caption
Fig. 1: (a) Schematic of hybrid InAs nanowire showing EuS and Al layers on overlapping facets of a hexagonal InAs nanowire. (b) Scanning electron micrograph of wire 1. Enhanced false coloration highlights the 400 nm island measurement setup. (c) Differential conductance, GG, as a function of voltage bias, VV, and upper-gate voltage, VUV_{\rm U}, for the 400 nm island on wire 1 at zero applied magnetic field. Steps in conductance indicated by arrows alternate between V=30V=30 and 90μ90~\muV.

Recently, a new class of triple-hybrid materials was realized based on semiconducting InAs nanowires with strong spin-orbit coupling and large gg factor, coated with epitaxial superconducting Al, and ferromagnetic insulator EuS shells Krogstrup et al. (2015); Liu et al. (2020a). We investigate nanowires with hexagonal cross-sections and partly overlapping two-facet shells, as shown schematically in Fig. 1(a). Tunneling spectroscopy into the ends of long grounded hybrid wires Vaitiekėnas et al. (2021) showed signatures consistent with topological superconductivity, as recently investigated theoretically Woods and Stanescu (2020); Maiani et al. (2021); Escribano et al. (2020); Liu et al. (2020b); Langbehn et al. (2021); Khindanov et al. (2021).

Here, we report transport through Coulomb islands, 400 and 800 nm in length, made from the same batch of wires with normal metal leads and several top- and side-gate electrodes that independently control tunnel-barrier conductances and charge occupancy [Fig. 1(b)]. We observe characteristic features in Coulomb blockade that indicate gate-dependent, discrete subgap states whose energy can be tuned to zero. Qualitative comparison of cotunneling spectra to theoretical models suggest that the subgap states are spin polarized at zero magnetic field, as discussed in detail below.

Spectroscopy of four Coulomb island devices fabricated on two wires (denoted wire 1 and wire 2) showed similar results. Measurements were carried out using standard low-noise lock-in techniques in a dilution refrigerator with a base temperature of 20 mK, equipped with a three-axis vector magnet (see Supplemental Material Sup ).

Differential conductance, G=dI/dVG=dI/dV, of the 400 nm island on wire 1 as a function of source-drain voltage bias, VV, and upper-gate voltage, VUV_{\rm U}, showed Coulomb diamonds of alternating height [Fig. 1(c)]. Once the tunneling barriers were coarsely tuned, this behavior is typical of all measured devices. Within a Coulomb valley, low-bias GG was suppressed below the experimental noise floor. At higher bias, GG showed a step-like increase at an alternating bias, as seen in Fig. 1(c). The value of VV at which this first-step feature occurs could be tuned using the lower-gate voltage, VLV_{\rm L}. A less pronounced second step in GG at higher bias [around V=120μV=120~\muV in Fig. 1(c)] did not alternate from valley to valley nor varied with VLV_{\rm L} (see Fig. S1 in Supplemental Material Sup ). The charging energy, EC=300μE_{\rm C}=300~\mueV, measured from the Coulomb diamonds, is larger than the superconducting gap of the parent Al shell, ΔAl=230μ\Delta_{\rm Al}=230~\mueV hence also larger than the induced gap, Δ\Delta, which is reduced by the coupling to EuS Vaitiekėnas et al. (2021). The 800 nm island on wire 1 showed similar even-odd periodic Coulomb blockade with step-like cotunneling features at finite bias (see Fig. S2 in Supplemental Material Sup ).

To understand the conductance features and relate them to ABSs and spin, we model transport through a superconducting Coulomb island, including a single subgap state, spin-split by Zeeman energy, EZE_{\rm Z}. Sequential single-electron tunneling through an ABS on the island yields characteristic Coulomb diamonds Higginbotham et al. (2015); van Heck et al. (2016). To account for intermediate strengths of tunnel couplings to both leads we also include cotunneling processes Ekström et al. (2020) through a next-to-leading order expansion in the T-matrix (see Supplemental Material Sup ). Elastic cotunneling gives a bias independent background conductance, while inelastic cotunneling leaves the system in an excited state yielding steps in GG when the bias matches excitation energies.

Refer to caption
Fig. 2: (a) Calculated conductance, gg, as a function of voltage bias, vv, and charge offset, nGn_{\rm G}, for a superconducting island with a superconducting gap Δ\Delta, charging energy EC=5ΔE_{\rm C}=5\Delta, Zeeman energy EZΔE_{\rm Z}\gg\Delta (shown here for EZ=100ΔE_{\rm Z}=100\Delta), and a spin-polarized subgap state at energy ε=Δ/4\varepsilon=\Delta/4 (see Supplemental Material Sup ). Steps in gg due to cotunneling are visible at v=Δεv=\Delta-\varepsilon in odd (o) and v=Δ+εv=\Delta+\varepsilon in even (e) valleys as well as at v=2Δv=2\Delta in both valleys. The calculations were done for finite temperature kBT=0.03Δk_{\rm B}\,T=0.03\,\Delta. (b) and (c) Schematic superconducting density of states for odd (b) and even (c) valleys in (a) with excitations indicated by arrows. (d) In the odd ground state, a quasiparticle in the bound state exits the island, while another quasiparticle from a lead tunnels into the continuum. The excitation energy of Δε\Delta-\varepsilon is supplied by the voltage bias. (e) In both valleys, a voltage bias of 2Δ2\Delta can break a Cooper pair. In that case, one quasiparticle leaves the island, while the other, together with a quasiparticle from a lead, is excited to the continuum. (f) In the even ground state, a spin-down quasiparticle from a broken Cooper pair is excited into the empty bound state and the spin-up quasiparticle leaves the island, while another quasiparticle tunnels into the continuum. The total energy cost of the process is Δ+ε\Delta+\varepsilon.

Theoretical values for differential conductance, gg, of a Coulomb island containing a single spin-split ABS as a function of voltage bias, vv, and gate-induced charge offset, nGn_{\rm G}, is shown in Fig. 2(a), where ε>Δ\varepsilon_{\uparrow}>\Delta and εε\varepsilon_{\downarrow}\equiv\varepsilon are energies of the two spin branches. The main experimental features are captured qualitatively in this simple theoretical model. In particular, the bias value of the conductance step alternates between even (e) and odd (o) island parities. Steps in differential conductance, marked by the red and blue ticks, correspond to transitions between ground and lowest excited states, as illustrated in Figs. 2(b) and 2(c). Red lines marking Δε\Delta-\varepsilon for odd valleys and Δ+ε\Delta+\varepsilon for even valleys correspond to processes that change the parity of the subgap state, while the blue lines marking 2Δ2\Delta for both valleys correspond to processes that break Cooper pairs without changing parity. Cotunneling processes involving higher-energy intermediate states with ±1\pm 1 charge on the island are shown in Figs. 2(d)–2(f).

Theoretical spectra for spin-degenerate or weakly spin-split ABS show a denser pattern of cotunneling steps associated with excitations to spin-flipped states at fixed charge, as shown in Fig. S3 in Supplemental Material Sup . A qualitative comparison shows that the experimental data from Fig. 1(c) agree better with the spin-polarized rather than the spin-degenerate theory (see the discussion and Figs. S4 and S6 in Supplemental Material Sup ). Previous measurements of a similar hybrid island, but without the EuS shell, had shown Coulomb spectra that are consistent with the theory for a spin-degenerate bound state, see Fig. S7 in Supplemental Material Sup .

