Abstract
In typical metallic tunnel junctions, the tunneling events occur on a femtosecond timescale. An estimation of this time requires current measurements at optical frequencies and remains challenging. However, it has been known for more than 40 years that as soon as the bias voltage exceeds one volt, the junction emits infrared radiation as an electrically driven optical antenna. We demonstrate here that the photon emission results from the fluctuations of the current inside the tunneling barrier. Photon detection is then equivalent to a measurement of the current fluctuations at optical frequencies, allowing to probe the tunneling time. Based on this idea, we perform optical spectroscopy and electronic current fluctuation measurements in the far from equilibrium regime. Our experimental data are in very good agreement with theoretical predictions based on the Landauer Büttiker scattering formalism. By combining the optics and the electronics, we directly estimate the socalled traversal time.
Introduction
Among the most intriguing questions in quantum mechanics, the traversal time τ_{T} that a particle takes to tunnel through a barrier has been the subject of a longstanding debate. Although several theoretical expressions have been proposed^{1}, the approach using a physical clock such as a Larmor’s clock is now well accepted^{2} and yields \(\tau _{\mathrm{T}} = {\int}_{\cal F} {\kern 1pt} dz[m{\mathrm{/}}(\hbar \kappa (z))]\) where κ(z) is the magnitude of the imaginary wave vector in the classically forbidden region \({\cal F}\). Experimentally, this question has mainly been addressed in attosecond physics. Taking advantage of strong laserfield techniques, fundamental questions related to tunneling time in atom ionization have been clearly addressed^{3,4,5}. As an alternative, electronic transport measurements through a tunnel junction allow for experimental control of the tunneling parameters and seem to be a natural way to investigate τ_{T}. However, the experiments so far focused on the I(V) characteristics which mainly probe the tunneling rate related to the transmission of the barrier \({\cal T}\). The traversal time thus only appears as a correction to the tunneling current due to the dynamical feedback of the electromagnetic environment^{6,7}.
An alternative way to estimate the traversal time by measuring the current–current correlations on a timescale comparable to the duration of the tunneling process. This is equivalent to measuring the current noise spectral density S_{ii} at frequencies ν ~ 1/τ_{T}. For a typical aluminum oxide barrier, τ_{T} is of the order of femtoseconds and necessarily implies a measurement of S_{ii} at optical frequencies. Note that for a rectangular barrier, the time for an electron to cross the barrier is given by \(\tau _{\mathrm{T}}\) = \(d\sqrt {m{\mathrm{/}}\left( {2(U  eV)} \right)}\) where U is the barrier height, d its thickness, m the effective mass of electron and V the bias voltage. For a common aluminum oxide barrier U ~ 2.5 eV, d ~ 2 nm and m ~ 3.5 × 10^{−31} kg which gives τ_{T} ~ 2 × 10^{−15} s at 1 V. In the following, we probe the traversal time by studying the infrared radiation emitted from a macroscopic planar junction^{8,9}. This emitted light is broadband with a highfrequency cutoff determined by the applied voltage hν < eV and was first detected in 1976 by Lambe and McCarthy^{10}. It was immediately attributed to the inelastic excitation of surface plasmon polariton (SPP) junction modes which coupled to photons due to the roughness of the metal electrodes. The tunneling current flowing through the junction is indeed accompanied by the generation of light which originates from the highfrequency component of current shot noise in the tunnel barrier^{8,9}. At such voltage (~1 eV), the junction is in the far from equilibrium regime (FFER). Even if light has been observed from biased junctions prepared by electromigration^{11} and from those formed between a biased scanning tunneling microscope (STM) tip and metallic sample^{12,13,14}, heating of electron gas in such nanoscopic junctions make the quantitative study of the transport properties of the junction difficult. From a theoretical point of view, the work of Lesovik and Loosen^{15} could explain the reduction in light emission observed in such STM experiments as a reduction of the quantum shot noise.
In this article, we first discuss the validity of the theory given by the outofequilibrium generalized fluctuation dissipation relation (FDR). We show that it provides information about temporal correlations between electrons but fails to describe the dynamics of the tunneling process on the femtosecond timescale in the FFER. We then generalize the existing theoretical expression of the shot noise spectral density S_{ii} to optical frequencies by using a Landauer–Büttiker (LB) scattering approach. By using electronic and optical measurements, we use the universal expression for shot noise in the radio frequency range (\(\nu \ll 1{\mathrm{/}}\tau _{\mathrm{T}}\)) to experimentally prove the validity of the tunneling limit at high bias voltage in the FFER. Finally, we demonstrate that the FDR breaks down at optical frequencies. We show that our experimental results are in very good agreement with the generalized expression of S_{ii} and that they give a very good estimation of τ_{T}, the onedimensional (ID; longitudinal) traversal time. This defines the unique characteristic time for the tunnel junction^{16} which can thus be seen as a promising electrically driven source of SPP for future plasmonic circuits.
