Exploring the energy spectrum of a four-terminal Josephson junction: Towards topological Andreev band structures Tommaso Antonelli, Marco Coraiola, David Christian Ohnmacht, Aleksandr E. Svetogorov, Deividas Sabonis, Sofieke C. ten Kate, Erik Cheah, Filip Krizek, Rüdiger Schott, Juan Carlos Cuevas, Wolfgang Belzig, Werner Wegscheider, Fabrizio Nichele Abstract Hybrid multiterminal Josephson junctions (JJs) are expected to harbor a novel class of Andreev bound states (ABSs), including topologically nontrivial states in four-terminal devices. In these systems, topological phases emerge when ABSs depend on at least three superconducting phase differences, resulting in a three-dimensional (3D) energy spectrum characterized by Weyl nodes at zero energy. Here, we realize a four-terminal JJ in a hybrid Al/InAs heterostructure, where ABSs form a synthetic 3D band structure. We probe the energy spectrum using tunneling spectroscopy and identify spectral features associated with the formation of a tri-Andreev molecule, a bound state whose energy depends on three superconducting phases and, therefore, is able to host topological ABSs. The experimental observations are well described by a numerical model. The calculations predict the appearance of four Weyl nodes at zero energy within a gap smaller than the experimental resolution. These topological states are theoretically predicted to remain stable within an extended region of the parameter space, well accessible by our device. These findings establish an experimental foundation to study high-dimensional synthetic band structures in multiterminal JJs, and to realize topological Andreev bands. I Introduction Josephson junctions (JJs) are key elements of superconducting circuits, used in quantum technology applications and fundamental research. In a superconducting-normal conductor-superconducting junction, supercurrent transport is mediated by Andreev bound states (ABSs), electron-hole superposition states confined within the normal area. The ABS energy depends on the phase difference between the superconducting wave functions of the two leads [ 1 , 2 , 3 , 4 ] . Recently, these electronic modes have been the subject of extensive study [ 5 , 6 , 7 , 8 , 9 , 10 , 11 , 12 ] , including the coherent manipulation of ABSs [ 13 , 14 , 15 , 16 ] and the exploration of topological superconductivity [ 17 , 18 , 19 , 20 ] . In multiterminal JJs (MTJJs), where three or more superconducting terminals are linked to a single normal scattering region, ABSs form synthetic band structures which are expected to host a wide range of properties not attainable in two-terminal devices. Among the most intriguing prospects is the potential to realize topologically nontrivial phases in the three-dimensional (3D) band structure of four-terminal JJs (4TJJs), with Weyl nodes arising in the energy spectrum [ 21 , 22 , 23 , 24 , 25 , 26 , 27 ] . Topological phases in these systems are inherently robust with respect to external perturbations [ 21 ] , making them particularly appealing for applications in quantum information processing [ 28 , 29 ] and spintronics [ 30 ] . A first set of studies on MTJJs focused on their transport properties, including the signatures of Cooper pair quartets [ 31 , 32 , 33 , 34 , 35 ] and the flow of supercurrents across multiple superconducting leads [ 36 , 37 , 38 , 39 , 40 , 41 , 42 ] . Recently, MTJJs have been proposed as a platform to realize Andreev molecules—a system where ABSs hybridize due to the spatial overlap of their wave functions [ 43 , 44 , 45 , 46 , 47 ] , resulting in delocalized states that extend across all leads and exhibit nonlocal Josephson effect [ 43 , 48 , 49 , 50 , 51 ] . In Andreev molecules, two primary coherent transport processes occur: double elastic cotunneling and double crossed Andreev reflection [ 52 , 31 , 43 ] , both essential for the generation of Cooper pair multiplets [ 31 , 32 , 53 ] and for engineering Kitaev chains [ 54 ] in quantum dot arrays [ 55 , 56 , 57 , 58 , 59 , 60 , 61 , 62 , 63 , 64 , 65 ] . Detailed understanding of Andreev band structures in multiterminal devices can be gained through local spectroscopy, which has been employed to probe hybridized ABSs [ 66 , 67 ] , as well as spin-split energy levels and ground state parity transitions [ 68 , 69 ] . In these experiments, phase biasing allowed the exploration of ABS spectra as a function of up to two superconducting phase differences [ 70 , 66 , 69 ] . However, realizing topological phases with nontrivial Chern numbers strictly requires independent tuning of three phase degrees of freedom [ 21 , 71 , 72 , 24 ] , a challenge that remains to be addressed. Figure 1: (a) False-colored scanning electron micrograph of the four-terminal device under study, showing the three-loop geometry. (b) Close-up view of the scattering area taken prior to the gate deposition. Gates structures are drawn