Introduction

The universal relation between the frequency fJ of the oscillating current and an applied DC voltage bias V across a superconducting weak link1 is determined solely by natural constants.

$$\frac{{f_{\mathrm{J}}}}{V} = \frac{{2e}}{h} = {{\Phi }}_0^{ - 1} = 483.6\,{\mathrm{MHz}}\,{\mathrm{\mu V}}{\kern 1pt} ^{ - 1},$$
(1)

where e is the single electron charge, h is the Planck constant and Φ0 is the superconducting flux quantum. This relation, describing the conventional, 2π-periodic Josephson effect, can be understood as the tunneling of Cooper pairs with a net charge e* = 2e coupled to photons of energy hf2. This coupling, referred to as the AC Josephson effect, has first been measured in superconducting tunnel junctions3 and has been shown to persist in metallic weak links4, carbon nanotubes5 and semiconductor channels6,7, as well as in high critical temperature superconductors8.

In proximitized semiconductor nanowires, an effective superconducting gap with a p-wave symmetry arises due to the breaking of the time-reversal symmetry above a threshold magnetic field9,10,11,12,13,14,15,16. When a weak link is formed between two leads, the p-wave component leads to a factor of two increase in the flux periodicity, giving rise to the so-called 4π-periodic Josephson effect17,18. Phenomenologically, this phase periodicity is equivalent to an effective tunneling charge e* = e instead of 2e in Eq. (1). Therefore, in this Majorana zero mode (MZM) regime, the frequency at a given voltage bias V drops by a factor of two, fMZM(V) = fJ(V)/2, providing a robust signature of the topological phase transition in the superconducting leads. In real devices, however, the finite size of the topological regions19, poisoning events9,18 and Landau–Zener (LZ) tunneling to the quasiparticle continuum20 can effectively restore the 2π-periodic trivial state. The latter two parity-mixing effects cause the system to relax to its ground state, effectively constraining the system in the lowest topological energy branch (red solid lines in Fig. 1a). Nevertheless, out-of-equilibrium measurements performed at rates faster than these equilibration processes can still capture the 4π-periodic nature of topological junctions19,20,21. In contrast, finite-size effects can be avoided by biasing the junction at voltages large enough to overcome the Majorana hybridization gap εM20.

Fig. 1
figure 1

Principles of the experiment. a Energy dispersion of topologically trivial (dashed green line) and nontrivial (solid red line) Andreev levels inside a NW Josephson junction as a function of the phase difference across the junction. The gap εM arises from the finite MZM wavefunction overlap. b Equivalent circuit diagram of the device. The NW junction (in blue box) is capacitively coupled to the superconducting tunnel junction (red box) via the capacitors CC. The microwave losses and stray capacitance are modeled by the RC element enclosed by the dashed black box, see text. The applied DC bias voltages are \(V_{{\mathrm{NW}}}^{\mathrm{S}}\) and \(V_{{\mathrm{DET}}}^{\mathrm{S}}\) with an effective internal resistance rNW and rDET, respectively. c Principle of the frequency-sensitive detection based on photon-assisted tunneling: an absorbed photon with an energy hf gives rise to quasiparticle current if hf > 2ΔDET - eVDET. d Scanning electron micrograph of the NW junction placed on three electrostatic gates. A false color micrograph of the junction is shown in the inset, with the epitaxial Al shell highlighted in cyan. e Bright-field optical image of the coupling circuitry between the NW junction (blue box) and the detector junction (red box). f Micrograph of the split tunnel junction detector. The junctions are encircled. g Measured INW(VNW) characteristics of the NW junction at zero in-plane magnetic field exhibiting a supercurrent branch and multiple Andreev reflections. h Measured IDET(VDET) trace of the detector split junction at zero in-plane magnetic field with a minimized switching current. The insets in (g) and (h) show the large scale I(V) trace of each junction. The normal state conductance GN is given in the units of G0 = 2e2/h. All images and data were taken on device NW1. The scale bars denote 1 μm (d), 10 μm (e), and 1 μm (f), respectively

Full size image

Here, we report the direct observation of a magnetic field-induced halving of the Josephson radiation frequency22 in indium arsenide nanowire (InAs NW) junctions partially covered with an epitaxially grown aluminium shell (Fig. 1d). In this system, possessing a hard induced superconducting gap23, previous direct transport experiments suggest parity lifetimes above 0.1 μs24 and hybridization energies εM ≲ 1 μeV for leads longer than 1.5 μm25. Thus, a frequency-sensitive measurement in the microwave domain is expected to reveal the 4π-periodic Josephson effect26,27.

