## Introduction

The study of topological phases has been one of the central topics in condensed matter physics in the past decade1,2. Among the rich class of topological phases, topological Weyl and Dirac semimetals, which are characterized by discrete gapless nodes in the bulk spectra, have attracted wide attention3,4,5. Referred to as Weyl or Dirac points, these gapless nodes in the bulk can be connected by open strings formed by topologically protected surface states on the boundaries, called Fermi-arc states6. Due to their anomalous electromagnetic responses7,8,9,10,11,12,13,14, as well as the interesting interplay between bulk and surface Fermi arcs15,16, topological semimetals have been studied intensively in recent years5. In parallel to the rapid exploration of Dirac properties in topological semimetals, the study of their superconducting states has become an important topic aiming for possible unconventional superconductivity17,18,19,20. It was predicted that Fermi-arc states in a Dirac semimetal such as Cd3As2 could be used to realize Majorana fermions20,21. This is due to the fact that the Fermi-arc states in Cd3As2 stem from helical edge states of quantum spin Hall insulators embedded in the time-reversal invariant semimetal22. By proximitizing the Fermi-arc states with s-wave superconductors, Majorana flat band can be possibly created at the interface of a π-Josephson junction20,23. While the signature of superconductivity has been found in Cd3As224,25,26, the role of Fermi-arc states in these pairing phases remains unclear.

Despite the appealing proposals for creating Majorana fermions, the proposed scheme relies on the assumption20,21 that the Fermi arcs can acquire hard superconducting gaps from the parent superconductor. In topological insulator/superconductor (TI/SC) junctions, it is generally believed that the induced superconducting gap in the surface states is small due to high interfacial barriers and Fermi surface mismatch27,28,29. In a Dirac semimetal Cd3As2, the unavoidable coexistence of low-energy bulk excitations and Fermi-arc states5 causes further complications to distinguish the roles of bulk states and surface states in the superconducting state. Therefore, a prerequisite experimental evidence of strong superconducting pairing on the surface is clearly desirable. In particular, due to the highly non-uniform spatial distribution of surface states in Cd3As2, the superconducting proximity on the surface and bulk may exhibit distinctive anisotropic signatures.

Here, we report the observation of strong proximity-induced superconductivity on the surfaces of Cd3As2 based on two superconducting characteristics: the proximity-induced superconducting gap and the spatial distribution of supercurrents on surfaces. Using four-terminal transport measurements on Nb/Cd3As2 hybrid structures, we observe a pronounced proximity-induced gap (Δs) on the surfaces, with its size comparable to the parent superconducting gap (ΔNb) of Nb. The sizable surface proximity gap manifests itself as a flat conductance plateau in differential conductance (dI/dV) spectra. In contrast, the proximity gap in bulk states is a few times smaller (Δb ~ 0.14Δs) and featured by a zero-bias broad peak (ZBBP) in the dI/dV spectra. The relation between the observed conductance plateau/ZBBP and proximity-induced gaps in surface/bulk states are further confirmed by our theoretical calculations with a qualitatively good agreement. Upon changing surface-bulk contributions in Cd3As2 samples with different thicknesses, the conductance plateau/ZBBP exhibit evolutionary behaviors consistent with their surface/bulk origins. Furthermore, using superconducting quantum interference (SQI) measurements on Nb/Cd3As2/Nb Josephson junctions (JJs), we observe a SQUID pattern in the surface-state-dominated Cd3As2 JJs, which indicates that the supercurrent density is predominantly on top and bottom surfaces.

## Results

### Proximity-induced superconductivity in Nb/Cd3As2

The proximity-induced superconductivity in Nb/Cd3As2 hybrid structure is displayed in Fig. 1a (device #01). The 250 nm-thick Nb film, colored in green in the false-color scanning electron microscopy (SEM) image, is deposited on top of a ~200 nm-thick Cd3As2 nanoplate (blue color). A four-terminal measurement across the interface was performed to detect the interface resistance, denoted as RInterface (see Supplementary Note 1, Supplementary Fig. 1 for sample characterizations and Supplementary Note 2, Supplementary Figs. 2 and 3 for measurement details).