Results
Fluctuation dissipation relation
In the tunneling limit, the noise spectral density at a finite frequency ν usually reads^{17,18,19,20}:
where I(V) is the dc characteristic of the voltagebiased tunnel junction, \(N(\epsilon )\) = \(1{\mathrm{/}}\left( {{\mathrm{exp}}\left( {\epsilon {\mathrm{/}}k_{\mathrm{B}}T} \right)  1} \right)\) the Bose–Einstein distribution and e denotes the elementary charge of electron. It is referred to as the FDR and is generally valid as shown in refs. ^{19,20}. The FDR is indeed in quantitative agreement with numerous experiments in the microwave regime. These experiments range from simple linear tunnel junctions^{21} to tunnel junctions showing nonlinear characteristics due to superconducting electrodes^{22} or to dynamical Coulomb blockade effect^{23}. According to Eq. (1), the noise spectral density S_{ii} vanishes for frequencies ν > eV/h as long as thermal fluctuations are negligible \(\left( {h\nu \gg k_{\mathrm{B}}T} \right)\). This has a simple interpretation: in the wavepacket picture^{24,25}, the electrons are emitted from a constant voltage source with an average time separation τ_{Q} = h/eV. It is thus impossible to measure current–current correlations on a timescale smaller than τ_{Q} (Fig. 1). τ_{Q} characterizes the temporal correlations between successive attempts of the electrons to cross the junction. However, if τ_{Q} is a characteristic time of the FDR, the traversal time τ_{T} is not. This means that the FDR does not provide a complete picture of the tunneling process. The reason arises from the fact that tunneling events are actually supposed to be instantaneous in the FDR (see Methods). We thus expect a breakdown of the FDR at frequency 1/τ_{T} ~ ν ≲ 1/τ_{Q} proving the existence of the traversal time τ_{T}. The central result of this article is the observation of this breakdown by comparing the measured noise spectral density S_{ii}(eV, hν) to the FDR deduced from the measured characteristics I(V). This result constitutes direct evidence of the existence of a characteristic time associated with tunneling.
Nonlinear tunneling transport
The FFER is reached when the applied bias voltage is of the order of the tunnel barrier height U. In this regime, the tunneling barrier is modified by the bias voltage which leads to an intrinsic nonlinear conductance. Without a careful study of the Coulomb interactions in the tunnel barrier, gauge invariance (invariance of the current under a global voltage shift applied on both electrodes) is not systematically satisfied^{26,27}. It is indeed necessary to determine the electrical potential which depends on the applied bias voltage and the possible charge accumulation in the conductor. Instead of the transfer Hamiltonian formalism, here we use the LB description to properly describe the ballistic transport through the barrier (Supplementary Note 3). Its transmission \({\cal T}\) is thus necessarily both energy and voltage dependent and its I(V) characteristic reads:
where \(f(\epsilon )\) = \(1{\mathrm{/}}\left( {1 + {\mathrm{exp}}\left( {\left( {\epsilon  \epsilon _{\mathrm{F}}} \right){\mathrm{/}}k_{\mathrm{B}}T} \right)} \right)\) is the Fermi–Dirac distribution with \(\epsilon _{\mathrm{F}}\) the Fermi energy. In the tunneling limit, the voltage dependence of \({\cal T}\) can be deduced by using the freeelectron density of states and the Wentzel–Kramers–Brillouin (WKB) transmission coefficient^{28} and by considering a total potential including the potential barrier U(z) and the biasing energy U_{bias}(z, V) = −eV(1 − z/d) as depicted in Fig. 2. Note that the effects of image charge could be considered in the potential barrier U: in that case, these effects would only renormalize the barrier height. They will be taken into account only to estimate the tunneling current. It is worth emphasizing that the biasing energy is essential to explain the nonsymmetric I(V) characteristics depicted in Fig. 2 (Supplementary Fig. 4). We now consider the current fluctuations characterized by the nonsymmetrized shot noise spectral density \(S_{\alpha ,\beta }\) = \(\left\langle {\hat I_\alpha (  \nu )\hat I_\beta (\nu )} \right\rangle {\mathrm{\Delta }}f\) where \(\hat I_\alpha (\nu )\) is the Fourier component of the current operator measured in the electrode α, β = L, R and Δf the measurement bandwidth. For ν > 0, this quantity refers to the emission quantum noise which is measured in a passive detection scheme such as the photon detector used here^{15,27}. Using the scattering LB approach for a single quantum channel of conduction in the tunneling limit (\({\cal T} \ll 1\)), we get for α ≠ β^{27}
where \(f_{\mathrm{L}}(\epsilon )\) = \(f(\epsilon + eV)\) and \(f_{\mathrm{R}}(\epsilon )\) = \(f(\epsilon )\). In the zero frequency limit and using the expression \(f\left( {\epsilon + \epsilon _0} \right)\left( {1  f(\epsilon )} \right)\) = \(N\left( {\epsilon _0} \right)\left( {f(\epsilon )  f\left( {\epsilon + \epsilon _0} \right)} \right)\), the above expressions simplify to S_{LL} = S_{RR} = −S_{LR} = \(S_{{\mathrm{ii}}}^{{\mathrm{(FDR)}}}\) and the FDR holds even in the nonlinear regime. However, at finite frequency, the energy and voltage dependence of the transmission \({\cal T}\) leads to charge accumulation in the barrier and the noise spectral density depends on the electrode where it is evaluated (S_{LL} ≠ S_{RR} ≠ −S_{LR})^{27,29}. A key question then arises: which noise spectral density is measured at optical frequencies? Following Nyquist^{30}, the noise spectral density S_{ii} related to the radiated spectral power P_{ν} emitted by the tunnel junction is given by:
where the radiation impedance \({\cal R}(\nu )\) stands for the coupling between the current in the conductor and the farfieldradiating electromagnetic modes. This radiation impedance takes into account the timescales related to the plasmonic in the metallic contacts and the stemming from the optical setup. It clearly does not depend on the dc voltage and is therefore not relevant to the assessment of the traversal time. At optical frequencies, this coupling takes place in the insulating barrier because of the screening of the electromagnetic field in the metallic electrodes (Supplementary Fig. 10). An estimation of this coupling requires a microscopic description of the charge transfer. Even though it should be necessary to solve the coupled system of Schrödinger and Poisson equations to calculate the tunneling current \(\hat I_{\mathrm{T}}\), the screening in the metallic electrodes permits a simple description of \(\hat I_{\mathrm{T}}\). The bare electron inside the tunneling barrier induces a polarization charge e(1 − z/d) and ez/d in the left and right electrodes respectively (Fig. 3c). We can thus assume that the charge accumulation on the surface of the electrodes is equal on average during the tunneling event: \(\hat Q_{\mathrm{L}} = \hat Q_{\mathrm{R}} = \hat Q\) with \(\left\langle {\hat Q} \right\rangle = e{\mathrm{/}}2\). The continuity equation \(d\hat Q{\mathrm{/}}dt\) = \(\hat I_{\mathrm{L}}  \hat I_{\mathrm{T}}\) = \(\hat I_{\mathrm{T}} + \hat I_{\mathrm{R}}\) then implies \(\hat I_{\mathrm{T}}\) = \(\left( {\hat I_{\mathrm{L}}  \hat I_{\mathrm{R}}} \right){\mathrm{/}}2\) with the conventional direction of the current (Fig. 2). The current noise spectral density \(S_{{\mathrm{ii}}}\) = \(\left\langle {\hat I_{\mathrm{T}}(  \nu )\hat I_{\mathrm{T}}(\nu )} \right\rangle {\mathrm{\Delta }}f\) in Eq. (4) is then given by:
Although this expression is similar to that written in refs.^{14,31}, it should be noted that the authors describe the charge conservation in a central dot (molecular junction) using a transfer Hamiltonian model. They are mainly interested in transfer rate and not in tunneling time. In our case, we are interested in the tunneling region described by the scattering formalism. Our formalism takes into consideration not only the effect of the Coulomb interactions in the barrier but also the time delay of the electron crossing the barrier. This is hidden in the strong energy dependence of the tunneling transmission. Under these conditions, the FDR cannot be satisfied when ν ~ 1/τ_{T} (Methods).
Experimental setup
Our experimental setup is shown in Fig. 3c. Electronic and optical measurements are performed in a cryogenic environment at T ~ 100 K to prevent junction breakdown and to reduce the thermal noise in the infrared photon detector. The sample is a 100 × 100 μm^{2} planar Al/AlO_{x}/Al tunnel junction deposited on a sapphire substrate (Fig. 3a and Supplementary Note 1). Because of the layered structure of the junction, the electromagnetic modes known as SPPs in the junction should not radiate to the free space. However, the total thickness of the junction a ~ 10 nm is smaller than the penetration depth of the SPP in the metal: δ_{p} = c/ω_{p} ≃ 13 nm with ω_{p} = 14.7 eV the plasma frequency of aluminum. This enables the coupling between the SPP mode localized at the interface electrode/vacuum (Fig. 3c) and the propagating mode in the substrate (Supplementary Figs. 8–11). This corresponds to the Kretschmann configuration where the coupling appears at a specific angle θ_{p} ≃ arcsin(1/n) ≃ 35° where n stands for the refractive index of sapphire^{32}. We use total internal reflection in a conical prism to collect the emitted photons (Fig. 3b, c). The current noise S_{ii}(eV, hν) at an optical frequency ν is measured at two different frequencies corresponding to the wavelengths λ = c/ν = 0.9 ± 0.02 μm and 1.3 ± 0.015 μm. The current noise at zero frequency S_{ii}(eV, hν = 0) is measured at radio frequencies with a standard crosscorrelation technique (Supplementary Note 2 and Supplementary Figs. 1–3).
Noise measurement in the radio frequency range of 20 to 100 kHz
At high voltage, the I(V) characteristic shown in Fig. 2 exhibits a strong nonlinearity: the differential resistance varies by more than one order of magnitude ranging from 6 kΩ at low bias to 150 Ω at high bias. From the theoretical expectation of Eq. (2) and using the WKB approximation assuming an homogenous trapezoidal barrier^{33} to evaluate the transmission of the tunnel junction, we estimate the mean barrier height U ~ 2.7 eV, its asymmetry ΔU ~ 2.9 eV and its thickness d ~ 2 nm (lower inset of Fig. 2 and Supplementary Figs. 5 and 7). Although the junction is symmetrical (identical electrodes), the strong asymmetry of the barrier can be explained by the roughness of the thin electrodes as has already been reported in the literature^{34}. The thickness d is in agreement with the capacitance of the junction ~0.5 nF.