Returning to experiment, transport data for the 400 nm device on wire 2 shown in Fig. 1(c) yield Δ+ε=120μ\Delta+\varepsilon=120~\mueV and Δε=60μ\Delta-\varepsilon=60~\mueV, giving ε=30μ\varepsilon=30~\mueV and Δ=90μ\Delta=90~\mueV, consitent with the 2Δ2\Delta feature at 180μ180~\mueV. The data in Fig. 1(c) gives a slightly smaller Δ=60μ\Delta=60~\mueV. We note that Δ\Delta can be gate-voltage dependent. In general, the induced gap is considerably smaller than the parent Al gap in these wires. The deduced EC=430μE_{\rm C}=430~\mueV is larger than Δ\Delta, consistent with the even-odd periodic Coulomb pattern. The sharp spectral features at the degeneracy points indicate a discrete subgap state.

Decreasing VLV_{\rm L} from +0.2 V to 0 modifies the Coulomb blockade peaks from distinctly even-odd to 1e1e-periodic at zero bias, with consecutive diamonds differing only by the intensity of step features at finite bias, as seen in Fig. 3(b). The onsets of the lower-energy steps in both valleys align with Δ=90μ\Delta=90~\mueV, indicating that ε0\varepsilon\approx 0.

We interpret the evolution as reflecting a subgap state that gradually decreases to zero energy as VLV_{\rm L} is varied. In other words, the gate-induced electric fields change the electrostatic environment of the hybrid nanowire, thus modifying the parameters of the subgap state and effectively changing its energy. In the present context, the evolving spin-mixing angle at the superconductor-ferromagnetic insulator interface Hübler et al. (2012) contributes to the gate dependence of ε\varepsilon.

Similar measurements at various VLV_{\rm L} are shown in Fig. S8 in Supplemental Material Sup . The even-odd structure in the amplitude of the finite bias conductance steps is expected theoretically, and reflects the relative phase difference between electron and hole components of the subgap state (see the discussion and Fig. S9 in Supplemental Material Sup ).

Refer to caption
Fig. 3: (a) Differential conductance, GG, as a function of source-drain bias, VV, and upper gate voltage, VUV_{\rm U}, for the 400 nm island on wire 2. A clear even-odd Coulomb diamond pattern is visible with an inelastic onset in VV at Δ+ε=120μ\Delta+\varepsilon=120~\mueV for the bigger diamond and Δε=60μ\Delta-\varepsilon=60~\mueV for the smaller one, as well as an additional step at 2Δ=180μ2\Delta=180~\mueV in both diamonds, giving Δ=90μ\Delta=90~\mueV and ε=30μ\varepsilon=30~\mueV. The data were taken at a fixed lower-gate voltage VL=0.2V_{\rm L}=0.2 V. (b) Similar to (a) measured at VL=0V_{\rm L}=0, giving nearly 1ee-periodic Coulomb diamonds with two steps in GG for all diamonds at Δ=90μ\Delta=90~\mueV and 2Δ=180μ2\Delta=180~\mueV, indicating ε0\varepsilon\approx 0. The measured charging energy EC=430μE_{\rm C}=430~\mueV.

To investigate ε\varepsilon dependence on the electrostatic environment we look over a wider range of gate voltages. The observed even-odd pattern crosses smoothly through 1ee periodicity, reflecting the continuous evolution of ε\varepsilon across zero. This is shown in Fig. 4(a) as a function of upper-gate voltage, VUV_{\rm U}, for the 800 nm island on wire 2. Both the onsets of the high-bias features, at values cic_{\rm i}, and the peak spacings, sis_{\rm i}, alternate in magnitude. Subscripts i=1i=1 and 2 denote the two different charge occupancies of the island. We define c1=(Δ+ε)/ec_{1}=(\Delta+\varepsilon)/e and c2=(Δε)/ec_{2}=(\Delta-\varepsilon)/e, then take the difference between consecutive cic_{\rm i} to extract the subgap-state energy ε\varepsilon, as a function of VUV_{\rm U} as shown in Fig. 4(b). Within the measured range of VUV_{\rm U}, ε\varepsilon decreases monotonically from +10μ+10~\mueV to 10μ-10~\mueV. Independently, values for ε\varepsilon were extracted from Coulomb peak spacing at zero bias Higginbotham et al. (2015); Albrecht et al. (2016); Shen et al. (2018). For EC>Δ>εE_{\rm C}>\Delta>\varepsilon, peak spacings are given by s1=(EC+ε)/eηs_{1}=(E_{\rm C}+\varepsilon)/e\eta and s2=(ECε)/eηs_{2}=(E_{\rm C}-\varepsilon)/e\eta, where η\eta is a dimensionless lever arm measured from the slopes of the Coulomb diamonds. The subgap-state energy inferred from the Coulomb peak spacing difference agrees well with ε\varepsilon deduced from finite bias steps, as shown in Fig. 4(b). Good qualitative agreement between measured and computed spectra is shown in Fig. S10 in Supplemental Material Sup . Similar analysis for ε\varepsilon as a function of VLV_{\rm L}, where the subgap state approaches but does not cross zero energy, is shown in Fig. S11 in Supplemental Material Sup .

The sign of ε\varepsilon depends on the definition of cic_{i} and sis_{i}. Assuming strong zero-field spin splitting at zero applied magnetic field leaves it ambiguous whether or not a level has crossed zero energy. We therefore cannot label the even and odd valleys with certainty. We note that while in principle the evolution of ε\varepsilon with applied magnetic field contains information on the spin projection of the bound state and hence the ground state parity, we are not able to determine if the field predominantly affects the Zeeman splitting or the magnetization of the EuS (see Ref. Vaitiekėnas et al. (2021)). Representative magnetic-field data for both islands on wire 2 are shown in Figs. S11 and S12 in Supplemental Material Sup .

Refer to caption
Fig. 4: (a) Differential conductance, GG, measured for the 800 nm island on wire 2 as a function of source-drain bias, VV, over an extended range of the upper-gate voltage, VUV_{\rm U}. The Coulomb blockade pattern evolves from even-odd at VU=1.47V_{\rm U}=-1.47 V, through 1ee around VU=1.43V_{\rm U}=-1.43 V, to even-odd periodicity again at VU=1.35V_{\rm U}=-1.35 V, visible in both inelastic cotunneling onsets cic_{\rm i} and peak spacings sis_{\rm i}, where i=1i=1 and 2 denote Coulomb valleys. The larger i=1i=1 diamonds on the negative side of the measured gate voltage range become smaller than the i=2i=2 diamonds on the more positive side. The measured charging energy EC=320μE_{\rm C}=320~\mueV is smaller than for the 400 nm island. The data were taken at VL=0.7V_{\rm L}=-0.7 V. (b) Subgap state energy, ε\varepsilon, inferred from step heights (black) and peak-spacing differences (red). ε\varepsilon decreases monotonically from roughly 10μ10~\mueV through 0 to 10μ-10~\mueV as the gate voltage is increased. Black error bars represent standard errors from the cic_{\rm i} measurement at the positive and negative VV; red error bars were estimated by propagation of error from the Lorentzian peak fitting and lever arm, η\eta, measurement.

Finally, we note that for specific gate configurations there are no inelastic cotunneling steps present in Coulomb diamonds (see Fig. S14 in Supplemental Material Sup ). This can be understood within the model as resulting from the condition ε>Δ\varepsilon>\Delta, yielding a cotunneling background for all voltage-bias values within odd valleys and non-zero conductance above 2Δ2\Delta in even valleys.