Temperature-dependent resistance (RT) curves at T ≥ 2.0 K across the junction are shown in Fig. 1b (black curve) with three drops at Tc1 = 8.4 K, Tc2 ~ 8.2 K, and Tc3 ~ 4.0 K corresponding to Nb superconducting for Tc1 and proximity-induced superconductivity for Tc2 and Tc3, respectively, since Nb has been in zero-resistance state at 8.3 K (Fig. 1b inset, also see details in Supplementary Note 3 and Supplementary Fig. 4). Then, we carry out the four-terminal dI/dV measurements to show the induced superconductivity in Cd3As2 in Fig. 1b–d. A bias-independent conductance plateau (BICP) appears in the vicinity of ±1.20 mV (Δs) with a ZBBP around ±0.19 mV (Δb) and an above-gap dip at high bias (Δd). For temperatures lower than 4 K, a very sharp zero-bias conductance peak (ZBCP) emerges which is superimposed on the broad peak (Fig. 1c). Control experiments exclude the effect from Nb or Cd3As2 (Supplementary Note 3 and Supplementary Figs. 5 and 6). Evidently, the ZBBP and ZBCP correspond to the second transition (Tc2) and third transition (Tc3) in the RT curve while the first transition (Tc1) mostly corresponds to the conductance plateau with a small resistance drop by the above-gap dip in Fig. 1b. The observation of the superconducting proximity effect and conductance plateau suggests that the interface is of high transparency. Fits to the broad peaks in black lines by standard Blonder–Tinkham–Klapwijk (BTK) theory30 in Fig. 1c, d confirm the proximity-induced superconductivity in Cd3As2. The BTK fittings yield important parameters of Δb = 0.19 meV and Z = 0 (Supplementary Note 4 and Supplementary Figs. 7 and 8). Obviously, the amplitudes of BICP, ZBBP and ZBCP decrease with increasing temperature and perpendicular magnetic field. Figure 1e, f, respectively, show temperature- and magnetic field-dependent gaps which are fitted to the Bardeen–Cooper–Schrieffer (BCS) theory (Supplementary Note 5). Δs is extracted to be 1.34 meV at 0 T, comparable to the superconducting gap of Nb estimated from the BCS theory31 Δ = 1.76kBTc ~ 1.30 meV, which further verifies the proximity effect in the Nb/Cd3As2 system and s-wave pairing in Cd3As2.

The above-gap dip Δd and ZBCP are, respectively, attributed to depairing effect near the interface and the proximity-induced superconductivity in another band of Cd3As2 such as the band projected onto [112] direction (Supplementary Note 6 and 7 and Supplementary Figs. 911).

Next, we probe the relation between BICP, ZBBP and proximity-induced superconductivity. For voltage bias V lower than the Nb superconducting gap, the transport process is dominated by Andreev reflection (AR), in which an incoming electron from the Dirac semimetal is converted to a reflected hole at the Nb/Cd3As2 interface. Thus, the presence of both ZBBP and BICP for sub-gap voltage bias suggests two types of Andreev reflection channels in Cd3As2: one exists only within a narrow energy window given by the width of the broad peak, while the other has a much wider energy range measured by the width of the conductance plateau. Since the energy window for Andreev reflections is directly related to the interfacial coupling between the channels of Cd3As2 and Nb, the contrasting features of the broad peak and plateau also suggest very different proximity effects in two types of channels. By placing Cd3As2 next to the superconducting Nb, the superconducting proximity effect is expected to occur in both the bulk and the surface region, as shown in Fig. 2a. Importantly, since the wave functions of the surface states are predominantly localized near the interface, the surface states can strongly couple to Nb than the bulk. As a result, it is reasonable to expect that the proximity effect in the surface states can be much stronger than that in the bulk states. Besides, when the Fermi level is close to the Dirac points, AR amplitude for bulk-state channels is expected to be small and the Cooper pair wave function would decay rapidly in the bulk region as shown in Fig. 2b.

To further distinguish the proximity effect on surface and bulk channels, we acquired the continually evoluting features of BICP and ZBBP by tuning the thickness to change the specific surface area with bulk/surface states proportion in Cd3As2 (Fig. 2c–f and Supplementary Note 8, Supplementary Figs. 1215). We measured dI/dV spectra of Cd3As2 hybrid structure with different thickness at 2 K as shown in Fig. 2c–f. Devices # 01 and 02 with a thickness of ~200 nm both show BICP and ZBBP while the plateau remains pronounced while the ZBCP becomes strongly suppressed and hardly observable in thinner samples #03 and #04 (<150 nm), as shown in Fig. 2e, f, indicating that the surface channels dominate the proximity effect in thin samples. Thus, the channels responsible for the plateau are indeed from Cd3As2 surface states while the broad peak results from low-lying bulk channels in Cd3As2.