In the FFER, the bias voltage is of the order of the tunnel barrier height and it is legitimate to wonder if the tunneling limit is still valid and if the traversal time still has a meaning. As previously discussed, the FDR is universal at zero frequency and may be used to test the tunneling regime^{35}. Current fluctuations S_{ii}(eV, hν = 0) are measured in the radio frequency range with low noise voltage amplifiers giving access to voltage fluctuations S_{vv} = g(Z_{setup}(eV)^{2}S_{ii} + S_{vv,setup}(eV)) where g is the global gain of the amplifier chain, Z_{setup} is the transimpedance of the measurement setup and S_{vv,setup} its excess noise. Because of the large variation of the tunneling resistance, a careful calibration is required to extract the current noise S_{ii}. The voltagedependent transimpedance Z_{setup} and the excess noise S_{vv,setup} are determined by using an external noise source while g is deduced from the measurement of the shot noise in the linear regime (Supplementary Note 2). Figure 4 shows the current noise S_{ii} in the FFER. Although the tunnel resistance is strongly nonlinear, S_{ii} clearly satisfies the FDR at zero frequency,
In the high bias limit \(eV \gg k_{\mathrm{B}}T\), the current noise is linearly proportional to the dc current. The Fano factor F = S_{ii}(eV, hν = 0)/(eI) is then equal to one and confirms that electronic transport through the junction operates in the tunneling limit even at high bias voltage. This rules out the presence of pinholes in the barrier since it would imply a significant reduction of the shot noise with F < 1. Let us assume that the static resistance decreases from R = 6 kΩ to R = 0.85 kΩ because of the appearance of N pinholes with an average transmission \({\cal T}_{\mathrm{p}}\sim 1\). This would imply \(N{\cal T}_{\mathrm{p}}\) ≃ 13 at high bias voltage. The Fano factor related to the parallel association of tunneling channels and pinholes would then be \(F\) = \(1  {\textstyle{{2R} \over {R_{\mathrm{K}}}}}N{\cal T}_{\mathrm{p}}^2\), with R_{K} = h/e^{2} ≃ 25.8 kΩ the quantum of resistance. Finally, \(F \simeq 1  0.85{\cal T}_{\mathrm{p}}\) ~ 0.15, this is not what we observe. We notice systematic errors at high positive bias. They cannot be attributed to Joule heating since they should also be observed for negative bias. The fact that the calibration is off by ~10% is attributed to parasitic capacitances of the measurement setup which are not included in Z_{setup}.
Noise and traversal time estimation at optical frequencies
Figure 3d shows an image of the light emission pattern from the tunnel junction when the camera is focused on the conical prism. In the center, a small amount of light comes directly from the tunnel junction (zoom in Fig. 3d). This is due to the surface roughness of the electrodes allowing SPP scattering at the surface of the upper electrode^{36}. The homogenous light intensity indicates that the electrontophoton conversion in the tunnel junction is also homogenous over the surface of the junction confirming the absence of pinholes in the barrier. However, the bright ring in Fig. 3d reveals that more than 98% of the light is emitted at the specific angle θ_{p} as expected in the Kretschmann configuration (Supplementary Note 6). The light power P_{ν}Δν is plotted as a function of the bias voltage for two different wavelengths in Fig. 5a. The inset of Fig. 5a displays the relationship between the light power for the two wavelengths on a loglog plot^{37}. Data points do not fit the blackbody radiation law (solid line in inset) and, as previously mentioned, the Joule heating cannot be responsible for the observed photon emission. As already observed by Lamb and McCarthy^{10}, the light power exhibits a sharp voltage crossover at eV = hc/λ: electrons crossing the tunnel junction relax their energy by emitting photons at a frequency ν ≤ eV/h. This crossover is predicted by both the FDR and the LB theories. However, our data clearly disagree with the FDR (dashed line in Fig. 5a) and are in very good agreement with the LB relation of Eq. (5) (solid line in Fig. 5a). This proves the existence of the traversal time. The LB approach explains the biaspolarity dependence of the light emission which has previously been observed but not justified^{36,38,39}. It also allows one to extract the radiation impedance according to Eq. (4). This gives \({\cal R}\)(λ = 0.9 μm) = 2.5 ± 0.1 mΩ and \({\cal R}\)(λ = 1.3 μm) = 2.3 ± 0.1 mΩ, about a factor of four higher than our rough estimation in the limit \(\nu {\mathrm{/}}\nu _{\mathrm{p}} \ll 1\)^{9}. Unlike the low frequency noise which is bonded by the RC frequency cutoff (1/(2RC) ~ 100 kHz), the spectral power density at optical frequencies involves the radiation impedance \({\cal R}(\nu )\) which does not exhibit any highfrequency cutoff (Supplementary Note 6)^{40}:
where β = tanh(a/δ_{p}) ≃ 0.69, ν_{p} = ω_{p}/2, n = 1.75 is the refractive index of sapphire and the alumina dielectric barrier, d ≃ 2 nm is the thickness of the barrier and Z_{vac} ≃ 376 Ω is the vacuum impedance. This underestimation can not only be attributed to the approximative values of the thickness and the refractive index of the dielectric barrier but also to the interband transition at λ_{inter} = 0.825 μm in aluminum. The 0.1 mΩ uncertainty on radiative resistance \({\cal R}(\nu )\) comes from the fit of the data but one has to keep in mind that an error on the detection efficiency could induce a systematic error of about 20%. We assume here that the coupling between the current fluctuations and the electric field takes place in the insulating barrier. This is justified by the screening of the electric field in the metal. If we only consider the coupling in the electrodes, we should expect a radiation impedance in the μΩ range, three orders of magnitude smaller than the observed one (Supplementary Note 6).