We thank Claus Sørensen for contributions to materials growth and Shivendra Upadhyay for assistance with nanofabrication. We acknowledge support from the Danish National Research Foundation, European Research Council (Grants Agreement No. 716655 and No. 856526), Microsoft, NanoLund, QuantERA (Project ”2D hybrid materials as a platform for topological quantum computing”), and a research grant (Project 43951) from VILLUM FONDEN.

References

  • Meservey and Tedrow (1994) R. Meservey and P. Tedrow, Phys. Rep. 238, 173 (1994).
  • Buzdin (2005) A. I. Buzdin, Rev. Mod. Phys. 77, 935 (2005).
  • Sau et al. (2010) J. D. Sau, R. M. Lutchyn, S. Tewari,  and S. Das Sarma, Phys. Rev. Lett. 104, 040502 (2010).
  • Eschrig (2011) M. Eschrig, Phys. Today 64, 43 (2011).
  • Li et al. (2014) J. Li, H. Chen, I. K. Drozdov, A. Yazdani, B. A. Bernevig,  and A. MacDonald, Phys. Rev. B 90, 235433 (2014).
  • Linder and Robinson (2015) J. Linder and J. W. Robinson, Nat. Phys. 11, 307 (2015).
  • Strambini et al. (2017) E. Strambini, V. Golovach, G. De Simoni, J. Moodera, F. Bergeret,  and F. Giazotto, Phys. Rev. Mater. 1, 054402 (2017).
  • Bergeret et al. (2018) F. S. Bergeret, M. Silaev, P. Virtanen,  and T. T. Heikkilä, Rev. Mod. Phys. 90, 041001 (2018).
  • Manna et al. (2020) S. Manna, P. Wei, Y. Xie, K. T. Law, P. A. Lee,  and J. S. Moodera, PNAS 117, 8775 (2020).
  • Vaitiekėnas et al. (2021) S. Vaitiekėnas, Y. Liu, P. Krogstrup,  and C. Marcus, Nat. Phys. 17, 43 (2021).
  • Higginbotham et al. (2015) A. P. Higginbotham, S. M. Albrecht, G. Kiršanskas, W. Chang, F. Kuemmeth, P. Krogstrup, T. S. Jespersen, J. Nygård, K. Flensberg,  and C. M. Marcus, Nat. Phys. 11, 1017 (2015).
  • Prada et al. (2020) E. Prada, P. San-Jose, M. W. de Moor, A. Geresdi, E. J. Lee, J. Klinovaja, D. Loss, J. Nygård, R. Aguado,  and L. P. Kouwenhoven, Nat. Rev. Phys. 2, 575 (2020).
  • Pillet et al. (2010) J. Pillet, C. Quay, P. Morfin, C. Bena, A. L. Yeyati,  and P. Joyez, Nat. Phys. 6, 965 (2010).
  • Nichele et al. (2020) F. Nichele, E. Portolés, A. Fornieri, A. Whiticar, A. Drachmann, S. Gronin, T. Wang, G. Gardner, C. Thomas, A. Hatke, M. Manfra,  and C. Marcus, Phys. Rev. Lett. 124, 226801 (2020).
  • Grove-Rasmussen et al. (2009) K. Grove-Rasmussen, H. I. Jørgensen, B. M. Andersen, J. Paaske, T. S. Jespersen, J. Nygård, K. Flensberg,  and P. E. Lindelof, Phys. Rev. B 79, 134518 (2009).
  • Chang et al. (2013) W. Chang, V. E. Manucharyan, T. S. Jespersen, J. Nygård,  and C. M. Marcus, Phys. Rev. Lett. 110, 217005 (2013).
  • van Dam et al. (2006) J. A. van Dam, Y. V. Nazarov, E. P. Bakkers, S. De Franceschi,  and L. P. Kouwenhoven, Nature 442, 667 (2006).
  • Jørgensen et al. (2007) H. I. Jørgensen, T. Novotnỳ, K. Grove-Rasmussen, K. Flensberg,  and P. Lindelof, Nano Lett. 7, 2441 (2007).
  • Lee et al. (2014) E. J. Lee, X. Jiang, M. Houzet, R. Aguado, C. M. Lieber,  and S. De Franceschi, Nat. Nanotechnol. 9, 79 (2014).
  • Shen et al. (2018) J. Shen, S. Heedt, F. Borsoi, B. van Heck, S. Gazibegovic, R. L. O. het Veld, D. Car, J. A. Logan, M. Pendharkar, S. J. Ramakers, et al., Nat. Comm. 9, 1 (2018).
  • Heinrich et al. (2018) B. W. Heinrich, J. I. Pascual,  and K. J. Franke, Prog. Surf. Sci. 93, 1 (2018).
  • Whiticar et al. (2021) A. Whiticar, A. Fornieri, A. Banerjee, A. Drachmann, S. Gronin, G. Gardner, T. Lindemann, M. Manfra,  and C. Marcus, arXiv:2101.09706  (2021).
  • Hays et al. (2020) M. Hays, V. Fatemi, K. Serniak, D. Bouman, S. Diamond, G. de Lange, P. Krogstrup, J. Nygård, A. Geresdi,  and M. Devoret, Nat. Phys. 16, 1103 (2020).
  • Tokuyasu et al. (1988) T. Tokuyasu, J. A. Sauls,  and D. Rainer, Phys. Rev. B 38, 8823 (1988).
  • Hübler et al. (2012) F. Hübler, M. Wolf, T. Scherer, D. Wang, D. Beckmann,  and H. v. Löhneysen, Phys. Rev. Lett. 109, 087004 (2012).
  • Diesch et al. (2018) S. Diesch, P. Machon, M. Wolz, C. Sürgers, D. Beckmann, W. Belzig,  and E. Scheer, Nat. Commun. 9, 5248 (2018).
  • Krogstrup et al. (2015) P. Krogstrup, N. Ziino, W. Chang, S. Albrecht, M. Madsen, E. Johnson, J. Nygård, C. Marcus,  and T. Jespersen, Nat. Mater. 14, 400 (2015).
  • Liu et al. (2020a) Y. Liu, S. Vaitiekėnas, S. Marti-Sanchez, C. Koch, S. Hart, Z. Cui, T. Kanne, S. A. Khan, R. Tanta, S. Upadhyay, M. Espineira Cachaza, C. M. Marcus, J. Arbiol, K. A. Moler,  and P. Krogstrup, Nano Lett. 20, 456 (2020a).
  • Woods and Stanescu (2020) B. D. Woods and T. D. Stanescu, arXiv:2011.01933  (2020).
  • Maiani et al. (2021) A. Maiani, R. S. Souto, M. Leijnse,  and K. Flensberg, Phys. Rev. B 103, 104508 (2021).
  • Escribano et al. (2020) S. Escribano, E. Prada, Y. Oreg,  and A. Levy Yeyati, arXiv:2011.06566  (2020).
  • Liu et al. (2020b) C.-X. Liu, S. Schuwalow, Y. Liu, K. Vilkelis, A. Manesco, P. Krogstrup,  and M. Wimmer, arXiv:2011.06567  (2020b).
  • Langbehn et al. (2021) J. Langbehn, S. A. González, P. W. Brouwer,  and F. von Oppen, Phys. Rev. B 103, 165301 (2021).
  • Khindanov et al. (2021) A. Khindanov, J. Alicea, P. Lee, W. S. Cole,  and A. E. Antipov, Phys. Rev. B 103, 134506 (2021).
  • (35) See Supplemental Material for more detailed methods and simulation description, as well as additional measurements.
  • van Heck et al. (2016) B. van Heck, R. M. Lutchyn,  and L. I. Glazman, Phys. Rev. B 93, 235431 (2016).
  • Ekström et al. (2020) J. Ekström, P. Recher,  and T. L. Schmidt, Phys. Rev. B 101, 195420 (2020).
  • Albrecht et al. (2016) S. M. Albrecht, A. P. Higginbotham, M. Madsen, F. Kuemmeth, T. S. Jespersen, J. Nygård, P. Krogstrup,  and C. M. Marcus, Nature 531, 206 (2016).
  • Liu et al. (2020c) Y. Liu, A. Luchini, S. Martí-Sánchez, C. Koch, S. Schuwalow, S. A. Khan, T. Stankevič, S. Francoual, J. R. Mardegan, J. Krieger, V. N. Strocov, J. Stahn, C. A. F. Vaz, M. Ramakrishnan, U. Staub, K. Lefmann, G. Aeppli, J. Arbiol,  and P. Krogstrup, ACS Appl. Mater. Interfaces 12, 8780 (2020c).
  • Grove-Rasmussen et al. (2018) K. Grove-Rasmussen, G. Steffensen, A. Jellinggaard, M. H. Madsen, R. Žitko, J. Paaske,  and J. Nygård, Nat. Comm. 9, 2376 (2018).
  • Bruus and Flensberg (2004) H. Bruus and K. Flensberg, Many-body quantum theory in condensed matter physics: an introduction (Oxford university press, 2004).
  • Koller et al. (2010) S. Koller, M. Grifoni, M. Leijnse,  and M. R. Wegewijs, Phys. Rev. B 82, 235307 (2010).
  • Koch et al. (2004) J. Koch, F. von Oppen, Y. Oreg,  and E. Sela, Phys. Rev. B 70, 195107 (2004).
  • Ozaki (2007) T. Ozaki, Phys. Rev. B 75, 035123 (2007).