Results

Frequency-sensitive detection of Josephson radiation

As a frequency-sensitive microwave detector, we utilized a superconducting tunnel junction with a quasiparticle gap of ΔDET, wherein the photon-assisted electron tunneling (PAT) current contributed to the DC current above a voltage bias threshold eVDET > 2ΔDET - hf22,28 (Fig. 1c). This on-chip detector29, coupled via capacitors CC to the NW junction (see Fig. 1b for the schematics and Fig. 1e for an optical image of the device), was engineered to result in an overdamped microwave environment characterized by a single fc = (2πRC)-1 ≈ 28 GHz cutoff frequency with R = 538 Ω and C = 10.4 fF (see Supplementary Fig. 2). The resulting broadband coupling to the detector7 inhibited higher order photon emission, which could mimic the 4π-periodic Josephson effect30.

The nanowire was deterministically deposited on a set of three gates covered by 30-nm thick SiNx dielectric as shown in Fig. 1d. The Josephson weak link, where the Al shell was removed by wet chemical etching, was located above the central gate (see inset of Fig. 1d). We investigated devices with junction lengths ranging from 86 to 271 nm. The high quality of the nanowire junction is apparent from the presence of distinct multiple Andreev reflection steps in its INW(VNW) characteristics31 (Fig. 1g), which is a signature of the hard superconducting gap in the nanowire23. The observed curves and linear conductance also establish that no conductive mode with a transmission close to unity exists in the channel, which could contribute to the 4π-periodic signal even in the absence of topological ordering20.

The microwave detector, presented in Fig. 1f, was fabricated using two angle-evaporated32 Al/AlOx/Al tunnel junctions, forming a superconducting quantum interference device. This geometry allowed us to minimize the Josephson energy of the detector by applying Φ = Φ0/2 flux through the loop (see Fig. 1h) and thus to limit its back action to the nanowire. The respectively 8- and 11-nm thick Al layers set an in-plane critical magnetic field of the detector in excess of 1 T, well above the measured topological transition in similar devices25. Nevertheless, increasing subgap currents limited our experimental field range to 325–650 mT for different devices. The circuit parameters and fabrication details are given in the Supplementary Tables and in the Methods section, respectively.

In the presence of a voltage spectral density SV(f), the DC current contribution of the PAT process is as follows22,28 in the subgap regime, where eVDET < 2ΔDET:

$$I_{{\mathrm{PAT}}}\left( {V_{{\mathrm{DET}}}} \right) = {\int}_0^\infty {\kern 1pt} {\mathrm{d}}f\left( {\frac{e}{{hf}}} \right)^2S_V(f)I_{{\mathrm{QP,0}}}\left( {V_{{\mathrm{DET}}} + \frac{{hf}}{e}} \right).$$
(2)

Here, IQP,0(VDET) is the tunnel junction current in the absence of absorbed radiation, SV(f) = 0 (see Fig. 1h). Note that the quasiparticle gap edge at eVDET = 2ΔDET resulted in a sharp increase of IQP,0(VDET). In the presence of monochromatic radiation with a frequency f0, SV(f) ~ δ(f - f0), IPAT(VDET) thus developed a step-like feature at hf0 = 2ΔDET - eVDET. With a phenomenological effective charge e* of the AC Josephson effect, we write this condition in terms of the voltage drop on the nanowire, VNW.

$$e^ \ast V_{{\mathrm{NW}}} = hf_0 = 2{\mathrm{\Delta }}_{{\mathrm{DET}}} - eV_{{\mathrm{DET}}},$$
(3)

where e* = 2e for conventional junctions (see Eq. (1)) and e* = e in the 4π-periodic regime. To extract e* and thus determine the periodicity of Josephson radiation, we tracked the transconductance peak dIPAT/dVNW(VNW, VDET) measured by standard lock-in techniques at a frequency of 17.7 Hz (see Supplementary Fig. 1). The experiments were performed at the base temperature of a dilution refrigerator (~20 mK).