The shot noise spectral density at optical frequencies gives direct information about the current fluctuations on the femtosecond timescale. By using the scattering LB approach for the twodimensional tunnel barrier, we obtain an estimation of the onedimensional traversal time given by the WKB approximation (Methods):
where \(\tilde F\) = S_{ii}(eV, hν)/(eI(V)) is a voltage and frequencydependent Fanolike factor. It then only involves measurable quantities S_{ii}(eV, hν) and I(V) where S_{ii} is deduced from the light power and the value of the radiation impedance previously determined. Figure 5b shows g(eV, hν)/(2πν) as a function of the bias voltage for the two different wavelengths. This expression is only defined for sufficiently high voltage when \(1{\mathrm{/}}4 < \tilde F \le 1{\mathrm{/}}2\). However, the bias voltage has to be kept smaller than 2 V to avoid damaging the junction. The estimation τ_{T} ~ g(2hν, hν)/(2πν) ~ 1.1 fs can thus only be deduced from the measurement at λ = 1.3 μm with eV = 2hν = 1.9 V. This value is in very good agreement with the theoretical onedimensional traversal times calculated for a trapezoidal barrier in the energy range \([\epsilon _{\mathrm{F}},\epsilon _{\mathrm{F}} + h\nu ]\) (Methods).
Discussion
We have experimentally demonstrated that the quantum shot noise spectral density cannot be simply expressed by the universal FDR. If this discrepancy is formally due to both energy and voltage dependence of the transmission \({\cal T}\) (Methods), the key reason is that the charge inside the tunneling barrier fluctuates and this reflects the fact that electrons spend a certain time inside the barrier. To take into account this time spent in the barrier, we have developed a theory using the formalism of LB which also gives insights on the mechanism at the origin of the optical radiation emitted by biased planar tunnel junction. The radiation impedance in the mΩ range is indeed small and appears as a central quantity in the understanding of the poor light emission efficiency of tunnel junctions. This leads us to redefine the efficiency with respect to the dissipated Joule power: \(\eta = {\int}_0^{ + \infty } {\kern 1pt} P_\nu d\nu {\mathrm{/}}(V \times I)\) ~ 4 × 10^{−8}. According to this definition, we can show that the efficiency is now directly related to the radiation impedance: \(\eta \sim \eta _0{\cal R}(eV{\mathrm{/}}h){\mathrm{/}}R_{\mathrm{K}}\) where R_{K} = h/e^{2} ≃ 25.8 kΩ is the quantum of resistance and η_{0} ≃ 0.047 is a constant slightly dependent on the details of the barrier (Supplementary Fig. 12). We emphasize that this definition contrasts with the usual one which is given by the electrontophoton conversion rate. We find the former more appropriate since it reflects the fact that, in metallic tunnel junctions, electrons with energy smaller than bias voltage can contribute to the current. Unlike semiconductors, the lack of a band gap in metals indeed implies that each electron crossing the barrier emits a bunch of photons in a spectral range 0 < ν < eV/h with a radiated spectral power P_{ν} proportional to the current. The emitted light power is then proportional to the Joule power V × I.
Regarding the mechanism of photon emission in metallic tunnel junctions, it is usually attributed to the spontaneous emission in the barrier due to inelastic electron tunneling. Although it has recently been demonstrated in a Van der Waals quantum tunnel junction^{41} that part of the emission processes is due to direct photon emission, usually a twostep process is assumed with tunneling: electrons inelastically excite SPP which then couple to photons. The inelastic tunneling process is characterized by an electrontoplasmon conversion rate which roughly describes the coupling to the electromagnetic environment^{16,42,43,44}. In our case, we give a microscopic description of the tunneling by using the LB scattering approach. It is worth noting that the LB approach which is used here only considers elastic tunneling processes. In this description, the energy relaxation formally takes place in the electrodes and corresponds to electron–hole pair recombination specified by S_{LL}, S_{RR} and S_{LR}^{29}. Nevertheless, by considering the coupling to the electric field only in the dielectric layer, we implicitly assume a relaxation in the tunneling barrier associated with the noise spectral density S_{ii} = (S_{LL} + S_{RR} − 2S_{LR})/4. Thus, our approach is not in contradiction with the inelastic interpretation. More precisely, by only considering the elastic tunneling current to calculate the radiation impedance, we just neglect the feedback of the electromagnetic environment on the tunneling processes. This feedback, known as dynamical Coulomb blockade, is responsible for a correction of the current of the order of the ratio \({\cal R}(\nu ){\mathrm{/}}R_{\mathrm{K}}\)^{45,46,47}. In our case, this effect is negligible, which validates a posteriori our approach using the elastic current.
In conclusion, we have measured the current fluctuations S_{ii} in a metallic tunnel junction in the optical domain and deduced an estimation of the traversal time τ_{T} ~ 1.1 fs. In this regime, S_{ii} may no longer be described by the usual fluctuation dissipation relation because of the energy and voltage dependence of the tunneling transmission. We have shown how this dependence can be incorporated into the LB formalism to ensure the gauge invariance of the I(V) characteristics in the farfromequilibrium regime and thus describe the quantum fluctuations of the current at optical frequencies. This theoretical description is in good agreement with our experimental results, it allows us to give an approximative value of the traversal time by a fittingfree model and sheds light on the estimation of the quantum efficiency of metallic tunnel junction as a light emitter. Our experimental approach demonstrates that optical measurements are a powerful tool to study the quantum electronic transport at high energy (~1 eV). Such measurements extend the range of applicability of mesoscopic electronic transport. The traversal time, which is the average time that the particle spends in the barrier^{48}, could for instance be compared to the average time given by the electric waiting time distribution which has been recently calculated in mesoscopic conductors^{49,50}.