I Supplemental Material

II Sample preparation

InAs wires with hexagonal cross section were grown to a length of 10μ\sim 10~\mum and diameter of 120\sim 120 nm using molecular beam epitaxy Krogstrup et al. (2015). Partly-overlapping, two-faceted EuS (as grown thickness 8 nm) and Al (as grown thickness 6 nm) shells were grown in situ using electron beam evaporation, as shown in the main-text Fig. 1(a) Liu et al. (2020c, a). Devices were fabricated on a Si substrate with 200 nm SiOx capping. Coulomb islands were formed by wet-etching (MF-321 photoresist developer, 30 s, room temperature) and contacting the exposed InAs/EuS with Ti/Au (5/150 nm) after in situ Ar milling (15 W, 7 min). Devices were then coated with a HfO2 (8 nm) dielectric layer, followed by the deposition of Ti/Au (5/150 nm) gates patterned using electron beam lithography. Additional details can be found in Ref. Vaitiekėnas et al. (2021).

III Measurements

The samples were cooled to base temperature of the dilution refrigerator at zero applied magnetic field. Following cooldown, an external magnetic field was applied μ0H=150\mu_{0}H_{\parallel}=150 mT along the wire axis then returned to zero. Unless otherwise noted, the measurements were carried out at zero applied field. Differential conductance, G=G= dI/I/dVV, was measured by sourcing voltage bias, VV, through one of the outer leads, floating the opposite end of the wire, and draining the current, II, through the common lead between the two islands, as shown for the 400 nm island in the main-text Fig. 1(b) and for the 800 nm island in Fig. S2(a). The islands were tuned into the Coulomb blockade regime using negative voltages applied to the cutter gates, VC1V_{\rm C1} and VC2V_{\rm C2}, and back gate, VBGV_{\rm BG} [labeled only in Fig. S2(a)]. Over a range much larger than the Coulomb peak spacing, VBGV_{\rm BG} also tuned the chemical potential of the island. The upper-gate voltage, VUV_{\rm U}, on the side coated with the Al shell, was predominantly used to tune the occupancy of the hybrid island, whereas the lower-gate voltage, VLV_{\rm L}, on the EuS shell side, was used to tune the charge carrier density in the semiconductor.

IV Model

We consider a superconducting island hosting a subgap state that can be spin-slit (for example, due to exchange-coupling to the ferromagnetic insulator) by an energy EZE_{\rm Z}. For simplicity, we describe the continuum of states as a spin-degenerate quasiparticle state at energy Δ\Delta, using the so-called zero bandwidth model Grove-Rasmussen et al. (2018). We take the charging energy of the island, ECE_{\rm C}, to be the largest and the coupling to the leads to be the smallest energy scales in the system, allowing us to treat the electron tunneling in a perturbative way.

The Hamiltonian of the system is given by

H=HL+HI+HT,H=H_{\rm L}+H_{\rm I}+H_{\rm T}\,, (1)

where the leads are described by

HL=ν,k,σξνkσcνkσcνkσ,H_{\rm L}=\sum_{\rm\nu,k,\sigma}\xi_{\nu k\sigma}c^{\dagger}_{\nu k\sigma}c_{\nu k\sigma}\,, (2)

with the energy, ξνkσ\xi_{\nu k\sigma}, and the annihilation operator, cνkσc_{\nu k\sigma}, of an electron in lead ν{L,R}\nu\in\{L,R\} with momentum kk and spin σ{,}\sigma\in\{\uparrow,\downarrow\}. We assume that each lead remains in internal equilibrium described by a Fermi-Dirac distribution with chemical potential μν\mu_{\nu}.

The superconducting island is described by

HI=j,σϑjσγjσγjσ+Eel(N),H_{\rm I}=\sum_{j,\sigma}\vartheta_{j\sigma}\gamma_{j\sigma}^{\dagger}\gamma_{j\sigma}+E_{\rm el}(N)\,, (3)

where j{0,Δ}j\in\{0,\Delta\} is the state index, ϑjσ\vartheta_{j\sigma} is the (spin-dependent) energy of subgap and continuum states, γjσ\gamma_{j\sigma} is the Bogoliubov quasiparticle annihilation operator, and EelE_{\rm el} is the electrostatic repulsion term that depends on the number of electrons in the island, N=2NC+nqpN=2N_{\rm C}+n_{\rm qp}, where NCN_{\rm C} is the number of Cooper pairs and nqpn_{\rm qp} is the total charge in the quasiparticle states. For the subgap state, we take ϑ0=ε\vartheta_{0\downarrow}=\varepsilon and ϑ0=ε+EZ\vartheta_{0\uparrow}=\varepsilon+E_{\rm Z}. For simplicity, we take spin-degenerate continuum states at ϑ1=ϑ1=Δ\vartheta_{1\downarrow}=\vartheta_{1\uparrow}=\Delta. Similar results are found for spin-split continuum states. The annihilation operator for an electron in the island, djσd_{j\sigma}, is related to γjσ\gamma_{j\sigma} by

γjσ=ujdjσρσvjeiϕdjσ¯,\gamma_{j\sigma}=u_{j}d_{j\sigma}-\rho_{\sigma}\,v_{j}e^{-i\phi}d_{j\bar{\sigma}}^{\dagger}\,, (4)

where uju_{j} and vjv_{j} are the Bogoliubov coefficients, ρ/=±1\rho_{\uparrow/\downarrow}=\pm 1 depending on the spin, ϕ\phi is the superconducting phase operator such that eiϕe^{i\phi} creates a Cooper pair on the island, and σ¯\bar{\sigma} is the spin opposite to σ\sigma. The electrostatic repulsion term is given by

Eel(N)=EC(NnG)2,E_{\rm el}(N)=E_{\rm C}(N-n_{\rm G})^{2}\,, (5)

where nGn_{\rm G} is the dimensionless gate-induced charge offset.