Next, we discuss the possible origin of surface superconducting channels. In device #01, the bulk states and Fermi-arc surface states can coexist (Table 1). As reported previously16,32, the Fermi surface property in Cd3As2 is thickness-dependent. In particular, we note that in samples with thickness >150 nm, the Fermi level is generically well above the Dirac points. Thus, both bulk states and Fermi-arc states contribute to electronic transport, while in samples thinner than 150 nm, the Fermi level is closer to the Dirac cone and Fermi-arc states dominate. Since both the electronic transport and proximity effects are closely related to the Fermi surface property, we anticipate that the surface channels possibly correspond to Fermi-arc surface states.

Both devices # 01 and 02 have shown the bulk states and Fermi arcs in transport from SdH oscillation measurements. Here, we use the relative amplitude of the surface oscillations compared with the bulk oscillation AS/AB to estimate the surface/bulk channels (S/B) domination. In contrast, thinner samples (# 03 and 04) exhibit low bulk domination. Based on the analysis of the SdH oscillations (Supplementary Note 9, Supplementary Fig. 18 and Supplementary Table 1), the Fermi surface property is confirmed to be dominated by the Fermi-arc states. Therefore, through the different weight of surface-bulk conduction, we experimentally attribute the flat conductance plateau to the superconducting proximity effect possibly in the Fermi-arc states and the broad peak to the relatively weak proximity effect in the bulk states. Besides, the surface and bulk contribution to Andreev reflection is also consistent with the conductance enhancement (Supplementary Note 10 for details).

In the thick limit of Cd3As2 where bulk states dominate, weak Andreev reflection occurs with strong tunneling behavior as a result of insufficient electrons from the bulk states to participate in AR30 (Supplementary Note 8, Supplementary Figs. 1214, device #05, >300 nm thick). At the interface, the AR electron density is proportional to $$\frac{{{\mathrm{d}}I}}{{{\mathrm{d}}V}} \propto G\left( {E_{\mathrm{F}}} \right) \propto \frac{{4e^2}}{h}N\left( {E_{\mathrm{F}}} \right) \propto k_{\mathrm{F}}^2 + k_0^2$$, where k0 is Fermi-arc length and can be estimated by15

$$F_{\mathrm{s}} = E_{\mathrm{F}}k_0/(e\pi v_{\mathrm{F}}),$$
(1)

where Fs is surface frequency. The Fermi velocity vF = kF/m* can be extracted from the bulk SdH oscillations. In device #01, kF ~ 0.0249 Å−1 and k0 ~ 0.8 nm−1 (ref. 32), and kF = 0.032 Å−1 in device #05 with negligible k0, thus the AR electron density in device #01 is much higher than that in device #05. Furthermore, scattering effects between the bulk and surface states can be significant due to disorders, and the surface states are no longer well-localized on the surfaces33 which reduce the effective coupling between the surface and Nb. Besides, for high Fermi levels, there could possibly exist a significant Fermi surface mismatch between the superconductor and the Cd3As2, which could further suppress the interfacial coupling.

We now analyze the proximity-induced superconducting gaps in Cd3As2. Two superconducting gaps against the relative amplitude of AS/AB are presented in Fig. 2g and Table 1. The wave functions of the surface states are localized at the surface, while these of the bulk states are predominantly in the bulk. Effectively, the surface states couple much more strongly to the superconductor than the bulk states. In thin Cd3As2, the transparent interface with enough electron states from surface channels ensures that the proximity effect arises. Hence, the bulk-state superconducting gap Δb drops fast while the surface superconducting gap Δs increases slightly with a larger surface domination. The penetration depth for each surface can be estimated with an assumption that the superconducting gap decreases following34

$${\mathrm{\Delta }}\left( z \right)\sim {\mathrm{\Delta exp}}\left( { - \frac{z}{{\xi _{\mathrm{N}}}}} \right),$$
(2)

where ξN is the superconducting coherence length in Cd3As2 (Table 1) and z is the distance in Cd3As2 from the interface. At the surface regime, $${\mathrm{\Delta }}_{{\mathrm{Nb}}} = {\mathrm{\Delta }}_{\mathrm{s}}{\mathrm{exp}}\left( { - \frac{{z_{\mathrm{s}}}}{{\xi _{\mathrm{N}}}}} \right)$$, we can evaluate the surface penetration depth zs in Cd3As2 of 10–50 nm, which is similar to the previous report for Fermi arcs in Cd3As216 and agrees with other numerical simulations based on the low-energy model of Dirac semimetals35 as an evidence of possible Fermi-arc superconductivity.