on the image (yellow). (c) Tunneling spectroscopy measurement, showing the differential conductance G as a function of bias V_{\rm{sd}} and the current I_{\rm{L}} flowing in flux line L. (d) Differential conductance map measured at V_{\rm{sd}}=-175~\mu\rm{V} as a function of the flux-line currents I_{\rm{L}} and I_{\rm{R}} . (e) As in (d), but measured as a function of the magnetic fluxes \Phi_{\rm{L}} and \Phi_{\rm{R}} after the current-to-flux remapping (see text). The schematics on the right of (d) and (e) represent the orientations of the corresponding maps with respect to the cubic unit cell in the 3D flux space. In this work, we realize a 4TJJ where three phases are independently controlled through flux biasing. We probe the ABS energy spectrum of the system across the entire 3D phase space using tunneling spectroscopy. Moreover, we observe the simultaneous hybridization of three ABSs, i.e., the formation of a so called tri-Andreev molecule, when the three phases are tuned close to \pi . Our findings are supported by a theoretical model, which qualitatively reproduces the main features of the measured Andreev spectra. Furthermore, our simulations indicate that the Andreev bands undergo a phase-controlled topological transition in which hybridization induces a band inversion accompanied by the appearance of Weyl nodes. Due to the finite resolution of the tunneling spectroscopy, with a linewidth of \sim 15~\mu eV, the gapless states (Weyl nodes) cannot be experimentally distinguished from the gapped states. Finally, we study the robustness of the topological phase under variations of experimentally addressable parameters, finding that the regime best describing our device is well within the topological region. Overall, our work provides access to Andreev band structures in three synthetic dimensions, creating an experimental platform and practical guidelines for the realization of topological states in hybrid multiterminal devices. II Experimental setup and 3D phase control The device under study, shown in Fig. 1 (a), consists of a 4TJJ embedded in a triple-loop geometry. It is realized in an InAs/Al heterostructure [ 73 , 74 ] where the epitaxial Al layer is selectively removed to expose the III–V semiconductor below. Three flux-bias lines are patterned on top of a uniform dielectric layer to generate the external magnetic fluxes \Phi_{i} (i=\rm{L,M,R}) threading the three interconnected superconducting loops (L, M, R). This enables us to control the phase differences \phi_{i} between the four terminals (T1-T4). The latter couple with a common semiconducting region [see Fig. 1 (b)] where a superconducting island with diameter \sim 90 nm is left at its center to partially screen the probe gate voltages. By design, the minimum distance between neighboring terminals is 50 nm, while the distance between T1 and T4 is 220 nm. All these lengths are small in comparison with the superconducting coherence length in the InAs 2DEG, estimated to be 600~\rm{nm} [ 49 ] . Four gate electrodes on the dielectric layer are energized by voltages V_{g} ( g\in {TL, TR, H, JJ}), allowing for electrostatic tuning of the electron density in the InAs layer below. Tunneling spectroscopy of the scattering area is performed by measuring the differential conductance G across a tunneling barrier, formed by depleting the InAs region below the gates V_{\rm{TL}} and V_{\rm{TR}} (see Supplemental Material [ 75 ] , Sec. I for additional details). The device was measured in a dilution refrigerator with a base temperature of about 10~\mathrm{mK} using lock-in techniques, with a DC voltage bias V_{\rm{sd}} and an AC voltage bias of amplitude \delta V_{\rm{sd}}=3~\mu\mathrm{V} applied between the probe and the four terminals. More information about materials, fabrication and measurement setup is available in Ref. [ 66 ] . Figure 2: (a) Tunneling differential conductance G measured as a function of DC voltage bias V_{\rm{sd}} and magnetic flux \Phi_{\rm{L}} , with \Phi_{\rm{M}}=\Phi_{\rm{R}}=0 . (b) As in (a), but varying \Phi_{\rm{M}} with \Phi_{\rm{L}}=\Phi_{\rm{R}}=0 . (c) Tunneling differential conductance measured at fixed V_{\rm{sd}}=-200 \mu\rm{V} and plotted in the 3D flux unit cell for values G\geq 0.095\times 2e^{2}/h . The unit cell axes are also labeled using the crystallographic-like notation used to define specific directions in the flux space. (d,e) Flux-flux maps of two cube faces [as indicated by the colored squares in (c)], measured at V_{\rm{sd}}=-200 \mu\rm{V} , showing avoided crossing induced by the formation of bi-Andreev molecules. (f,g) G measured as a function of V_{\rm{sd}} along the directions (110) and (101), as indicated by the white arrows in (d) and (e), respectively. Figure 1 (c) shows a typical tunneling spectrum measured by varying the flux-line current I_{\rm{L}} and the DC bias V_{\rm{sd}} . The flux-dependent ABS spectrum