The tunneling Hamiltonian is described by

HT=ν,k,j,σ(tνkσcνkσdjσ+h.c.)=ν,k,j,σ[tνkσcνkσ(ujγjσ+σvjeiϕγjσ¯)+h.c.],\begin{split}H_{T}&=\sum_{\nu,k,j,\sigma}\left(t_{\nu k\sigma}c_{\nu k\sigma}^{\dagger}d_{j\sigma}+h.c.\right)\\ &=\sum_{\nu,k,j,\sigma}\left[t_{\nu k\sigma}c_{\nu k\sigma}^{\dagger}\left(u_{j}\gamma_{j\sigma}+\sigma v_{j}e^{-i\phi}\gamma_{j\bar{\sigma}}^{\dagger}\right)+h.c.\right]\,,\end{split} (6)

where tt denotes the tunneling amplitudes between the island and leads.

V Formalism

The transport properties through the superconducting island are calculated using the T-matrix formalism Bruus and Flensberg (2004). The transition rate between initial, |i\left|i\right\rangle, and final, |f\left|f\right\rangle, states can be computed by

Γif=2π|f|T|i|2Wifδ(EiEf),\Gamma_{i\to f}=2\pi\left|\left\langle f\left|T\right|i\right\rangle\right|^{2}W_{if}\delta(E_{i}-E_{f})\,, (7)

where WifW_{if} weights the rate through thermal distributions of initial and final states at energies EiE_{i} and EfE_{f}, respectively, δ\delta is the Dirac delta function and

T=HT+HT1EiHLHI+i0+T,T=H_{\rm T}+H_{\rm T}\frac{1}{E_{i}-H_{\rm L}-H_{\rm I}+i0^{+}}T\,, (8)

which can be truncated at the desired order. Here the term linear in HTH_{\rm T} describes the sequential tunneling; the higher order terms describe the cotunneling contributions, which become progressively more important when the tunneling coupling between the island and the leads increases.

The quantum state of the island is described by |a\left|a\right\rangle=|N,NC,n\left|N,N_{\rm C},\textbf{n}\right\rangle, where n is a vector representing the occupation of the subgap and continuum excited states. The time derivative of the occupation probability of a given state can be written as

P˙a=b[ΓabPa+ΓbaPb],\dot{P}_{a}=\sum_{b}\left[-\Gamma_{a\to b}P_{a}+\Gamma_{b\to a}P_{b}\right]\,, (9)

which describes the stationary probability of occupation, PbstatP^{\rm{stat}}_{b}, by imposing P˙bstat=0\dot{P}^{\rm{stat}}_{b}=0 and bPbstat=1\sum_{b}P^{\rm{stat}}_{b}=1. The current through the device can be computed using the stationary distribution of probabilities and the rates. The tunneling current is given by

I=a,b[s2Γbaseq+δn(LR)Γbacot]Pbstat,\displaystyle\begin{split}I=\sum_{a,b}\left[\frac{s}{2}\Gamma^{\rm{seq}}_{b\to a}+\delta n(L\to R)\,\Gamma^{\rm{cot}}_{b\to a}\vphantom{\frac{1}{2}}\right]P^{\rm{stat}}_{b}\,,\end{split} (10)

where s=+1s=+1 for rightwards sequential tunneling and 1-1 for the opposite direction, δn(LR)\delta n(L\to R) is the net charge transferred from left to right in a cotunneling process, and the sum runs over all the possible rates connecting the island state |b\left|b\right\rangle with any state |a\left|a\right\rangle.

VI Sequential tunneling rates

The sequential rates are given by

Γ|N,NC,n|N+1,NC,n’seq=Γν|uj|2nF(EfEiμν)Γ|N,NC,n|N1,NC,n’seq=Γν|uj|2nF(μν+EfEi)Γ|N,NC,n|N+1,NC+1,n’seq=Γν|vj|2nF(EfEiμν)Γ|N,NC,n|N1,NC1,n’seq=Γν|vj|2nF(μν+EfEi),\displaystyle\begin{split}\Gamma^{\rm{seq}}_{\left|N,N_{\rm C},\textbf{n}\right\rangle\to\left|N+1,N_{\rm C},\textbf{n'}\right\rangle}&=\Gamma_{\nu}\left|u_{j}\right|^{2}n_{\rm F}(E_{f}-E_{i}-\mu_{\nu})\,\\ \Gamma^{\rm{seq}}_{\left|N,N_{\rm C},\textbf{n}\right\rangle\to\left|N-1,N_{\rm C},\textbf{n'}\right\rangle}&=\Gamma_{\nu}\left|u_{j}\right|^{2}n_{\rm F}(\mu_{\nu}+E_{f}-E_{i})\,\\ \Gamma^{\rm{seq}}_{\left|N,N_{\rm C},\textbf{n}\right\rangle\to\left|N+1,N_{\rm C}+1,\textbf{n'}\right\rangle}&=\Gamma_{\nu}\left|v_{j}\right|^{2}n_{\rm F}(E_{f}-E_{i}-\mu_{\nu})\,\\ \Gamma^{\rm{seq}}_{\left|N,N_{\rm C},\textbf{n}\right\rangle\to\left|N-1,N_{\rm C}-1,\textbf{n'}\right\rangle}&=\Gamma_{\nu}\left|v_{j}\right|^{2}n_{\rm F}(\mu_{\nu}+E_{f}-E_{i})\,,\end{split} (11)

where nFn_{\rm F} is the Fermi-Dirac distribution function and μν\mu_{\nu} is the chemical potential of the lead ν\nu. Here, we have used the wideband approximation, where the tunneling rates are energy independent and Γν=2πρF|tν|2\Gamma_{\nu}=2\pi\rho_{F}|t_{\nu}|^{2}, with the lead density of states at the Fermi level ρF\rho_{F}. Note that for these rates n and n\textbf{n}^{\prime} differ by the occupation of one state.

VII Cotunneling rates

We consider the cotunneling rates transferring an electron from one lead to the other, which are the dominant ones in the limit ECmax(Δ,vμLμR)E_{\rm C}\gg\max\left(\Delta,\,v\equiv\mu_{\rm L}-\mu_{\rm R}\right), where vv is the source-drain voltage bias used in calculations. These processes, which conserve the charge on the island, can be expressed by

F(ω1)|f|T|i|2=|m1,νf|HT(ν)|m1m1|HT(ν¯)|iEm1Eiω1+m2,νf|HT(ν¯)|m2m2|HT(ν)|iEm2Ef+ω1|2,\begin{split}F(\omega_{1})&\equiv\left|\left\langle f\left|T\right|i\right\rangle\right|^{2}=\\ &\left|\sum_{m_{1},\nu}\frac{\left\langle f\left|H_{\rm T}(\nu)\right|m_{1}\right\rangle\left\langle m_{1}\left|H_{\rm T}(\bar{\nu})\right|i\right\rangle}{E_{m_{1}}-E_{i}-\omega_{1}}\right.\\ &\left.+\sum_{m_{2},\nu}\frac{\left\langle f\left|H_{\rm T}(\bar{\nu})\right|m_{2}\right\rangle\left\langle m_{2}\left|H_{\rm T}(\nu)\right|i\right\rangle}{E_{m_{2}}-E_{f}+\omega_{1}}\right|^{2}\,,\end{split} (12)

where HT(ν)H_{\rm T}(\nu) describes the tunneling between ν\nu lead and the island, ν¯\bar{\nu} denotes a lead opposite to ν\nu, and m1,2m_{1,2} are virtual intermediate states. To derive this expression we have imposed energy conservation, which leads to a function dependent on the energy of the tunneling electron from/to one of the leads, ω1\omega_{1}. The cotunneling rate can be written as