### Theoretical calculations on Andreev reflection

Our experimental observations above are further supported by theoretical calculations of the dI/dV spectra for our Nb/Cd3As2 hybrid structure. Using numerical Green’s function method based on a four-band tight-binding model of Cd3As2, we calculate the Andreev reflection amplitude of the Nb/Cd3As2 junction with the parent superconductor Nb modeled by a usual s-wave superconductor with the schematic set-up in Fig. 3a. Details of the Hamiltonian and the Green’s function are presented in the Method section, Supplementary Note 11, Supplementary Figs. 19 and 20 and Supplementary Table 2.

It is worth noting that due to the low-energy Dirac spectrum of Cd3As2, the bulk density of states is expected to increase monotonically as a function of EF measured from the Dirac points located at (0, 0, ±k0). In contrast, Fermi-arc states with different energies are connected on the surface, thus the number of Fermi-arc channels on the junction interface is expected to depend weakly on the Fermi level. Therefore, tuning the chemical potential EF of Cd3As2 in our model allows us to theoretically investigate the roles of surface states and bulk states in the Andreev reflection processes.

First, we consider the case that the location of chemical potential is at the Dirac points (EF = 0 meV), where the bulk density of states is very low and the Andreev reflections driven by bulk channels are negligible. This simulates the scenario in thin samples (devices #03 and #04) where the Fermi surface property is dominated by Fermi-arc states. As shown in Fig. 3b, in this case only a flat conductance plateau is found in our simulations. Being consistent with our experimental observations, the signatures of ZBBP are hardly observable. Then, by tuning the Fermi level to EF = 70 meV from the Dirac points, the bulk density of states gets enhanced and its contribution to Andreev reflections cannot be ignored. This simulates the case of relative thick samples (devices #01 and #02) where the bulk states and surface Fermi arcs have comparable weights in transport. In this scenario, a ZBBP emerges on top of the plateau in the differential conductance spectrum (Fig. 3c). Therefore, our theoretical calculations of the dI/dV spectrum suggest that the Fermi-arc channel can result in the plateau, while the ZBBP arises from Andreev reflections in the bulk-state channel. Here, we briefly note that without Fermi-arc states, a well-defined flat conductance plateau cannot be found in the dI/dV spectrum. This further indicates that the plateau most likely originates from Fermi-arc states. The detailed results are presented in Supplementary Note 11, Supplementary Fig. 21.

To understand the physical mechanisms for the formation of zero-bias peak and flat plateau in the dI/dV curve, it is worth noting that the energy windows for Andreev reflections driven by surface/bulk-state channels are measured by the widths of the plateau/zero-bias peak, respectively. This indicates that the superconducting proximity effect in the Fermi-arc states is much stronger than that in the bulk states. To demonstrate the proximity effects in Fermi-arc and bulk-state channels, we integrate out the superconducting Nb and include its contribution as a self-energy term on the topmost layer of Cd3As2 which interfaces with superconducting Nb. Considering the scenario where the bulk states have a non-negligible density of states, we use the same parameters in obtaining Fig. 3c to calculate the local spectral density on the topmost layer, as shown in Fig. 3d. The color bar indicates the local density of states on a logarithmic scale, with red (blue) colors indicating high (low) density. Clearly, due to the proximity-coupled s-wave superconductor, the topmost layer of Cd3As2 acquires two different spectral gaps, as indicated by white and yellow double arrows. Notably, the larger pairing gap is induced in states predominantly on the surface (higher local density of states on the surface), with twice of the gap size 2Δs corresponding to the width of the conductance plateau due to Fermi arcs, which is comparable to twice value of the parent superconducting gap. In contrast, the smaller induced gap is formed in states with lower density on the surface, and its size 2Δb matches the width of the zero-bias peak due to bulk states. This further confirms that the Fermi-arc states couple strongly to the superconductor, and thus acquire a sizable superconducting gap (~70% of parent gap Δ). In contrast, the states living in the bulk couple relatively weakly to the superconductor, which results in a smaller proximity gap. Therefore, the flat conductance plateau in the dI/dV curve indicates Fermi-arc superconductivity as a possible interpretation, while the zero-bias peak arises from relatively small proximity gap in the bulk states, which opens a narrower window near zero bias for Andreev reflections driven by bulk channels.