is visible outside a transport gap of 350~\mu\rm{V}=2\Delta/e , that is due to the superconducting probe ( \Delta is the superconducting gap). Two flux-independent conductance peaks at V_{\rm{sd}}=\pm\Delta/e highlight the probe gap edges and are attributed to multiple Andreev reflection processes. Assuming a BCS-like density of states (DOS) for the superconducting probe with peaks at energy \pm\Delta , G is expected to have a resonance at a voltage \pm(\Delta+E)/e when a peak is present in the DOS of the scattering area at energy E . Consequently, spectroscopic features observed at eV_{\rm{sd}}=\pm\Delta/e=\pm 175~\mu\rm{eV} correspond to DOS peaks at zero energy ( E=0 ) in the scattering area. When half of a superconducting flux quantum ( \Phi_{0}=2h/e ) induced by the left flux line penetrates loop L, the superconducting phase difference \phi_{\rm{L}} is tuned to \pi and the ABS energy approaches zero energy as for a highly transparent two-terminal JJ [ 2 ] . Notably, the dispersion shows an additional modulation with a larger periodicity in I_{\rm{L}} , which is evident in the map shown in Fig. 1 (d) measured at constant V_{\rm{sd}}=-\Delta/e by varying I_{\rm{L}} and I_{\rm{R}} . Here, we observe a slower modulation along the diagonal direction I_{\rm{L}}=I_{\rm{R}} , caused by the magnetic flux generated by lines L and R impinging through the middle loop. A useful way to navigate within the 3D flux space is to consider the cubic unit cell defined by the three independent magnetic fluxes \Phi_{\rm{L}} , \Phi_{\rm{M}} , \Phi_{\rm{R}} , as schematically illustrated in the insets of Fig. 1 (d,e). Within this framework, the I_{\rm{L}} - I_{\rm{R}} map follows a tilted plane sketched in gray, whose orientation is defined by the 3\times 3 mutual inductance matrix M : \Phi=\begin{pmatrix}\Phi_{\rm{L}}\\ \Phi_{\rm{M}}\\ \Phi_{\rm{R}}\end{pmatrix}=M\begin{pmatrix}I_{\rm{L}}\\ I_{\rm{M}}\\ I_{\rm{R}}\end{pmatrix}+\Phi_{(0,0,0)}, (1) where \Phi_{(0,0,0)} is an offset defining the corner of a unit cell. In order to cut the unit cell in a controlled way, we compensate for the cross-coupling between loops and flux lines by simultaneously setting the three currents I_{\rm{L}} , I_{\rm{M}} and I_{\rm{R}} needed to reach a flux point \Phi , according to Eq. ( 1 ). Figure 1 (e) shows the differential conductance map measured in this way by sweeping the fluxes \Phi_{\rm{L}} and \Phi_{\rm{R}} and keeping \Phi_{\rm{M}}=0 . As a result, a periodic square net is obtained, demonstrating independent flux control over the three loops. More details about the current-to-flux remapping and how to extract the mutual inductance matrix elements are provided in Ref. [ 75 ] , Sec. II. Figure 3: (a) Slicing the cubic unit cell (see schematic) by measuring \Phi_{\rm{L}} - \Phi_{\rm{R}} maps at different values of \Phi_{\rm{M}} , with voltage bias V_{\rm{sd}} kept fixed at -175~\mu\rm{V} . (b,c) Cuts measured as a function of V_{\rm{sd}} along the (101) and (10\overline{1}) direction (see gray arrows in (a) at \Phi_{M}=0.5~\Phi_{0} ). In the schematics on the right of the plots, the currents flowing in the two outer loops of device are sketched, revealing that mutual inductive effects on the center loop are expected along \Phi_{101} and not along \Phi_{10\overline{1}} . III ABS energy dispersion in the 3D flux space Having established a measurement protocol that allows for independent flux control, we systematically map the ABS energy spectrum in the 3D flux space. Owing to the periodicity of the spectrum, we can restrict our investigation to a single 3D flux unit cell. We start with a simple case where we sweep the flux threading a loop while keeping the other two fluxes at zero, as shown in Fig. 2 (a) for varying \Phi_{\rm{L}} . The energy spectrum reveals the dispersion of a highly transparent ABS, that we identify as the mode formed between the terminal pair T1-T2. The large but finite transparency is expected to prevent the band from reaching zero energy, forming a minigap between the electron- and hole-like branches of the ABS spectrum [ 3 ] . However, this small energy gap is not experimentally revolved due to the sizable spectral broadening, estimated to be 15~\mu eV (see Ref. [ 75 ] , Sec. III). Notably, the ABS dispersion does not reach the superconducting gap edge, but it is reduced to \sim 0.4\Delta/e , potentially due to the repulsion with the lower-transmission states visible at V_{\rm{sd}}<-0.3 mV. Figure 2 (b) shows the dispersion of the ABSs formed between T2-T3, which are tuned by sweeping \Phi_{\rm{M}} . Similar to the previous case, the spectrum is dominated by a brighter highly transparent mode having the same reduced dispersion, and a low-transmission manifold of states at lower energies. The dispersion of the modes formed between T3 and T4 (shown in Ref. [ 75 ] , Sec. III) is qualitatively equivalent to the one along \Phi_{\rm{L}} . In