Methods
Planar tunnel junction
In the article, for the sake of simplicity, the LB formalism (Eqs. (2), (3a) and (3b)) is only considered for a 1D single channel of conduction. We demonstrate in the supplemental material that the threedimensional (3D) case can be deduced by summing over all the transversal modes, and the tunnel junction is then equivalent to the parallel association of the \(M = S{\mathrm{/}}\lambda _{\mathrm{F}}^2\) transversal modes of conduction contained in the tunnel junction area S with a mean transmission:
where \(T_{{\mathrm{WKB}}}\left( {\epsilon _{} = \epsilon  \epsilon _ \bot ,eV} \right)\) is the WKB transmission through a 1D homogeneous barrier and λ_{F} the Fermi wavelength. However, we will see in the following that the 3D case can also be obtained by an integration over the longitudinal energy \(\epsilon _{} = \epsilon  \epsilon _ \bot\). Using this integration, we will shown that the Fanolike factor at optical frequencies directly gives the 1D traversal time τ_{T} without any assumption about the dimensionality of the junction.
Shot noise spectral density and the FDR
For a 1D tunneling barrier, the tunneling transmission is related to the traversal time τ_{T} according to (Supplementary Note 4):
Using the scattering LB approach in the tunneling limit (\({\cal T} \ll 1\)), the integration over the longitudinal energy \(\epsilon _{}\) of the 2D versions of Eqs. (3a) and (3b) gives:
where \(\tilde f\left( {\epsilon _{}} \right)\) = \( {\textstyle{{Mk_{\mathrm{B}}T} \over {\epsilon _{\mathrm{F}}}}}{\mathrm{ln}}\left( {f\left( {  \epsilon _{}} \right)} \right)\) is a quasidistribution and corresponds to the integration of \(f\left( {\epsilon _ \bot + \epsilon _{}} \right)\) over \(\epsilon _{}\). Note that no approximations have been used so far and we have chosen here to integrate over \(\epsilon _{}\) instead of \(\epsilon _ \bot\). \(T_{{\mathrm{WKB}}}\left( {\epsilon _{},eV} \right)\) corresponds here to the one dimensional WKB transmission. Then, the FDR reads:
Equations (11) and (12) are equivalent only if \(T_{{\mathrm{WKB}}}\left( {\epsilon _{},eV} \right)\) + \(\sqrt {T_{{\mathrm{WKB}}}\left( {\epsilon _{} \pm h\nu ,eV} \right)}\) = \(2\sqrt {T_{{\mathrm{WKB}}}\left( {\epsilon _{},eV \mp h\nu } \right)}\). The FDR thus holds at finite frequency only if the transmission is of the form (Supplementary Note 5):
This form is only a good approximation of the tunneling transmission at small bias voltage \(eV \ll U\) with:
In this limit, \(\tau _{\mathrm{T}}\) = \(\sqrt {md^2{\mathrm{/}}(2U)}\) and \(h\nu\) ≲ \(eV \ll U\) implies \(\nu \tau _{\mathrm{T}} \ll  ({\mathrm{ln}}{\kern 1pt} {\cal T}_0){\mathrm{/}}8 < 1\) which corresponds to a negligible traversal time.
Traversal time estimation
According to Eqs. (10) and (11) and assuming that τ_{T} is rather constant on the energy scale \([\epsilon _{\mathrm{F}},\epsilon _{\mathrm{F}} + h\nu ]\), the zero temperature limit \(k_{\mathrm{B}}T \ll eV,h\nu\) gives (Supplementary Note 4 + Supplementary Fig. 6):
with Θ the Heaviside step function. Similarly, the LB formula Eqs. (2) and (10) give at the special bias voltages V = ±2hν/e:
Figure 6 shows the product of the distribution function \(\left[ {\tilde f\left( {\epsilon _{}} \right)  \tilde f\left( {\epsilon _{} \pm h\nu } \right)} \right]\) and the transmission \(T_{{\mathrm{WKB}}}(\epsilon , \pm 2h\nu )\) appearing in Eqs. (15) and (16). This distribution function is a broad peak spanning from \(\epsilon _{\mathrm{F}}\) to \(\epsilon _{\mathrm{F}} + h\nu\). Thus, the generalized Fano factor \(\tilde F(eV,h\nu )\) = S_{ii}(eV, hν)/(eI(V)) evaluated at V = ±2hν/e can be expressed in the zero temperature limit as a function of \(\tau _{\mathrm{T}}(\epsilon \sim h\nu {\mathrm{/}}2,eV = 2h\nu )\):
Finally, by inverting Eq. (17), we get:
Note that the estimated traversal time corresponds to the 1D motion through the barrier.
Data availability
The data that support the findings of this study are available from the corresponding author on reasonable request.
References
 1.
Landauer, R. & Martin, T. Barrier interaction time in tunneling. Rev. Mod. Phys. 66, 217–228 (1994).
 2.
Büttiker, M. & Landauer, R. Traversal time for tunneling. Phys. Rev. Lett. 49, 1739–1742 (1982).
 3.
Eckle, P. et al. Attosecond ionization and tunneling delay time measurements in helium. Science 322, 1525–1529 (2008).