### Surface supercurrent in Nb/Cd3As2/Nb Josephson junctions

Having established the proximity-induced surface superconductivity in Nb/Cd3As2, we next build Nb/Cd3As2/Nb Josephson junctions to directly analyze the spatial distribution of the supercurrent. This is enabled by SQI measurements in different directions.

Figure 4a schematically shows the lateral Cd3As2 Josephson junctions with closely spaced superconducting Nb electrodes on the top surface. We choose 120-nm-thick Cd3As2 with 140-nm-thick Nb electrodes to study the surface dominated supercurrent. The inset of Fig. 4b shows an SEM image of device #06. The length and width of the superconducting channel are L = 500 nm and W = 7 μm, respectively. Figure 4b shows the RT curve of the junction with two transitions Tc1 and Tc2 at zero magnetic field. Tc1 ~ 7 K originates from the Nb superconducting transition while Tc2(on) ~ 3 K comes from the superconducting proximity effect. The resistance continues to decrease as the junction cools down and reaches the zero-resistance state below Tc2(off) ~ 1 K. The tail of the resistance drop can be explained in terms of the BKT transition36 as shown in green line by Halperin–Nelson equation37

$$R = R_0{\mathrm{exp}}\left\{ { - 2b\left( {\frac{{T_{{\mathrm{c}}0} - T}}{{T - T_{{\mathrm{BKT}}}}}} \right)^{1/2}} \right\},$$
(3)

where R0 and b are material parameters, which realizes a zero-ohmic-resistance state driven by the binding of vortex-antivortex pairs at the BKT transition temperature TBKT = 1.0 K. On the other hand, junction resistance decreases at the temperature above Tc2, leading to a broadened SC onset, which can be well reproduced by the Aslamazov–Larkin37 fit in red dashed line for the two-dimensional (2D) fluctuation conductivity. These behaviors are consistent with expectations for 2D superconductivity as an evidence for the 2D surface superconductivity in Cd3As2 JJ since the thickness of Cd3As2 is far away from the 2D condition. Figure 4c and inset display the current-voltage (IV) characteristics and differential resistance (dV/dI) of the junction measured at 50 mK, respectively. From the slope of the IV curve, the normal-state resistance Rn ~ 17 Ω is extracted. In the regime |I| < 1 μA, the voltage across the junction and the dV/dI are zero, indicating a robust Josephson effect. The IcRn product gives a characteristic voltage of ~17 μV which is lower than the transition temperature of 1 K (with superconducting gap Δ = 1.76kBTc ~ 150 μV). It indicates that the junction is in the long junction limit38, where the superconducting coherence length ξN is smaller than the effective width between two Nb electrodes, i.e., ξN < Leff (Supplementary Note 12).

The spatial distribution of supercurrent in a Josephson junction can be extracted by SQI measurements, where a magnetic field B perpendicular to the junction induces oscillations in the amplitude of the superconducting current. Measuring the dependence of the critical current $$I_{\mathrm{c}}^{{\mathrm{max}}}$$ on B provides a convenient way to extract the distribution of supercurrent which is widely used to probe edge-mode superconductivity in quantum spin Hall insulator and quantum Hall systems39,40,41. The particular shape of the critical current interference pattern depends on the phase-sensitive summation of the supercurrents traversing the junction. In the case of a symmetric supercurrent distribution, this integral takes the simple form:

$$I_{\mathrm{c}}^{{\mathrm{max}}}\left( B \right) = \left| {\mathop {\smallint }\limits_{ - \infty }^\infty J_{\mathrm{c}}\left( x \right){\mathrm{cos}}\left( {\frac{{2\pi L_{{\mathrm{eff}}}B_{\mathrm{x}}}}{{{\mathrm{\Phi }}_0}}} \right){\mathrm{d}}x} \right|,$$
(4)