the following, we focus on the three high-transmission ABSs that we label |\rm{L}\rangle , |\rm{M}\rangle and |\rm{R}\rangle as shown in Fig. 1 (b). To have an overview on how such states disperse in the flux space, we map the differential conductance G fixed at V_{\rm{sd}}=-\Delta/e-25~\mu\rm{V} within the whole 3D unit cell. In Fig. 2 (c) we plot G for values larger than 0.095\times 2e^{2}/h , i.e., where the DOS is nonzero. At this V_{\rm{sd}} value, the Andreev dispersions are cut twice around \Phi_{i}=0.5~\Phi_{0} , forming pairs of parallel conductance lines along the cube faces. At the center of each face, the ABSs do not cross each other, but they rather interact opening avoided crossings, as highlighted in the face maps of Fig. 2 (d,e). These avoided crossings are spectral signatures characteristic of bi-Andreev molecular states formed by the hybridization of two ABSs, which have recently been observed in three-terminal devices [ 66 ] . Notably, the four-terminal device presented here acts as an effective three-terminal system on the cube faces, i.e., when one flux is kept fixed to zero. To better define directions in the 3D flux space, we introduce a crystallographic-like notation, where the three flux axes ( \Phi_{\rm{L}} , \Phi_{\rm{M}} and \Phi_{\rm{R}} ) are denoted as (100) , (010) , and (001) , respectively. The hybridization lifts the degeneracy of the original two-terminal ABSs, splitting the dispersion in two bands as observed in the spectra in Figs. 2 (f) and 2 (g), measured along the directions (110) and (101) (white arrows in Figs. 2 (d) and 2 (e), respectively). A larger splitting is observed along the (110) direction (f) compared to (101) cut (g), indicating a stronger coupling between the nearest neighbor ABSs \mid L \rangle and \mid M \rangle . A weaker hybridization is instead expected between \mid L \rangle and \mid R \rangle , due to their smaller wavefunction overlap. In Fig. S4 [ 75 ] , additional tunneling spectra show that the energy splitting is much smaller along the perpendicular directions \Phi_{10\overline{1}} and \Phi_{1\overline{1}0} , as expected when the interacting ABSs have opposite phases [ 43 , 47 , 66 ] . Thus, our observations reveal a significant hybridization between all three ABSs, which couple in pairs to form bi-Andreev molecules on each face of the cubic unit cell. Figure 4: (a) Schematic of the model comprising three two-level quantum dots (blue, red, green) coupled to each other and to four superconducting terminals (T1-T4). Andreev bound states (ABSs) form in the quantum dots and have energies dispersing as a function of the superconducting phase differences \phi_{i} . The model parameters (see definitions in the text) are: \epsilon=0.005\Delta , \Gamma=0.37\Delta and t=0.12\Delta . (b) Phase-effective flux relation including mutual inductive coupling between the loops when the other phases \phi_{\rm{L}} and \phi_{\rm{R}} are kept constant to zero (dashed gray line) and when they are equal to \phi_{\rm{M}} (solid purple line). (c) Simulated energy spectrum along \Phi_{\rm{L}} , with \Phi_{\rm{M}}=\Phi_{\rm{R}}=0 . The energy axis E is shifted by \Delta=175~\mu\rm{eV} to align to the experimental data. (d) Simulated flux-flux map at E=0 corresponding to the measurement shown in Fig. 3 (a) for \Phi_{\rm{M}}=0.5~\Phi_{0} . (e) Conceptual illustration of the hybridization among three degenerate energy levels resulting in bonding, nonbonding and antibonding states separated by energy gaps, as in a tri-atomic molecule. (f) Simulated energy spectra along \Phi_{10\overline{1}} for different \Phi_{\rm{M}} values, showing that hybridization between ABSs opens energy gaps at the crossing points between different levels. The colored squares refer to the arrows in Fig. 5 (f). IV Exploring the center of the unit cell The device configurations discussed so far reproduce the behavior of either two-terminal devices (when two phase differences are kept to zero, i.e., along the unit cell edges) or three-terminal ones (along the unit cell faces, where only one phase difference is kept to zero). Inside the unit cell, all the phase differences are nonzero, leading to a more complex ABS spectrum achievable only with four or more leads. To explore such configurations, we slice the cubic cell from the \Phi_{\rm{L}} - \Phi_{\rm{R}} face to the center along parallel planes measured at different \Phi_{\rm{M}} values and at V_{\rm{sd}}=-\Delta/e , as shown in Fig. 3 (a). At this V_{\rm{sd}} value, we probe the energy spectrum near the maxima of the ABSs, resulting in one conductance line per state. By increasing \Phi_{\rm{M}} , the conductance becomes asymmetric around the center of the plot and develops a maximum at \Phi_{\rm{L}}=\Phi_{\rm{R}}=0.7~\Phi_{0} . More detailed discussions of this figure are presented in Ref. [ 75 ] , Sec. V. Moving \Phi_{\rm{M}} further to 0.5~\Phi_{0} , the map recovers