Josephson radiation as a function of the magnetic field

Typical experimental datasets are shown in Fig. 2 for two nanowire junctions (NW1 and NW2, respectively) as the source of Josephson radiation. We limited the detector voltage range by the condition dIDET/dVDET < 10 μS, where the subgap quasiparticle current was still negligible, typically IDET ≲ 1 nA. A lower limit of the emitter junction voltage was defined by the phase diffusion regime33 characterized by periodic switching and retrapping events, which breaks the validity of Eq. (1) (see Supplementary Note 3). We therefore did not consider the low VNW regime, within the supercurrent peak. We show this range, excluded from the linear fits, shaded in gray in Figs. 2 and 3 (see Supplementary Note 2 on the characterization of these limits). We fit the peak positions using Eq. (3) in order to extract e* and ΔDET as a function of the applied in-plane magnetic field. The typical standard deviation was 3.6 GHz for each frequency data point (see Supplementary Note 2). The error bars of the fitted parameters were determined using the bootstrapping method34 (see Supplementary Note 2), and they show the full width at half maximum yielding a confidence level of 75% for a Gaussian lineshape.

Fig. 2
figure 2

Magnetic field-induced 4π-periodic Josephson radiation. Differential transconductance dIPAT/dVNW as a function of VNW and VDET for devices NW1 (a, b) and NW2 (c, d) at zero and finite magnetic fields, respectively. The position of the transconductance peak maps the frequency of the monochromatic Josephson radiation. A linear fit e*VNW = 2ΔDET - eVDET through these peaks is shown as an orange line. Dashed green and red lines show linear fits with a fixed slope corresponding to e* = 2e and e* = e, respectively. The shaded regions show the regimes where the fit of the transconductance peak is not reliable, see text. Two normalized and smoothed horizontal linecuts are plotted, where arrows point to the position of the extracted peaks. The orange, green, and red dots denote the position of the best fit, the e* = 2e fit, and the e* = e fit, respectively. The evolution of e*(B) and 2ΔDET(B) is presented in (e, f) for NW1 and in (g, h) for NW2. For the calculation of the error bars, see text and Supplementary Note 2. The transition from the 2π- to 4π-periodic Josephson radiation is observed between 175 and 300 mT as e* evolves from values near 2e (green circles) to values close to 1e (red triangles). For all devices, 2ΔDET(B) drops monotonically (black dashed line, see text), independently of the change in e*

Full size image
Fig. 3
figure 3

Gate tuning of the 4π-periodic radiation regime. Differential transconductance dIPAT/dVNW as a function of VNW and VDET for device NW3 at gate setting A (a, b) and setting B (c, d) at zero and finite magnetic fields, respectively. A linear fit and fits with fixed slopes e* = 2e and e* = e are shown as an orange line, a dashed green line, and a dashed red line, respectively. Two normalized and smoothed horizontal linecuts are also presented, in which the arrows point to the position of the extracted peaks. The evolution of e*(B) and ΔDET(B) is shown in (e, f) for setting A and in (g, h) for setting B. A transition from 2π- to 4π-periodic Josephson radiation is observed for gate setting A, but the radiation remains 2π-periodic for setting B. The gate voltage values are shown in Supplementary Table 2. For the calculation of the error bars, see text and Supplementary Note 2