Γifcot=2π𝑑ω1F(ω1)nF(ω1μL)×nF(μR+EfEiω1).\begin{split}\Gamma^{\rm{cot}}_{i\to f}=2\pi\int d\omega_{1}\,&F(\omega_{1})\,n_{\rm F}(\omega_{1}-\mu_{\rm L})\,\\ \times\,&n_{\rm F}(\mu_{\rm R}+E_{f}-E_{i}-\omega_{1})\,.\end{split} (13)

This expression for the cotunneling rates is divergent due to the appearing sequential tunneling. To avoid the divergent behaviour, we regularize the divergences as explained in Ref. Koller et al. (2010). The resulting integral can be formally solved analytically Koch et al. (2004), which leads to a complicated expression involving special functions. In the limit of T/EC103T/E_{\rm C}\gtrsim 10^{-3}, it turns out to be more computationally efficient to simplify it by expanding nFn_{\rm F} into a sum of complex Matsubara-Ozaki frequencies Ozaki (2007)

nF(ωμ)12=αrα1ωμ+iβα,n_{\rm F}(\omega-\mu)-\frac{1}{2}=\sum_{\alpha}r_{\alpha}\frac{1}{\omega-\mu+i\beta_{\alpha}}\,, (14)

where βα\beta_{\alpha} and rαr_{\alpha} are the approximated Matsubara frequencies and residues, respectively. Finally, Eq. (13) can be evaluated using the residue theorem yielding

Γifcot= 8πImα=0rα{nF(μR+EfEiμL+iβα)×[F(μLiβα)F(μR+EfEi+iβα)]}.\begin{split}\Gamma^{\rm{cot}}_{i\to f}=&\,8\pi\,\mbox{Im}\sum_{\alpha=0}^{\infty}r_{\alpha}\left\{n_{\rm F}(\mu_{\rm R}+E_{f}-E_{i}-\mu_{\rm L}+i\beta_{\alpha})\right.\\ &\left.\times\left[F(\mu_{\rm L}-i\beta_{\alpha})-F(\mu_{\rm R}+E_{f}-E_{i}+i\beta_{\alpha})\right]\right\}\,.\end{split} (15)

This sum can be truncated at α100\alpha\approx 100 Matsubara frequencies for the parameters used in the calculations.

VIII Transport calculations

Calculated Coulomb spectra for a spin-degenerate (EZ=0E_{\rm Z}=0) and weakly spin-split (EZ<ΔE_{\rm Z}<\Delta) subgap states are shown in Figs. S3(a) and S3(g). The colored ticks mark the onset of inelastic cotunneling events that excite the system into a higher-energy state, resulting in a step-like increase in conductance. The transitions in the odd and even valleys are represented separately in panels (b) and (c) for the spin-degenerate, and panels (h) and (i) for the weakly spin-split cases. Some examples of the cotunneling transport mechanisms for the two cases are illustrated in panels (d)–(f) and (j)–(l), respectively. In general, there are four cotunneling lines in both spin-degenerate and weakly spin-split cases (except for the specific instances where ε\varepsilon is fine tuned such that two transitions are degenerate in energy). In contrast, our experimental data show only two cotunneling steps (see, for example, Fig. 3 in the main text). A qualitative line-cut comparison between experimental data and spin-polarized as well as spin-degenerate cases is shown for the 400 nm island on wire 1 in Fig. S4. Aggregated line-cuts from several consecutive even and odd Coulomb valleys measured for the 400 nm island on wire 2 display two cotunneling steps, independent of lower-gate voltage and subgap-state energy, see Fig. S5.

The measured cotunneling features decrease in energy without splitting as an external magnetic field is applied, see Fig. S6(b). For comparison, we calculate conductance for a spin-polarized and two spin-degenerate cases—with and without magnetic hysteresis. As an input, we use approximated expressions for the field dependence of the superconducting gap and the subgap states. We describe the gap closing as Δ(H)=Δ0[1(H/HC)2]\Delta(H_{\parallel})=\Delta_{0}\left[1-\left(H_{\parallel}/H_{\rm C}\right)^{2}\right], with critical field HC=60H_{\rm C}=60 mT and the zero-field gap Δ0=60μ\Delta_{0}=60~\mueV, see the black curves in Fig. S6(c), (e), and (g). For simplicity we assume that the subgap state depends linearly on the field, see the red and blue curves in Fig. S6(c), (e), and (g). To account for the asymmetry in the measured data, we include magnetic hysteresis of 10 mT (consistent with previous experiments Vaitiekėnas et al. (2021)) in the spin-polarized and one of the spin-degenerate cases. The magnetic field in the spin-degenerate case leads to the appearance of additional cotunneling features that are not observed experimentally, see Fig. S6(f) and (h). We therefore find that the measurements agree best with the spin-polarized spectrum shown in Fig. S6(d).

We note that the number of cotunneling steps increases in similar hybrid island devices without EuS shell (see the replotted data from Ref. Higginbotham et al. (2015) in Fig. S7) consistent with Fig. S3(a).

These observations together suggest that the investigated hybrid islands are in the strongly spin-split limit (EZΔE_{Z}\gtrsim\Delta), discussed in the main-text Fig. 2.

The relative height of the cotunneling steps in the even and odd Coulomb valleys depends on the relative phase, φ\varphi, between the Bogoliubov components of the subgap state, u0|u0|exp(iφ)u_{0}\equiv|u_{0}|\exp(i\varphi) and v0|v0|v_{0}\equiv|v_{0}|, as illustrated in Fig. S9 for EZΔE_{\rm Z}\gg\Delta and ε=0\varepsilon=0. The effect can be explained by the interference between different cotunneling mechanisms that involve the same initial and final states. However, φ\varphi cannot be determined unambiguously in our setup, as the strong spin splitting and gate-dependent ε\varepsilon makes the global ground state unknown. This remains an open problem, relevant for distinguishing trivial and topological states.

In the experiment, the magnitude and, in some cases, the sign of ε\varepsilon can be tuned using electrostatic gate electrodes (Fig. 4 in the main text). In our model, we account for this behavior by changing ϑ0σ\vartheta_{0\sigma} in Eq. (3). A change in the sign of ε\varepsilon is equivalent to exchanging the Bogoliubov components of the corresponding subgap state. In Fig. S10 we show the calculated conductance in the strongly spin-split case for ε\varepsilon greater than, equal to, and less than zero. For ε=0\varepsilon=0, the Coulomb blockade is 1ee-periodic—the onsets of the lowest cotunneling steps in both odd and even valleys align at v=Δv=\Delta. For ε\varepsilon away from zero, the spectrum is even-odd periodic with the inelastic onset at v=Δεv=\Delta-\varepsilon in the odd and v=Δ+εv=\Delta+\varepsilon in the even valleys. The sign of ε\varepsilon determines the relative size of odd and even Coulomb diamonds.