its inversion symmetry, featuring two lobes of low conductance around the center. In this configuration, one would expect to measure constant conductance across the entire plane, since |\rm{M}\rangle is fixed at its energy maximum. Instead, the energy cut along \Phi_{101} [Fig. 3 (b)] reveals that the state |\rm{M}\rangle has a relatively weak dispersion just below V_{\rm{sd}}=-0.175 mV. We explain the weak influence of \Phi_{\rm{L,R}} on this ABSs in terms of mutual inductive coupling between loops. When \Phi_{\rm{L}} and \Phi_{\rm{R}} are swept in-phase ( \Phi_{101} ), two opposite currents flow along the two long sides of loop M, as shown in Fig. 3 (c). These currents induce two parallel flux contributions to \Phi_{\rm{M}} , providing an additional phase difference between T2 and T3. When \Phi_{\rm{L}} and \Phi_{\rm{R}} are swept with opposite sign, i.e., along \Phi_{10\overline{1}} , the two currents flow in the same direction as sketched in Fig. 3 (e), and the two induced fluxes cancel each other out. Indeed, the spectrum measured along this direction [Fig. 3 (d)] shows that |\rm{M}\rangle forms a flat band independent of the other two fluxes just below V_{\rm{sd}}=-0.175 mV. In Fig. 3 (d), we also observe two dispersive bands representing the |\rm{L}\rangle-|\rm{R}\rangle hybridized states and having their maxima at \Phi_{10\overline{1}}\sim 0.5\Phi_{0} . Here, they interact with the |\rm{M}\rangle -derived flat band which significantly decreases its energy to V_{\rm{sd}}\sim-0.2 mV. This indicates that an additional gap between the electron- and hole-like branches of the overall ABS spectrum is opened in addition to the minigap formed by the finite junction transparency. In the following, we show that this spectral feature marks the hybridization among three two-terminal ABSs occurring when they are tuned to the same energy, i.e., the formation of a tri-Andreev molecule. Figure 5: Experimental (a) and simulated (b) zero-energy plane (at fixed V_{\rm{sd}}=-175~\mu\rm{V} or E=0 , respectively) along a diagonal of the cubic unit cell (top-left schematic). Experimental (c) and simulated (d) energy spectrum along the diagonal direction of (a,b), i.e., along the flux direction \Phi_{111} [gray arrow in (a)]. (e,f) As in (a,b), but along another diagonal plane of the cubic unit cell \Phi_{10\overline{1}} - \Phi_{\rm{M}} (bottom-left schematic). Experimental (g) and simulated (h) energy spectrum along the flux direction \Phi_{1\overline{1}\overline{1}} [gray arrow in (e)] V Theoretical model To better understand how the hybridization between ABSs reshapes the complex Andreev band structure observed in the previous paragraph, we develop a theoretical model schematically represented in Fig. 4 (a). The model features four superconducting leads coupled to a normal scattering region described by means of three coupled quantum dots. The superconducting phase differences \phi_{\rm{L,M,R}} (or equivalently magnetic fluxes \Phi_{\rm L,M,R}=\Phi_{0}\phi_{\rm L,M,R}/2\pi ) are defined between the leads using the same convention introduced in Fig. 1 (a). Each dot contains two noninteracting spin-degenerate levels of energy \epsilon^{\pm}_{i} ( i=1,2,3 ), and is coupled to the neighboring leads with a coupling strength characterized by the parameter \Gamma , as well as to the other dots, described by the parameter t . All dot-lead couplings and interdot couplings are assumed to be equal. We note that the choice of a quantum dot network provides a convenient description of a normal region hosting discrete ABSs, with the degree of hybridization among them determined by t . No Coulomb interaction or charging energy are introduced in the system, as the dots are strongly coupled to the leads. To compare with the experimental data, we compute the total DOS projected onto the three dots as a function of the energy E and the three superconducting phase differences \phi_{\rm{L,M,R}} (fluxes \Phi_{\rm L,M,R} ), indicated in Fig. 4 (a). Furthermore, to include the effect of the mutual inductive coupling between the superconducting loops, which causes a nonlinear cross-dependence between the phases (magnetic fluxes), we remap them to effective phases \phi_{i}^{\rm eff} (effective magnetic fluxes \Phi_{i}^{\rm eff} ) using the relations: \displaystyle\phi_{\rm{L,R}}^{\rm eff}=\left[\phi_{\rm{L,R}}+\alpha f\left(\phi_{\rm{M}}\right)\right] \displaystyle\phi_{\rm{M}}^{\rm eff}=\left\{\phi_{\rm{M}}+\alpha\left[f(\phi_{\rm{L}})+f(\phi_{\rm{R}})\right]\right\}, where \phi_{i}^{\rm eff}/2\pi=\Phi_{i}^{\rm eff}/\Phi_{0} for i={\rm L,M,R} . Here, f(\phi)=-2\partial_{\phi}\left[1-\tau\sin^{2}(\phi/2)\right]^{-1/2} , \alpha=0.2 is the strength of the mutual coupling and \tau=0.9 introduces nonsinusoidal character to f(\phi) . Additional details on the model are discussed in Ref. [ 75 ] , Secs. VI and VII. Figure 4 (b) illustrates the influence of \phi_{\rm{L,R}} on \Phi_{\rm{M}}^{\rm eff}(\phi_{\rm{M}}) as a consequence of the mutual coupling, highlighting a nonlinear behavior when \phi_{\rm{L,R}} are varied together with \phi_{\rm{M}} . The simulated DOS as a function of energy and of \Phi_{\rm{L}} while \Phi_{\rm{M}}=\Phi_{\rm{R}}=0 is shown in Fig. 4 (c). The two energy levels in each of the three dots give rise to two distinct manifolds comprising three modes each. As expected, only one state per manifold has a significant energy dispersion, while the others remain mostly constant in energy. Similar to the tunneling spectra in Fig. 2 (a,b), the simulated band structure exhibits one resonance that approaches E=0 at \Phi_{\rm{L}}=0.5\Phi_{0} , forming a cusp consistent with an isolated high-transmission ABS. Figure 4 (d) displays the simulated \Phi_{\rm{L}} - \Phi_{\rm{R}} plane for constant flux \Phi_{\rm M}=0.5\Phi_{0} and constant energy E=0 , which corresponds to the measurement shown in Fig. 3 (a). The presence of two lobes of minimum DOS around \Phi_{\rm{L}}=\Phi_{\rm{R}}=0.5\Phi_{0} is captured by the model as a result of the mutual coupling between the phases. To focus on the effects of ABS hybridization, we consider the \Phi_{10\overline{1}} direction (in which mutual inductance effects are negligible) and simulate the Andreev spectra as a function of E for different values of \Phi_{\rm{M}} [Fig. 4 (f)]. At \Phi_{\rm{M}}=0 , \mid M \rangle remains flat at high energy, while both \mid R \rangle and \mid L \rangle form dispersing bands which are nearly degenerate in energy. By increasing \Phi_{\rm{M}} , the flat state approaches zero energy and forms avoided crossings with the dispersing states. These three nondegenerate ABSs resemble the energy levels of a tri-atomic molecule, where three molecular orbitals (bonding, nonbonding and antibonding) are split in energy, as conceptually depicted in Fig. 4 (e). At \Phi_{\rm{M}}=0.5\Phi_{0} , the state \mid M \rangle reaches the energy closest to zero and remains constant around that energy. Here, its dispersion bends downwards from zero energy as observed in the corresponding measurement in Fig. 3 (c). These results support the interpretation that our device hosts three ABSs hybridizing among each other, forming a tri-Andreev molecule delocalized over the four superconducting terminals. Figure 6: (a) ABS spectra extracted from the density of states as a function of the phase differences \phi_{\rm L} and \phi_{\rm R} for fixed \phi_{\rm M}=0.21\pi . The energy gap between the highest band with E<0 and the lowest band with E>0 vanishes in one point, where a Weyl node forms. (b) Enlargement of (a) around zero energy, highlighting the Weyl node. (c) As in (b), but for \phi_{\rm M}=0.58\pi . A topological gap is present between the bands. (d) As in (b), but for \phi_{\rm M}=0.96\pi , showing a second Weyl node with opposite topological charge where the gap closes. The parameters are the same as in Figs. 4 and 5 . (e) Positions of the positively (blue) and negatively (red) charged Weyl nodes in the 3D phase space. (f) Chern number C^{\rm{LR}} calculated as a function of \phi_{\rm M} with \tilde{\epsilon}\approx 0.01 and \tilde{t}\approx 0.32 , showing extensive topologically nontrivial regions where |C^{\rm{LR}}|=1 (green shading). (g) Topological phase diagram displaying the \phi_{\rm M} range of the topological region, \rho , as a function of the model parameters \tilde{\epsilon} and \tilde{t} . The diagram shows the robustness of the topological phase, that is present for \rho>0 . The black dot indicates the parametric point corresponding to the model used in Figs. 4 and 5 . VI tri-Andreev molecule Supported by our theoretical model, we investigate more complex spectral features measured along the diagonals of the cubic unit cell, when all three fluxes are varied. Figure 5 (a) shows a map measured at V_{\rm{sd}}=-\Delta/e along the \Phi_{101} - \Phi_{M} plane sketched in the schematic on the left. As previously discussed in Sec. III, \mid L \rangle and \mid R \rangle are already hybridized along the \Phi_{101} direction, as also visible by the splitting of the conductance line at \Phi_{\rm{M}}=0 and \Phi_{101}=0.5\Phi_{0} . When \Phi_{\rm{M}} is tuned to 0.5\Phi_{0} , these hybridized states mix with \mid M \rangle forming an avoided crossing which is well reproduced by the simulation displayed in Fig. 5 (b). Taking an energy cut along the (1,1,1) direction [gray arrow in (a)], we observe how the energy spectrum is affected by the hybridization, as shown in Fig. 5 (c). Here, the bands split in energy forming an M-shape dispersion close to zero energy. Notably, such effect is much stronger compared to the one observed for the bi-Andreev molecules in Fig. 2 (f,g). By simulating the same energy spectrum [Fig. 5 (d)], we reproduce a similar M-shape of the topmost band and reveal the energy splitting between the three bands representing the