Full size image

At zero magnetic field (Fig. 2a, c), the emitted Josephson radiation is always 2π-periodic with an extracted effective charge close to e* = 2e, as shown by the good agreement between the orange line and the dashed green line (best fit with fixed e* = 2e). In contrast, NW1 and NW2 exhibit the 4π-periodic Josephson effect above a threshold magnetic field (Fig. 2b, d), where e* ≈ e. The full evolution is shown in Fig. 2e, g, where a sharp transition is visible from e* ≈ 2e (green circles) to e* ≈ e (red triangles). Finally, the fitted ΔDET (Fig. 2c, f) shows a monotonic decrease described by \(2{\mathrm{\Delta }}_{{\mathrm{DET}}}(B) = 2{\mathrm{\Delta }}_0\sqrt {1 - B^2{\mathrm{/}}B_{\mathrm{c}}^2}\) for all devices (dashed lines), with no additional feature at the transition field. In contrast with the nanowire junctions, our control device, an Al/AlOx/Al tunnel junction, exhibits no transition in e* over the entire magnetic field range (see Supplementary Fig. 5).

Josephson radiation at different gate voltages

Figure 3 shows the magnetic field evolution of device NW3 at two distinct gate settings with similar GN and dIPAT/dVNW corresponding to similar Josephson couplings. By tuning the chemical potential in the nanowire via changing the gate voltages, it is possible to displace the position of the onset of the 4π-periodic Josephson radiation from ≈175 mT (Fig. 3b) to values larger than 375 mT (Fig. 3d). Note that the additional local maximum at high VNW values, also observed in earlier experiments7, is attributed to the shot noise of the nanowire junction.

The possibility to tune the nanowire devices into the 4π-periodic Josephson radiation regime with both magnetic field and chemical potential is consistent with the predicted phase diagram of this system9,10,18. We observed the same behavior in four distinct nanowire devices (see Supplementary Fig. 4 for device NW4), which we could interpret within the single sub-band model of the topological phase transition that takes place at a magnetic field B*, where \(E_{\mathrm{z}} = g\mu _{\mathrm{B}}B^ \ast {\mathrm{/}}2 = \sqrt {{\mathrm{\Delta }}_{{\mathrm{NW}}}^2 + \mu _{{\mathrm{NW}}}^2}\). Here g and μB are the Landé g factor and the Bohr magnetron, respectively. From our device parameters (see Supplementary Table 2), lower bounds on the g factors ranging from g ≈ 11 (B* = 175 mT) in device NW3 to g ≈ 35 (B* = 190 mT) in device NW4 were obtained, in agreement with the values reported in similar devices15,25,35. In contrast, an accidental crossing of a trivial Andreev bound state would be inconsistent with the observed field range of ΔB ~ 0.3 T of the 4π-periodic radiation, since within this range, a spinful Andreev level35 would evolve over the scale of the superconducting gap, ΔNW ~ gμBΔB, suppressing the 4π periodicity. We, however, did not observe a continuous variation of the onset magnetic field B* as a function of the applied gate voltages. This behavior is consistent with the calculations of the topological phase diagram based on realistic device simulations including orbital effects of the magnetic field36 and multiple spatial dimensions37,38 of the device.

We observed a single Josephson radiation frequency in the 4π-periodic regime, which was consistent with the supercurrent being predominantly carried by a single transmitting mode. While we were not able to reliably extract the transparency and the number of modes in our devices, the single-mode regime was observed earlier in similar InAs nanowires35,39,40. We also note that an upper bound on the channel transmission of τ = GN/G0 could be determined from the normal state conductance GN < G0, which was measured in the linear regime well above the superconducting gap. This value is shown in Figs. 2 and 3 for each device.

Numerical simulations of the Josephson radiation frequency

Next, we numerically evaluated the expected voltage spectral density seen by the detector junction in various regimes. We used the quasiclassical resistively and capacitively shunted junction model coupled to a stochastic differential equation describing the occupation of the single pair of Andreev levels in the NW junction. The equivalent circuit of the device in the microwave domain is shown in Fig. 1b, in which each element is experimentally characterized7 (see Supplementary Fig. 2 and Supplementary Tables). Note that we neglected the load of the detector on the circuit, which is justified by its negligible subgap conductance compared to that of all other elements in the circuit.