 4.
Torlina, L. et al. Interpreting attoclock measurements of tunnelling–times. Nat. Phys. 11, 503 (2015).
 5.
Camus, N. et al. Experimental evidence for quantum tunneling time. Phys. Rev. Lett. 119, 023201 (2017).
 6.
Guéret, P., Marclay, E. & Meier, H. Experimental observation of the dynamical image potential in extremely low GaAs/Al_{x}Ga_{1−x}/GaAs tunnel barriers. Appl. Phys. Lett. 53, 1617–1619 (1988).
 7.
Esteve, D. et al. Observation of the temporal decoupling effect on the macroscopic quantum tunneling of a Josephson junction. Phys. Scr. 1989, 121 (1989).
 8.
Hone, D., Mühlschlegel, B. & Scalapino, D. J. Theory of light emission from small particle tunnel junctions. Appl. Phys. Lett. 33, 203–204 (1978).
 9.
Laks, B. & Mills, D. L. Photon emission from slightly roughened tunnel junctions. Phys. Rev. B 20, 4962–4980 (1979).
 10.
Lambe, J. & McCarthy, S. L. Light emission from inelastic electron tunneling. Phys. Rev. Lett. 37, 923–925 (1976).
 11.
Buret, M. et al. Spontaneous hotelectron light emission from electronfed optical antennas. Nano. Lett. 15, 5811–5818 (2015).
 12.
Schneider, N. L., Schull, G. & Berndt, R. Optical probe of quantum shotnoise reduction at a singleatom contact. Phys. Rev. Lett. 105, 026601 (2010).
 13.
Peters, P.J. et al. Quantum coherent multielectron processes in an atomic scale contact. Phys. Rev. Lett. 119, 066803 (2017).
 14.
Schneider, N. L., Lü, J. T., Brandbyge, M. & Berndt, R. Light emission probing quantum shot noise and charge fluctuations at a biased molecular junction. Phys. Rev. Lett. 109, 186601 (2012).
 15.
Lesovik, G. B. & Loosen, R. On the detection of finitefrequency current fluctuations. Jetp Lett. 65, 295–299 (1997).
 16.
Parzefall, M. et al. Antennacoupled photon emission from hexagonal boron nitride tunnel junctions. Nat. Nanotechnol. 10, 1058–1063 (2015).
 17.
Rogovin, D. & Scalapino, D. Fluctuation phenomena in tunnel junctions. Ann. Phys. 86, 1–90 (1974).
 18.
Lee, H. & Levitov, L. S. Current fluctuations in a single tunnel junction. Phys. Rev. B 53, 7383–7391 (1996).
 19.
Sukhorukov, E. V., Burkard, G. & Loss, D. Noise of a quantum dot system in the cotunneling regime. Phys. Rev. B 63, 125315 (2001).
 20.
Roussel, B., Degiovanni, P. & Safi, I. Perturbative fluctuation dissipation relation for nonequilibrium finitefrequency noise in quantum circuits. Phys. Rev. B 93, 045102 (2016).
 21.
Gabelli, J. & Reulet, B. Dynamics of quantum noise in a tunnel junction under ac excitation. Phys. Rev. Lett. 100, 026601 (2008).
 22.
Basset, J., Bouchiat, H. & Deblock, R. Emission and absorption quantum noise measurement with an onchip resonant circuit. Phys. Rev. Lett. 105, 166801 (2010).
 23.
Parlavecchio, O. et al. Fluctuationdissipation relations of a tunnel junction driven by a quantum circuit. Phys. Rev. Lett. 114, 126801 (2015).
 24.
Martin, T. & Landauer, R. Wavepacket approach to noise in multichannel mesoscopic systems. Phys. Rev. B 45, 1742–1755 (1992).
 25.
Thibault, K., Gabelli, J., Lupien, C. & Reulet, B. Pauliheisenberg oscillations in electron quantum transport. Phys. Rev. Lett. 114, 236604 (2015).
 26.
Christen, T. & Büttiker, M. Gaugeinvariant nonlinear electric transport in mesoscopic conductors. EPL 35, 523 (1996).
 27.
Blanter, Y. & Büttiker, M. Shot noise in mesoscopic conductors. Phy. Rep. 336, 1–166 (2000).
 28.
Simmons, J. G. Generalized thermal JV characteristic for the electric tunnel effect. J. Appl. Phys. 35, 2655–2658 (1964).
 29.
Zamoum, R., Lavagna, M. & Crépieux, A. Nonsymmetrized noise in a quantum dot: Interpretation in terms of energy transfer and coherent superposition of scattering paths. Phys. Rev. B 93, 235449 (2016).
 30.
Nyquist, H. Thermal agitation of electric charge in conductors. Phys. Rev. 32, 110–113 (1928).
 31.
Lü, J.T., Christensen, R. B. & Brandbyge, M. Light emission and finitefrequency shot noise in molecular junctions: from tunneling to contact. Phys. Rev. B 88, 045413 (2013).
 32.
Kretschmann, E. Decay of non radiative surface plasmons into light on rough silver films. comparison of experimental and theoretical results. Opt. Commun. 6, 185–187 (1972).
 33.
Aref, T. et al. Characterization of aluminum oxide tunnel barriers by combining transport measurements and transmission electron microscopy imaging. J. Appl. Phys. 116, 073702 (2014).
 34.