where Leff is the effective length of the junction along the direction of the current, accounting for the magnetic flux threading through parts of the superconducting contacts over the London penetration depths. As shown schematically in Fig. 5a, the supercurrent density has an approximately uniform distribution along the z-axis in Cd3As2. Thus, for a magnetic field By applied along y direction, the uniform current density results in a single-slit pattern yields the single-slit Fraunhofer pattern $$\left| {{\mathrm{sin}}\left( {\frac{{\pi L_{{\mathrm{eff}}}BW}}{{{\mathrm{\Phi }}_0}}} \right)/\left( {\frac{{\pi L_{{\mathrm{eff}}}BW}}{{{\mathrm{\Phi }}_0}}} \right)} \right|$$ shown in the right panel. The slight asymmetry in the Fraunhofer pattern can be due to inversion symmetry breaking in the junction which is predicted in topological semimetals42. In contrast, due to the surface-dominated superconductivity, the supercurrent density along the y direction is predominantly localized on the edges as shown in Fig. 5b. When the magnetic field is applied along z direction, the Fraunhofer pattern has a more sinusoidal oscillation characteristic of a SQUID pattern. Notably, the central lobe width in this case shrinks to Φ0 when only the top and bottom surface states dominate. This is due to the destructive interference at half flux quantum in a SQUID-like geometry.

The corresponding supercurrent distributions along the z/y directions (Fig. 5c, e) are obtained by transforming the single-slit pattern to the real-space current density Jc(z) as shown, respectively, in Fig. 5d, f. The full details of the extraction procedure can be found in the Supplementary Note 13 and Supplementary Fig. 22. Clearly, for B applied along the y direction, the supercurrent spreads across the junction (Fig. 5d). On the contrary, when B is applied along z direction, the critical current envelope becomes similar to a sinusoidal oscillation in Fig. 5e. The shift towards a SQUID interference pattern corresponds to the development of sharp peaks in supercurrent density at the mesa edges. The periodicity of the critical current oscillations is contrasted to the observed Fraunhofer pattern in Supplementary Fig. 23 to prove the single period of Φ0 at low magnetic field. The deviation from the single slit in Fig. 5c, d may result from the fluctuations in junction length, position-dependent transparency of interface by a granularity of the Nb film and the variation of the surface carrier density. Despite the deviation, the Fraunhofer pattern is obviously different from the SQUID pattern. The first minimum of Ic at ~0.2 mT, extracted from the Fraunhofer pattern in Fig. 5c, gives the effective length of Leff ~ 1.4 μm, which is slightly larger than the distance of two Nb electrodes. This can be explained by the triangular shape of Nb along the out-of-plane direction in the cross-section Transmission Electron Microscopy image, as shown in Supplementary Fig. 3a where the thinner Nb, close to the edge of Nb electrodes, may become non-superconducting. Besides, the effective length Leff should also include the London penetration depth λL ~ 100 nm for Nb43 such that Leff = L + 2λL. These two reasons may contribute to the larger Leff. When the magnetic field is applied along z-axis, the periodicity of oscillations reaches $$\frac{{{\mathrm{\Phi }}_0}}{{L_{{\mathrm{eff}}}t}}\sim 13\,{\mathrm{mT}}$$ (t is the thickness of Cd3As2), which accords well with the SQUID pattern in Fig. 5e. We can then estimate the width of the supercurrent-carrying surface channel using a Gaussian line shape. The two surface-full regime depths zs are extracted to be 13 and 14 nm, respectively, which is consistent with our experimental results on Andreev reflections across the Nb/Cd3As2 junctions. As control experiments, thick Cd3As2 is found to be hard to achieve Josephson effect while it performs multiple Andreev reflections (MARs) (see Supplementary Note 14 and Supplementary Figs. 2427 for details).

## Discussion

Our experiments on the Nb/Cd3As2 and Nb/Cd3As2/Nb hybrid structures demonstrate the surface superconductivity with a large proximity gap from the parent superconductor, and a detailed supercurrent distribution is extracted from the SQI measurements. Here, we would like to discuss the physical origin of the observed surface superconductivity and its potential implications.