Refer to caption
Fig. S1: Differential conductance, GG, measured for the 400 nm island on wire 1 at various lower-gate voltages, VLV_{\rm L}, as a function of bias, VV, and upper-gate voltage, VUV_{\rm U}. All spectra display even-odd periodic Coulomb blockade and a step in conductance around V=120μV=120~\muV. Around zero and for positive VLV_{\rm L} settings, additional steps in conductance at valley-dependent VV can be seen. For more negative VLV_{\rm L} these steps fade out, and negative differential conductance becomes apparent.
Refer to caption
Fig. S2: (a) Scanning electron micrograph of wire 1 with color-enhanced setup for the longer, 800 nm island. (b) Differential conductance, GG, as a function of bias voltage, VV, and upper-gate voltage, VUV_{\rm U}, shows even-odd alternation with the conductance-step onsets alternating between V=30V=30 and 140μ140~\muV, as well as around V=170μV=170~\muV in both valleys. The data were taken at VL=1V_{\rm L}=-1 V. (c) Similar to (b) but taken at VL=0V_{\rm L}=0 with the lower energy conductance-step onsets alternating between roughly V=20V=20 and 160μ160~\muV from diamond to diamond and the higher energy onset around V=160μV=160~\muV in both valleys.
Refer to caption
Fig. S3: (a) Calculated differential conductance, gg, as a function bias, vv, and offset charge, nGn_{\rm G}, for the model of a Coulomb island with a superconducting gap Δ\Delta, charging energy EC=5ΔE_{\rm C}=5\Delta, a spin-degenerate subgap state ε=Δ/4\varepsilon=\Delta/4 with the Bogoliubov components u0=v0=1/2u_{0}=v_{0}=1/\sqrt{2}, and Zeeman energy EZ=0E_{\rm Z}=0. The odd valley displays a relatively higher background conductance down to v=0v=0. Steps in gg are visible at v=Δεv=\Delta-\varepsilon in odd, and v=2εv=2\varepsilon in even Coulomb valleys as well as at v=Δ+εv=\Delta+\varepsilon and 2Δ2\Delta in both Coulomb valleys. The calculations were done for finite temperature kBT=0.03Δk_{\rm B}\,T=0.03\,\Delta. (b) and (c) Schematic representation of superconducting density of states for odd (b) and even (c) valleys in (a) with possible inelastic cotunneling excitations indicated by arrows. (d) In the odd ground state, the quasiparticle in the bound state can exit the island, while another quasiparticle with an opposite spin from a lead tunnels back into the bound state. The process does not require energy as at EZ=0E_{\rm Z}=0 the subgap state is spin degenerate. (e) In both occupancies, a Cooper pair can be broken-up with the spin-down quasiparticle leaving the island and the spin-up quasiparticle being excited into the spin-degenerate bound state, while another quasiparticle from a lead tunnels into the continuum. The total energy cost of the process is Δ+ε\Delta+\varepsilon. (f) Similar to (e) but the quasiparticle entering from the lead is spin-down and tunnels into the bound state. (g)-(l) Similar to (a)-(f) but for a weakly spin-split subgap state, offsetting the processes marked with yellow and green by EZE_{\rm Z}, while the processes marked with red and blue remain unaltered. The spectrum in (g) is calculated using EZ=0.42ΔE_{Z}=0.42\,\Delta, while the other parameters are the same parameters as in (a).
Refer to caption
Fig. S4: (a) Black curve, left axis: Differential conductance, GG, line-cut taken from the main-text Fig. 1(c) at VU=0.474V_{\rm U}=-0.474 V, in the middle of a smaller Coulomb valley. GG increases in a step-like manner at V=30μV=30~\mueV and then again around 120μ120~\mueV. Red curve, right axis: Calculated conductance, gg, through a hybrid island with a superconducting gap Δ=60μ\Delta=60~\muV, spin-polarized subgap state at ε=30μ\varepsilon=30~\muV, and an odd offset charge, nG=5n_{\rm G}=5e, showing two cotunneling steps at Δε\Delta-\varepsilon and 2Δ2\Delta. Blue curve, right axis: Same as the red curve, but for a spin-degenerate bound state, showing a finite background conductance starting at V=0V=0 as well as three cotunneling steps at Δε\Delta-\varepsilon, Δ+ε\Delta+\varepsilon, and 2Δ2\Delta. (b) Same as (a) but the experimental data were taken from the middle of a larger Coulomb valley at VU=0.409V_{\rm U}=-0.409 V, showing two step-like increases at V=90V=90 and 120μ120~\muV; the calculations were done for an even offset charge, nG=6n_{\rm G}=6e, showing two cotunneling steps at Δ+ε\Delta+\varepsilon and 2Δ2\Delta for the spin-polarized case, and an additional third step at 2ε2\varepsilon in the spin-degenerate case.
Refer to caption
Fig. S5: (a) Same data as in the main-text Fig. 3(a): Differential conductance, GG, as a function of source-drain bias, VV, and upper gate voltage, VUV_{\rm U}, for the 400 nm island on wire 2. The data were taken at a fixed lower-gate voltage VL=0.2V_{\rm L}=0.2 V. A clear even-odd Coulomb blockade pattern displays two cotunneling steps in both valleys: around 60 and 180 μ\mueV in the smaller, and 120 and 180 μ\mueV in the larger valley. (b) Conductance line-cuts taken from (a) at VUV_{\rm U} values marked by the ticks, plotted against the absolute value of VV. Red (black) curves are from the smaller (larger) Coulomb diamonds. (c) Calculated conductance, gg, through a hybrid island with a superconducting gap Δ=90μ\Delta=90~\muV and spin-polarized subgap state at ε=30μ\varepsilon=30~\muV, for even (black) and odd (red) offset charges. Both curves display two cotunneling steps at Δ+ε\Delta+\varepsilon and 2Δ2\Delta for even occupancy, and Δε\Delta-\varepsilon and 2Δ2\Delta for odd occupancy. (d) Similar to (c) but for a spin-degenerate subgap state. Odd curve displays a relatively larger background conductance. Both even and odd curves display three cotunneling steps at Δε\Delta-\varepsilon, Δ+ε\Delta+\varepsilon, and 2Δ2\Delta. (e)-(h) Similar to (a)-(d) but for VL=0V_{\rm L}=0 and ε0\varepsilon\approx 0. The Coulomb blockade is 1ee periodic, with all valleys displaying two cotunneling steps around 90 and 180 μ\mueV. Independent of lower-gate voltage, the measured data display only two cotunneling steps in all Coulomb valleys, with comparable background conductances, suggesting transport trough spin-polarized subgap states.
Refer to caption
Fig. S6: (a) Differencial conductance, GG, as a function of voltage bias, VV, and upper-gate voltage, VUV_{\rm U}, for the 400 nm island on wire 1 at zero applied magnetic field and the same gate configuration as in the main-text Fig. 1(c). (b) GG as a function of VV and external magnetic field, μ0H\mu_{0}H_{\parallel}, applied parallel to the wire axis, measured at VU=0.421V_{\rm U}=-0.421 V. Sweep direction indicated by arrow. The onsets of the two conductance steps decrease in VV for HH_{\parallel} away from zero, but do not split. A small hysteresis of 10\sim 10 mT can be seen. (c) Superconducting gap, Δ\Delta, and the spin-up branch of a subgap state, ε\varepsilon_{\uparrow}, values as a function of an HH_{\parallel}, that were used as an input for spin-polarized model. In this case the subgap state is assumed to display 10 mT hysteresis. (d) Calculated conductance, gg, through a hybrid island with a spin-polarized subgap state and an even parity, as a function of HH_{\parallel}, taking Δ\Delta and ε\varepsilon_{\uparrow} from (c). The spectrum displays only two cotunneling features throughout the superconducting range. (e) and (g) Similar to (c) and (d) but assuming a subgap state that is spin-degenerate at H=0H_{\parallel}=0. The spectrum in (d) exhibits three cotunneling features that split in field. (g) and (h) Similar to (e) and (g) but with 10 mT hysteresis, showing a complicated pattern of up to four cotunneling features.