bonding, nonbonding and antibonding molecular states illustrated in Fig. 4 (e). As discussed in Sec. III for the bi-Andreev molecules, the energy splitting induced by the hybridization is strongly anisotropic in the phase space. Figure 5 (e) shows the conductance map along the \Phi_{10\overline{1}} - \Phi_{M} plane sketched in the schematic on the left. Here, the dispersions of \mid L \rangle and \mid R \rangle overlap with each other at \Phi_{\rm{M}}=0 , but split when \Phi_{\rm{M}} is swept towards 0.5~\Phi_{0} . Notably, the splitting appears larger along the (11\bar{1}) direction compared to (1\bar{1}\bar{1}) , since \mid L \rangle has a slightly larger transparency than \mid R \rangle (see [ 75 ] , Sec. IV). Also, the horizontal conductance resonance representing the maximum of \mid M \rangle significantly decreases in intensity approaching the center of the map. The simulation in Fig. 5 (f) reproduces the opening of a low conductance region at the crossing point, which ultimately derives from the splitting at E=0 observed in the simulated energy spectra in Fig. 4 (f). Indeed, those dispersions represent horizontal energy-dependent cuts of Fig. 5 (f), at \Phi_{\rm{M}} values indicated by the colored arrows. Taking an energy-dependent cut along the direction (1\bar{1}\bar{1}) [Fig. 5 (g)], we observe a sizable splitting into two bands, but smaller than the splitting observed along the (111) direction. The overall hybridization strength is indeed reduced along the direction (1\bar{1}\bar{1}) since two phases always have opposite values. Therefore, two bands are expected to remain nearly degenerate, as reproduced in the simulated spectrum in Fig. 5 (h). In summary, the dispersion of a tri-Andreev molecule is characterized by an anisotropic energy splitting, larger along the (111) direction and weaker along its perpendicular direction (1\bar{1}\bar{1}) . This is directly reflected into the different shapes of the avoided crossings observed at the center of the cube diagonal conductance maps of Fig. 5 (a,e). The spectral signatures of a tri-Andreev molecule discussed here are also observed in a second device (see [ 75 ] , Sec. VIII). VII Topological Andreev bands The hybridization between the three ABSs in our device leads to the formation of molecular-like states whose energy depends on all three superconducting phase differences. This property is one of the requirements for the formation of Weyl nodes, which would appear as zero-energy crossing points with linear dispersion as a function of all the three phases. In the following, we analyze the simulated Andreev band structure matching our experimental results. In Fig. 6 (a), we show the ABSs spectrum extracted from the maxima of the density of states for \phi_{\rm M}=0.21\pi . Here, the two particle-hole symmetric bands closest to the Fermi level form a zero energy crossing at \phi_{\rm L}=\phi_{\rm R}=0.89\pi as highlighted in the zoomed-in plot in Fig. 6 (b). Increasing \phi_{\rm M} , we obtain gapped states [see Fig. 6 (c)] with a small energy gap. The gap is closed again at \phi_{\rm M}=0.96\pi , where another zero-energy crossing appears at \phi_{\rm L}\sim\phi_{\rm R}\sim 0.6\pi [Fig. 6 (d)]. Within the whole cubic unit cell we find four zero-energy crossings appearing in two pairs (red and blue points in Fig. 6 (e)), which are related by time-reversal symmetry. As shown in Ref. [ 76 , 77 , 78 ] , the full information about the topology of the ABS spectrum is encoded in the topological Hamiltonian, H_{\rm top} , given by the inverse of the central Green’s functions G_{\rm C} evaluated at zero energy, i.e., H_{\rm top}=-G_{\rm C}^{-1}(E=0) . H_{\rm top} depends on the dimensionless parameters \tilde{\epsilon}=\epsilon/\Gamma and \tilde{t}=t/\Gamma [ 24 , 27 ] (see more details in Ref. [ 75 ] , Sec. VII). The values of t and \epsilon used for the DOS simulations matching our measurements correspond to \tilde{\epsilon}\approx 0.01 and \tilde{t}\approx 0.32 . By diagonalizing H_{\rm top} , we obtain the eigenvectors \ket{v_{i}} and the corresponding spin-degenerate eigenenergies. To establish the topological properties of the ABSs, we compute the topological invariant following the numerical method in Ref. [ 79 ] . The Chern number is given by C_{i}^{\rm LR}(\phi_{\rm M})=\frac{1}{2\pi}\int_{0}^{2\pi}\int_{0}^{2\pi}B_{i}^{\rm LR}(\phi_{\rm L},\phi_{\rm R},\phi_{\rm M})\,d\phi_{\rm L}d\phi_{\rm R}, (2) with the total Chern number C^{\rm LR}=\sum_{i=1,2,3}C_{i}^{\rm LR} as the sum over all occupied bands and B_{i}^{\rm LR}\equiv-2{\rm Im}\braket{\partial_{\phi_{\rm L}}v_{i}|\partial_{\phi_{\rm R}}v_{i}} represents the Berry curvature calculated at fixed \phi_{\rm M} . Figure 6 (f) shows two distinct nontrivial topological phases having Chern number equal to \pm 1 within extensive \phi_{\rm M} intervals indicated by the green areas. Therefore, the zero energy crossing points