Our model of the nanowire junction considers LZ tunneling between branches of the energy-phase dispersion shown in Fig. 1a, as well as tunneling to the continuum and stochastic quasiparticle poisoning events20. The probability of LZ tunneling is determined by the voltage drop VNW according to PLZ = exp(-V0/VNW), where \(eV_0 = 4\uppi\varepsilon _{\mathrm{M}}^2{\mathrm{/}}\left( {{\mathrm{\Delta }}_{{\mathrm{NW}}}\sqrt \tau } \right)\) is the characteristic voltage above which PLZ ~ 1. In this limit, 4π periodicity is observed despite the gap εM caused by finite-size effects25. Similarly, LZ tunneling to the continuum close to φ = 2π defines a voltage scale \(eV_1 = 2{\mathrm{\pi\Delta }}_{{\mathrm{NW}}}\left( {1 - \sqrt \tau } \right)^2{\mathrm{/}}\sqrt \tau\), above which 2π periodicity is restored20. We note that a trivial Andreev bound state in the short junction limit can be modeled similarly with eV0 =π ΔNW(1 - τ) and eV1 = 0.

Figure 4 shows representative plots obtained by numerically evaluating SV(f, VNW) (see Supplementary Note 3), which determines the photon-assisted tunneling current by Eq. (2). We observed that the numerical results agreed well with the characteristic features of the experimental data. We found that the circuit equations allowed for a phase-diffusion regime at low VNW values33, where e*VNW < hf, because the junction spent part of the time in the steady supercurrent state in which the voltage drop was zero. The calculations also reproduced the absence of higher harmonics in the radiation spectrum, attributed to the low transmission of the junction and the overdamped nature of the microwave environment30. This confirms our expectation of the suppression of multiphoton processes due to a low-quality factor, justifying the usage of the semiclassical junction model.

Fig. 4
figure 4

The calculated radiation spectrum. The voltage spectral density SV(f) incident on the detector junction, computed by numerically solving the system of stochastic differential equations shown in Supplementary Note 3. a, b Show results for a junction in the trivial regime (small transmission and large transmission, respectively), while (c, d) show the emission spectrum in the topological regime. V0 and V1 are voltage scales for Landau–Zener tunneling between branches of the junction bound state and for tunneling to the quasiparticle continuum, respectively; see text. Circuit parameters are set as rNW = 2.4 kΩ, RNW = 50 kΩ, R = 0.5 kΩ, C = 10 fF, CC = 400 fF, and \(I_{{\mathrm{NW}}}^{\mathrm{0}} = 8{\kern 1pt}\) nA. The noise temperature T = 150 mK and the quasiparticle poisoning rate is Γq = 100 MHz. As in Fig. 2, the dashed green and red lines show the frequency of Josephson radiation corresponding to e* = 2e and e* = e, respectively. The estimated phase-diffusion region is shaded in gray

Full size image

A key result of these simulations in a wide range of junction parameters is that with the circuit elements taking values representative of those in the experiment, the radiation frequency always reflects the internal dynamics of the nanowire Josephson junction both in the 2π-periodic (Fig. 4a, b) and in the 4π-periodic emission regimes (Fig. 4c, d). Finally, we note that our results are consistent with V0 ≲ 15 μeV translating to an avoided crossing εM ≲ 10 μeV. Using the exponential cutoff in ref. 25, this suggests that our devices have a continuous topological region of several hundreds of nanometers on each side of the nanowire junction, which is consistent with the scanning electron microscopic images of the devices.

Discussion

In conclusion, we observed the 4π-periodic Josephson effect in multiple InAs nanowires above a threshold magnetic field in a range of 175–300 mT. This effect, which can be suppressed by tuning the gate voltages, is consistent with the expected signatures of a topological phase transition. By observing the periodicity of Josephson effect using an on-chip microwave detector, we investigated this system while preserving its charge parity, in line with the requirements for prospective topological quantum computers. This experimental technique may also prove instrumental in identifying more exotic non-Abelian anyon states41,42, due to its proven sensitivity to the periodicity of Josephson effect, directly measuring the charge fractionalization of the anyon state43,44.