Refer to caption
Fig. S7: (a) Differential conductance, GG, for a 310 nm InAs/Al hybrid island without EuS as a function of gate voltage, VGV_{\rm G}, and source–drain voltage, VV, at zero external magnetic field. Two clear steps in conductance can be seen for both charge occupancies around V=210V=210 and 300μ300~\mueV. We associate these features with excitations at Δ+ε\Delta+\varepsilon and 2Δ2\Delta, respectively, yielding Δ=150μ\Delta=150~\mueV and ε=60μ\varepsilon=60~\mueV. Two additional, less pronounced features at the edges of the Coulomb diamonds (see dashed arrows) alternate between V=90V=90 and 120μ120~\mueV and can be associated with Δε\Delta-\varepsilon and 2ε2\varepsilon, respectively. The data are replotted from Ref. Higginbotham et al. (2015).
Refer to caption
Fig. S8: (a) Differential conductance, GG, measured for the 400 nm island on wire 2 as a function of bias, VV, and lower-gate voltage, VLV_{\rm L}, which is positioned on the side of the wire with the EuS shell [see the main-text Fig. 1(a)]. This gate tunes carrier density in the semiconductor as well as the island occupancy. The data span 36 Coulomb diamonds, taken at fixed VU=0.5V_{\rm U}=-0.5 V. (b) GG dependence on VUV_{\rm U} taken at VL=V_{\rm L}= -0.3 V. (c)-(e) Same as (b) but measured at VL=V_{\rm L}= -0.2 V (c), -0.1 V (d), and 0.1 V (e). Similar data taken at VL=0V_{\rm L}=0 and 0.2 V are shown in the main-text Fig. 3.
Refer to caption
Fig. S9: Calculated differential conductance, gg, as a function bias, vv, and offset charge, nGn_{\rm G}, for a superconducting island with bound state at ε=0\varepsilon=0 with the relative phase between the Bogoliubov components φ=0\varphi=0 (a), π/2\pi/2 (b), and π\pi (c), where φ\varphi is given by u0=exp(iφ)v0u_{0}=\exp(i\varphi)v_{0}. The other parameters are the same as in the main-text Fig. 2(a). Depending on the value of φ\varphi, the amplitude of the conductance step at v=Δv=\Delta in the odd valley can be (a) higher than, (b) equal to, or (c) lower than the step in the even valley.
Refer to caption
Fig. S10: Calculated differential conductance, gg, as a function of bias, vv, and offset charge, nGn_{\rm G}, for a superconducting island with a bound state at ε=Δ/4\varepsilon=\Delta/4 (a), 0 (b), and Δ/4-\Delta/4 (c). The other parameters are the same as in the main-text Fig. 2(a). Depending on the value of ε\varepsilon, the lowest inelastic onset in the odd valley can be at (a) lower, (b) same, or (c) higher value of vv compared to the lowest cotunneling onset in the even valley.
Refer to caption
Fig. S11: (a) Differential conductance, GG, for the 800 nm island on wire 2 as a function of bias, VV, over an extended range of lower-gate voltage, VLV_{\rm L}. The Coulomb blockade pattern evolves from even-odd at VL=0.7V_{\rm L}=-0.7 V, through 1ee around VL=0.69V_{\rm L}=-0.69 V, to even-odd periodicity again at VL=0.67V_{\rm L}=-0.67 V, visible in both inelastic cotunneling onsets cic_{\rm i} and peak spacings sis_{\rm i}, where i=1i=1 and 2 denote Coulomb valleys with the same charge occupancies. i=1i=1 diamonds are larger than the i=2i=2 diamonds on both ends of the measured gate voltage range. The data were taken at VU=1.43V_{\rm U}=-1.43 V. (b) ε\varepsilon deduced from the data shown in (a) using the inelastic cotunneling onsets (black) and the peak-spacing difference (red). ε\varepsilon decreases from roughly 10μ10~\mueV to 0, but then increases again to 10μ10~\mueV as the gate voltage is increased. The black error bars represent standard errors from the cic_{\rm i} measurement at the positive and negative VV, whereas the red error bars were estimated by propagation of error from Lorentzian peak fitting and lever arm, η\eta, measurement.
Refer to caption
Fig. S12: (a) Differential conductance, GG, as a function of bias, VV, and upper-gate voltage, VUV_{\rm U}, for the 400 nm island on wire 2, at zero applied magnetic field and lower-gate voltage VL=0V_{\rm L}=0. Coulomb blockade is nearly 1ee periodic with finite bias conductance steps at V=90V=90 and 180μ180~\mueV in each diamond as well as discrete peaks at the degeneracy points. (b) Similar to (a) taken at finite external magnetic field μ0H=150\mu_{0}H_{\parallel}=150 mT applied parallel to the wire axis. Coulomb blockade shows finite, featureless conductance outside of the diamonds and no sign of excited states. (c) Zero-bias GG as a function of VUV_{\rm U} and μ0H\mu_{0}H_{\parallel}. Sweep direction indicated by arrow. Both the peak amplitude and position in VUV_{\rm U} changes nonmonotonically with HH_{\parallel}. (d) Same as (c) with sweep direction from positive to negative field. (e) ε\varepsilon deduced from the difference of two consecutive peak spacings indicated in (c) allowing to track the subgap state evolution with both HH_{\parallel} and VUV_{\rm U}. (f) Same as (e) but for the data from (d).
Refer to caption
Fig. S13: (a) Differential conductance, GG, as a function of bias, VV, and upper-gate voltage, VUV_{\rm U}, for the 800 nm island on wire 2, at zero applied magnetic field and lower-gate voltage VL=0.7V_{\rm L}=-0.7 V. Coulomb blockade is nearly 1ee periodic with two inelastic cotunneling onsets at V=60V=60 and 120μ120~\mueV in each diamond as well as discrete peaks at the degeneracy points. (b) Similar to (a) taken at μ0H=150\mu_{0}H_{\parallel}=150 mT. Coulomb blockade shows finite, featureless conductance outside of the diamonds and no sign of excited states. (c) Zero-bias GG as a function of VUV_{\rm U} and μ0H\mu_{0}H_{\parallel}. Sweep direction indicated by arrow. Both the peak amplitude and position in VUV_{\rm U} changes nonmonotonically with HH_{\parallel}. (d) Same as (c) with sweep direction from positive to negative field. (e) ε\varepsilon deduced from the difference of two consecutive peak spacings indicated in (c) allowing to track the subgap state evolution with both HH_{\parallel} and VUV_{\rm U}. (f) Same as (e) but for the data from (d).
Refer to caption
Fig. S14: (a) Differential conductivity, GG, as a function of bias, VV, and upper-gate voltage, VUV_{\rm U}, for the 800 nm island on wire 2, at lower-gate voltage VL=0V_{\rm L}=0 V. The spectrum shows an even-odd periodic Coulomb blockade with the conductance suppressed below the noise floor in the larger valleys and small, but finite conductance throughout the smaller valleys. No further features are observed up to roughly V=180μV=180~\mueV where GG increases in both Coulomb diamonds. Note the logarithmic color scale. (b) Calculated differential conductance, gg, as a function of bias, vv, and offset charge, nGn_{\rm G}, for a modeled superconducting Coulomb island with ε=Δ\varepsilon=\Delta, EZ=0E_{\rm Z}=0, EC=5ΔE_{\rm C}=5\Delta, and asymmetric voltage bias given by the left- and right-lead potential ratio μL/μR=5\mu_{L}/\mu_{R}=5. The spectrum qualitatively agrees with the experimental data in (a).