Brinkman, W. F., Dynes, R. C. & Rowell, J. M. Tunneling conductance of asymmetrical barriers. J. Appl. Phys. 41, 1915–1921 (1970).
 35.
Esposito, M., Harbola, U. & Mukamel, S. Nonequilibrium fluctuations, fluctuation theorems, and counting statistics in quantum systems. Rev. Mod. Phys. 81, 1665–1702 (2009).
 36.
Hanisch, M. & Otto, A. Light emission from rough tunnel junctions in UHV. J. Phys. Condens. Matter 6, 9659 (1994).
 37.
Bonnet, I. & Gabelli, J. Probing Planck’s law at home. Eur. J. Phys. 31, 1463 (2010).
 38.
Kirtley, J., Theis, T. N. & Tsang, J. C. Light emission from tunnel junctions on gratings. Phys. Rev. B 24, 5650–5663 (1981).
 39.
Kirtley, J. R., Theis, T. N., Tsang, J. C. & DiMaria, D. J. Hotelectron picture of light emission from tunnel junctions. Phys. Rev. B 27, 4601–4611 (1983).
 40.
Ushioda, S., Rutledge, J. E. & Pierce, R. M. Theory of prismcoupled light emission from tunnel junctions. Phys. Rev. B 34, 6804–6812 (1986).
 41.
Parzefall, M. et al. Light from Van der Waals quantum tunneling devices. Preprint at https://arxiv.org/abs/1804.06163 (2018)
 42.
Persson, B. N. J. & Baratoff, A. Theory of photon emission in electron tunneling to metallic particles. Phys. Rev. Lett. 68, 3224–3227 (1992).
 43.
Berndt, R., Gimzewski, J. K. & Johansson, P. Inelastic tunneling excitation of tipinduced plasmon modes on noblemetal surfaces. Phys. Rev. Lett. 67, 3796–3799 (1991).
 44.
Chen, C., Bobisch, C. A. & Ho, W. Visualization of Fermi’s golden rule through imaging of light emission from atomic silver chains. Science 325, 981–985 (2009).
 45.
Xu, F., Holmqvist, C. & Belzig, W. Overbias light emission due to higherorder quantum noise in a tunnel junction. Phys. Rev. Lett. 113, 066801 (2014).
 46.
Xu, F., Holmqvist, C., Rastelli, G. & Belzig, W. Dynamical coulomb blockade theory of plasmonmediated light emission from a tunnel junction. Phys. Rev. B 94, 245111 (2016).
 47.
Altimiras, C., Portier, F. & Joyez, P. Interacting electrodynamics of short coherent conductors in quantum circuits. Phys. Rev. X 6, 031002 (2016).
 48.
Fertig, H. A. Traversaltime distribution and the uncertainty principle in quantum tunneling. Phys. Rev. Lett. 65, 2321–2324 (1990).
 49.
Albert, M., Haack, G., Flindt, C. & Büttiker, M. Electron waiting times in mesoscopic conductors. Phys. Rev. Lett. 108, 186806 (2012).
 50.
Dasenbrook, D., Hofer, P. P. & Flindt, C. Electron waiting times in coherent conductors are correlated. Phys. Rev. B 91, 195420 (2015).
Acknowledgements
We acknowledge fruitful discussions with M. Aprili, J. Basset, E. BoerDuchemin, J. Estève, JJ Greffet, E. Pinsolle, F. Portier, B. Reulet, I. Safi and P. Simon. We also thank A. Crépieux for useful insight. This work was funded by ANR11JS0400601, Investissements d’Avenir LabEx PALM (ANR10LABX0039PALM) and ANR15CE240020.
Author information
Affiliations
Contributions
J.G. designed the project and conceived the experiment. Both authors, P.F and J.G, fabricated the device, performed the measurements, analyzed the data and interpreted the results. J.G. wrote the manuscript.
Corresponding author
Correspondence to Julien Gabelli.
Ethics declarations
Competing interests
The authors declare no competing interests.
Additional information
Publisher’s note: Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.
Electronic supplementary material
Rights and permissions
Open Access This article is licensed under a Creative Commons Attribution 4.0 International License, which permits use, sharing, adaptation, distribution and reproduction in any medium or format, as long as you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons license, and indicate if changes were made. The images or other third party material in this article are included in the article’s Creative Commons license, unless indicated otherwise in a credit line to the material. If material is not included in the article’s Creative Commons license and your intended use is not permitted by statutory regulation or exceeds the permitted use, you will need to obtain permission directly from the copyright holder. To view a copy of this license, visit http://creativecommons.org/licenses/by/4.0/.
About this article
Cite this article
Février, P., Gabelli, J. Tunneling time probed by quantum shot noise. Nat Commun 9, 4940 (2018). https://doi.org/10.1038/s41467018073696
Received:
Accepted:
Published:
Further reading

A generalized selfconsistent model for quantum tunneling current in dissimilar metalinsulatormetal junction
AIP Advances (2019)

Optical antennas driven by quantum tunneling: a key issues review
Reports on Progress in Physics (2019)

Electron Traversal Times in Disordered Graphene Nanoribbons
Entropy (2019)

Antenna surface plasmon emission by inelastic tunneling
Nature Communications (2019)
Comments
By submitting a comment you agree to abide by our Terms and Community Guidelines. If you find something abusive or that does not comply with our terms or guidelines please flag it as inappropriate.