in (e) are positively (blue) and negatively (red) charged Weyl nodes appearing in the energy spectra (b,d) as the states cross the Fermi level. The eigenenergies of the topological Hamiltonian for the same parameters as in Fig. 6 are shown in Fig. S7 [ 75 ] . The eigenenergies exhibit zero-energy states exactly when one local maximum of the DOS is at zero energy, as expected. Notably, the spectrum shown in Fig. 6 (c) represents a topological insulating phase with a small energy gap. Since such features are masked by the relatively large spectral broadening in our measurements, we cannot experimentally confirm the predicted topological phase transitions. To evaluate the robustness of the topological regime in our simulations, we calculate the extension \rho of the region in \phi_{\rm M} where the Chern number differs from zero [green areas in (f)] as a function of the two key parameters of our model, \tilde{\epsilon} and \tilde{t} . The phase diagram in Fig. 6 (g) shows a first topological transition when the hybridization \tilde{t} exceeds \tilde{\epsilon} . Intuitively, the hybridization has to be large enough to push one state through zero energy, a mechanism analogous to the band inversion driven by spin-orbit coupling occurring in topological insulators [ 80 ] . By further increasing \tilde{t} , the two opposite charged Weyl nodes get closer to each other gradually reducing \rho to zero. At this point, the oppositely charged Weyl nodes annihilate with each other completely suppressing the topological phase. The parameters used for simulating our measurements corresponds to \tilde{\epsilon}=0.01 and \tilde{t}=0.32 , which places our system well within the calculated topological region as indicated by the black dot in Fig. 6 (g). VIII Discussion and conclusions In this work, we studied the hybridization between ABSs in a 4TJJ and demonstrated the formation of a tri-Andreev molecule. This state, whose energy is controlled by three superconducting phase differences, is expected to support topological Andreev bands. According to our model, Weyl nodes emerge when the hybridization shifts at least one ABS through zero energy, inducing an inversion between electron- and hole-like bands in certain regions of the phase space. By varying \phi_{\rm{M}} , the system undergoes four topological transitions marked by Weyl nodes as shown in Fig. 6 (f). The Andreev bands as a function of \phi_{\rm{L}} - \phi_{\rm{R}} have an energy gap ranging between 0 (at the Weyl nodes) and \sim 7~\mu\rm{eV} , depending on \phi_{\rm{M}} . The resolution of tunneling spectroscopy, approximately 15~\mu\rm{eV} , prevents us from experimentally resolving any gapped states in the low-energy spectrum. Thus, the spectral detection of Weyl nodes would be facilitated by a larger minigap, which could be obtained through an increase of the superconducting gap \Delta or by a enhanced device tunability. For example, a larger minigap would be obtained by making the coupling parameters [see \Gamma in Fig. 4 (a)] asymmetric or by increasing \epsilon , while keeping a sufficiently large hybridization strength. The phase diagram shown in Fig. 6 (g) provides a useful guideline for engineering these parameters while preserving the topological properties of the device. The large transparency and hybridization strength observed in our 4TJJ already fall well into the stability range of the topological state. Experimental techniques with higher energy resolution would be highly favorable for confirming the presence of Weyl nodes in the ABS spectrum, even in systems with such a large transparency. Microwave spectroscopy, in particular, offers sub- \mu\rm{V} resolution, making it well suited for this purpose. Furthermore, hybrid InAs/Al heterostructures are readily integrated in circuit QED architectures [ 81 , 82 , 14 , 83 , 11 , 84 ] , offering a tangible prospect for studies of topological Andreev band structure. The employment of polarized microwave radiation can also make this technique sensitive to the Berry curvature [ 24 ] . In summary, we realized a phase-controlled 4TJJ and studied its energy spectrum using tunneling spectroscopy. The measurement protocol based on independent flux control developed here allows the systematic exploration of the Andreev spectrum in the 3D phase space. We identified the spectral signatures of a tri-Andreev molecule resulting from the simultaneous hybridization of three ABSs. A numerical model reproduced the key experimental observation, and suggested that the current generation of devices already hosts topological Andreev bands. In the light of our results, phase-tunable MTJJs offer new opportunities for studying topological phases in high-dimensional synthetic band structures and developing novel superconducting quantum circuits [ 85 ] . IX Data availability The data that support the findings of this article will be available on Zenodo. X