First, we discuss alternative explanations of the surface superconductivity, which do not require the existence of surface states, in our Cd3As2 samples. These possibilities include surface doping by charge impurities and surface band-bending effects44,45. While our transport and SQI measurements alone cannot directly rule out these possibilities, a rather unusual anisotropy of these effects is generally required to be compatible with the surface supercurrent distribution found in our SQI experiments. Particularly, the charge impurities or surface band-bending should occur primarily on the top and bottom surfaces, which can hardly be met by the realistic conditions of our experimental set-up. In contrast, the intrinsic physical properties of surface states (i.e. Fermi arcs), for example, their anisotropic surface distribution and thickness-independent properties, show naturally consistent characteristic with the signatures of the observed surface superconductivity (see Supplementary Note 15 and Supplementary Fig. 28 for details).

We further contrast our result with previous reports on superconductivity in surface states of a usual 3D topological insulator (TI). Topologically nontrivial surface states also exist in TI and similar BICP due to surface states has been reported in Bi2Se3/NbSe2 hybrid structure46. The contribution of surface states and bulk states in TI systems is usually separated by tuning the Fermi level within the bulk gap using electric gating or chemical doping. Generally, sample resistance is a criterion to estimate the bulk states proportion47,48. In Dirac semimetal Cd3As2, an effective tuning of chemical potential32 becomes accessible to change the transport property from bulk-dominated to surface-dominated, which is also justified by two-frequency oscillations (see Supplementary Note 9 and Supplementary Figs. 16-18 for details). Moreover, even eliminating bulk states in TI by chemical doping such as Bi1.5Sb0.5Te1.7Se1.3 system, the SQI result is different in Cd3As2 and TI because surface states cover all the surfaces of TI. Specifically, in TI, due to the uniform surface states in all surfaces, the supercurrent density is edge-dominated along all three directions which results in a mixed Fraunhofer and SQUID-like pattern rather than a pure SQUID pattern49. Nevertheless, the surface states are generally only on two side-surfaces along the y-axis of Cd3As2. This enables the observation of SQUID pattern upon magnetic fields applied along the principal z-axis, which is spatially resolved as edge supercurrent along the y-direction as shown in Fig. 5.

As we introduced earlier, the Fermi-arc states in Cd3As2 originate from edge states of 2D TI embedded in the Dirac semimetal. In particular, Fermi arcs labeled by momentum kz (−k0, k0) stem from the slice of 2D TI indexed by kz. Thus, proximitizing the surface of Cd3As2 is equivalent to inducing pairing in different slices of 2D TI edges as displayed in Fig. 6. Therefore, the Cooper pairs are formed by opposite spin electrons from slices of 2D TIs with opposite kz. In other words, the surface supercurrent observed in our work can be regarded as a higher dimensional analog of edge supercurrent in 2D quantum spin Hall insulators (QSHIs). We note that with certain perturbations that break the C4-cymmetry of Cd3As2, Dirac points located along the principal axis can be gapped out. In this case, the system becomes a strong topological insulator50; in particular, the kz = 0 plane is still characterized by a nontrivial Z2 invariant of 2D QSHIs50. Thus, this subtlety does not affect our conclusion that the observed surface superconductivity provides signatures of higher dimensional edge supercurrents in 2D QSHIs.

The supercurrent distribution in our surface-dominated provides the first direct evidence of surface superconductivity in topological semimetals. While the observations in the current work cannot rule out alternative physical origins other than Fermi arcs, our results call for future efforts to elucidate the possible Fermi-arc origin of the surface superconductivity. Prospectively, when a short linear Josephson junction with π phase difference is formed on the surface of Cd3As2, one pair of Majorana bound states can be created at the junction interface of each slice of 2D TI23. This can result in a large number of non-dispersive Majorana fermions, called Majorana flat bands, that connects the separated Dirac points located at (0, 0, ±k0)20. The Majorana flat bands created by superconducting Fermi arcs can be signified by a characteristic sudden jump at π phase in the Josephson current–phase relation. Similar to the single Majorana bound state at a π-junction interface on quantum spin Hall edges23, the Majorana flat bands on the surface of Cd3As2 are protected by time-reversal symmetry and thus remain robust against non-magnetic disorder. While evidence of single Majorana bound states and chiral Majorana edge states has been spotted in recent experiments51,52, experimental realization of Majorana flat bands still remains unexplored. We believe that our experimental finding of strong proximity effect on surface may establish Cd3As2 to be a new promising platform for realizing Majorana flat bands.