Methods

Device fabrication

The devices were fabricated on commercially available undoped Si substrates with a 285-nm thick insulating SiOx layer in a similar fashion as in refs. 7,35. All etching and metal-deposition steps were realized using standard positive-tone electron-beam lithography techniques. At first, three Ti/Au (5 nm/15 nm) electrostatic gates and the coupling capacitor bottom plates were deposited (see Fig. 1 for design details). These were subsequently covered by an ~30-nm thick SiNx dielectric layer deposited by sputtering. Eleven 100-nm wide Cr/Pt (5 nm/25 nm) tracks were then defined. These ~100 Ω μm-1 resistive lines connected the gates, the (yet to be defined) Al/AlOx/Al detector, and the nanowire to the instrumentation setup. Next, the Al/AlOx/Al Josephson junctions were fabricated by evaporating 8- and 11-nm thick Al layers with an intermediate in-situ oxidation step at 0.5 mbar for 4 min using the Dolan bridge technique32. The nanowires were then deterministically deposited onto the electrostatic gates with a micro-manipulator setup equipped with an optical microscope. A gap in the nanowire Al shell was then created by wet etching using Transene D at a temperature of 48.2 °C for 12 s. Next, both the nanowire and the detector junctions were connected to the resistive lines with an 80-nm thick sputtered NbTiN film after an in-situ Ar plasma milling step. Finally, a Ti/Au (15/100 nm) layer was evaporated to define quasiparticle traps, the upper capacitor plates, and the contact pads. We note that no NbTiN film was used in device NW3. Instead, a Ti/Au (15/100 nm) layer was used to contact the nanowire and the detector. The dimensions and properties of each device are presented in Supplementary Table 1, and the experimental setup is described in Supplementary Fig. 1. We note that the detector was made of narrow and thin aluminum sections (see Fig. 1f) to limit the presence of vortices near the Al/AlOx/Al junctions, and thus to decrease the subgap current in finite magnetic field.

The InAs nanowires used in this work were grown via a two-step process by molecular beam epitaxy. The InAs nanowires were grown at 420 °C using the vapor–liquid–solid method with Au droplets as the catalyst. After cooling the system to -30 °C, Al was epitaxially grown on two of the six nanowire facets23.

The microwave environment of the InAs Josephson junction

We modeled the effective microwave environment of the nanowire Josephson junction with a parallel lumped resistor (R) and capacitor (C) element, which accounted for the low-pass nature of the coupling circuit (see inset of Supplementary Fig. 2a). We determined the effective RC values by measuring a sample in which the nanowire junction was replaced by an Al/AlOx/Al tunnel junction. The supercurrent peak was fitted against the Ivanchenko–Zil’berman model to find the RC values and the noise temperature of the circuit33 at zero magnetic field (see Supplementary Fig. 2a). The critical current as a function of the magnetic field was then found using the same model, with the RC and the noise temperature fixed at their zero field values (Supplementary Fig. 2b). We note that the same coupling circuit was used in ref. 7, leading to RC and noise temperature values in good agreement with the current ones. Thus, we conclude that the reproducibility is good for all the samples featured in the current study. These parameters are used to theoretically study the dynamics of Josephson radiation.

Reproducibility of the transition for nanowire devices

Supplementary Figure 3 shows every differential transconductance color plot from which the effective charge e* has been extracted in Fig. 2e. The color plots nominally follow the same trend as the ones presented in Fig. 2. Supplementary Figure 4 shows the magnetic field evolution of e* in device NW4. Device NW4 also exhibits a transition from 2π- to 4π-periodic Josephson radiation at B ~ 175 mT. As such, the observation of a magnetic field-induced transition in the periodicity of Josephson radiation has been observed in four distinct devices, showcasing the reproducibility of the observation.

Josephson radiation of an Al/AlOx/Al tunnel junction

Supplementary Figure 5 shows our measured data with a conventional Al/AlOx/Al superconducting tunnel junction as the source of Josephson radiation. Evaluating e* as a function of magnetic field in the same range as for Fig. 2 and Supplementary Fig. 4, we observed no transition in the periodicity of Josephson effect, confirming that the 4π-periodic Josephson radiation occurs only in nanowire junctions. We note that, in order to keep the circuit behavior similar, the normal state conductance of the tunnel junction was set to GN,T = 0.26G0.