## Methods

### Cd3As2 nanostructure growth

The Cd3As2 nanoplates were grown using Cd3As2 powders as the precursor in a horizontal tube furnace, in which argon was a carrier gas. Before the growth, the furnace was pumped and flushed with argon several times to remove water and oxygen. The temperature was ramped to the growth temperature within 15 min, held constantly for 20 min, and then was cooled down naturally over ~2 h in a constant flow of argon before the substrates were removed at room temperature. The precursor boat was placed in the hot center of the furnace (held at 500 °C), while the smooth quartz substrates were placed in the down-stream within a very small temperature range from 200 to 350 °C. The argon flow rate is 50 SCCM (standard cubic centimeters per minute). The smooth quartz substrates then appeared shining to the naked eyes. The largest crystal plane of as-grown Cd3As2 nanoplates is [112].

### Device fabrication

The Nb/Cd3As2 hybrid structures were fabricated by electron beam lithography (EBL) technique and wet-etched by standard buffered HF solution for 5 s in the electrode regime. For Nb/Cd3As2 device, we first fabricated Cr/Au (10/150 nm) bilayers electrodes on Cd3As2 side using magneton sputtering. Then, we use standard EBL method to deposit the Nb layer. For Nb/Cd3As2/Nb Josephson junction, we use the EBL method and magnetic sputtering to deposit 140 nm-thick Nb electrodes.

### Transport measurements

Four-terminal temperature-dependent transport measurements were carried out in a Physical Property Measurement System (PPMS) system (Quantum Design) (1.9 K) and dilute refrigerator (down to 35 mK) using lock-in amplifier (SR830) and Agilent 2912. The differential conductance (dI/dV) spectra were captured by ac-modulation technique. For the PPMS measurements, lock-in amplifier provides ac input in series with Agilent exporting dc voltage. Through a large resistor (0.05–1 MΩ) the input voltage converts to a constant current. Agilent 2912 and lock-in amplifiers with a low frequency (<50 Hz) were used for dI/dV spectra measurements. The R–T curves were measured by dI/dV spectra while the Agilent dc voltage is set to zero.

### Tight-binding model of Nb/Cd3As2 transport

In the Bloch basis formed by {|S1/2, 1/2〉, |P3/2, 3/2〉, |S1/2, −1/2〉, |P3/2, −3/2〉}, the momentum-space tight-binding Hamiltonian of the Dirac semimetal Cd3As2 used in the theoretical calculations of Andreev reflections is given by22:

$$H_{{\mathrm{DS}}}(k) = {\it{\epsilon }}_0\left( k \right)I_{4 \times 4} + \left[ {\begin{array}{*{20}{c}} {M(k)} & {A_ + (k)} & 0 & 0 \\ {A_ - (k)} & { - M(k)} & 0 & 0 \\ 0 & 0 & {M(k)} & { - A_ - (k)} \\ 0 & 0 & { - A_ + (k)} & { - M(k)} \end{array}} \right]$$

where the matrix elements are defined as:

$$\begin{array}{l}{\it{\epsilon }}_0\left( k \right) = C_0 - E_{\mathrm{F}} + 2C_1\left[ {1 - {\mathrm{cos}}\left( {k_{\mathrm{z}}c} \right)} \right] + 2C_2\left[ {2 - {\mathrm{cos}}\left( {k_{\mathrm{x}}a} \right) - {\mathrm{cos}}(k_{\mathrm{y}}a)} \right]\\ M\left( k \right) = M_0 - 2M_1\left[ {1 - {\mathrm{cos}}\left( {k_{\mathrm{z}}c} \right)} \right] - 2M_2\left[ {2 - {\mathrm{cos}}\left( {k_{\mathrm{x}}a} \right) - {\mathrm{cos}}\left( {k_{\mathrm{y}}a} \right)} \right]\\ A_ \pm \left( k \right) = A_0\left[ {{\mathrm{sin}}\left( {k_{\mathrm{x}}a} \right) \pm i{\mathrm{sin}}(k_{\mathrm{y}}a)} \right]\end{array}$$

Here, a = 3 Å, c = 5 Å refer to the lattice constants within the ab-plane and along the c-axis, respectively. For simplicity, we neglect off-block diagonal parts which only account for higher order or bulk inversion asymmetric terms22. The parameters C0, C1, C2, M0, M1, M2, and A0 are set in units of eV with their values given in Supplementary Table 2 of the Supplementary Information. EF is the Fermi level measured from the Dirac points in the Cd3As2. More details of the numerical Green’s function method for transport calculations in Fig. 4 are presented in Supplementary Note 11.