Can slow recombination in ordered superconductors explain the excess quasiparticle population? Eva Gurra, Douglas A. Bennett, Shannon M. Duff, Michael R. Vissers, Joel N. Ullom Abstract An excess density of quasiparticles is widely observed in superconducting films. This excess causes performance degradation in a variety of superconducting devices, including decoherence in qubits. In this Letter, we evaluate the hypothesis of [ 1 ] that the quasiparticle excess is caused by anomalously slow recombination at low quasiparticle densities due to localization in sub-gap states. We probe the density of states in aluminum and niobium films using current-voltage measurements of tunnel junctions and extract upper bounds on the energy scales of the sub-gap states and gap smearing. With these parameters, we evaluate the recombination times predicted by [ 1 ] and find that slow recombination is not predicted to occur at observed quasiparticle densities in aluminum- and niobium-based superconducting devices. These results suggest that the quasiparticle excess in ordered superconductors is primarily due to non-thermal sources of quasiparticle generation and not slow recombination. Introduction — The performance of superconducting quantum circuits depends on a pristine environment to ensure high coherence and reliability. One bottleneck is the presence of broken Cooper pairs known as quasiparticles. It is widely observed that the quasiparticle density in superconducting devices is higher than what is expected from the equilibrium Bardeen-Cooper-Schrieffer (BCS) theory, which predicts an exponential decrease in the density at low temperatures. Excess quasiparticles are a source of decoherence in Josephson-junction based superconducting qubits [ 2 , 3 , 4 , 5 ] , loss in superconducting resonators [ 6 , 7 , 8 ] , and noise in kinetic inductance detectors (KIDs) [ 9 , 10 , 11 ] . The steady-state quasiparticle population is reached when generation mechanisms are in balance with loss processes. BCS theory predicts the thermal quasiparticle density at finite temperature to be n_{qp}(T)=\nu_{0}\sqrt{2\pi k_{B}T\Delta}\exp{\left(-\Delta/k_{b}T\right)} (1) where \nu_{0} is the two-spin density of states, k_{B} is the Boltzmann constant, and \Delta is the superconducting gap energy. At experimentally achievable temperatures of 10 (50) mK, the expected thermal quasiparticle density in aluminum is \sim 10^{-95}\;\mu\text{m}^{-3} ( \sim 10^{-14}\;\mu\text{m}^{-3} ). In contrast, measured quasiparticle densities in aluminum, using a variety of techniques, are typically 1 – 10s \mu\text{m}^{-3} [ 10 , 6 , 12 , 13 ] and not lower than \sim 10^{-2}\;\mu\text{m}^{-3} [ 3 , 14 , 5 , 15 ] . One possible explanation for the excess quasiparticles is pair-breaking by photons, phonons, or particles with energy larger than 2\Delta that are not in thermal equilibrium with the superconductor. Examples include photons from warmer stages of the experimental apparatus [ 16 , 17 , 18 , 19 , 20 ] and phonons produced in the underlying substrate by background radiation [ 5 , 21 , 22 , 23 , 24 ] . However, observations of the quasiparticle excess persist despite a range of mitigation efforts such as careful filtering of electrical connections to the environment, and known pair-breaking mechanisms such as background radiation have not been shown to be sufficient to explain the excess. It is therefore interesting to consider other explanations for excess quasiparticles. An alternative explanation is that quasiparticle loss rates are slower than the predictions of BCS theory. The dominant loss process, recombination, occurs when two quasiparticles form a Cooper pair with the simultaneous emission of a phonon. Bespalov et al. [ 1 ] predict that recombination can be exponentially suppressed at low quasiparticle densities when quasiparticles are trapped in sub-gap states, which arise due to impurities and disorder in real superconducting films [ 25 , 26 , 27 , 28 ] . Ref. [ 1 ] further hypothesizes that this slow-down in recombination along with the non-equilibrium generation of quasiparticles, for example by background radiation, is sufficient to explain the quasiparticle excess. In this paper, we extract the key parameters needed to evaluate the theory of [ 1 ] and determine the feasibility of experimentally observing slow recombination in two commonly used ordered superconductors: elemental aluminum and niobium. We analyze current-voltage curves of superconducting tunnel junctions and extract the gap smearing, as described by Abrikosov – Gor’kov theory [ 29 ] , and an upper bound for the Lifshitz tail of localized states [ 25 ] which, to the best of our knowledge, have not been characterized in aluminum and niobium. Using these parameters, we calculate the range of quasiparticle densities where slow-recombination theory holds and find that the onset of slow recombination is predicted to occur at densities lower than those observed in various aluminum and niobium devices. Theory of slow recombination — In BCS theory, we have a gapped superconductor where quasiparticle excitations cannot exist inside the gap. Quasiparticles generated by a pair-breaking photon initially have an energy larger than the gap, but they relax back down via phonon emission. Since there are no available states below the gap, they recombine to maintain a steady state population. The recombination rate, \Gamma_{BCS} , is proportional to the quasiparticle density, n_{qp} [ 30 ] , \Gamma_{BCS}(n_{qp})=\bar{\Gamma}n_{qp} (2) where \bar{\Gamma} is a material-specific recombination constant. The theory of slow recombination presented in [ 1 ] considers spatial gap fluctuations caused by imperfections and disorder present in real materials that can generate a tail of states below the gap whose density decays exponentially with energy relative to the gap edge. Now quasiparticles relax back down to the gap edge via phonon emission but have available states below the gap to settle into. When a quasiparticle populates a sub-gap state, it is spatially localized and its decay terminates at an energy \varepsilon_{c} below the gap edge. This energy depends on the gap smearing and characteristic energy decay of the Lifshitz tail which we extract from density of states fits. From \varepsilon_{c} , a localization radius r_{c} is defined that sets the range of quasiparticle densities where slow recombination theory holds (see Appendices A and B). The recombination of these localized quasiparticles depends on the spatial overlap of their wavefunctions. Considering the average distance between quasiparticles, r , there are two limits: (1) the low quasiparticle density limit where r\gg r_{c} , so quasiparticles are further away from each other and recombination is exponentially suppressed, and (2) the high quasiparticle density limit where r\ll r_{c} , so the quasiparticles are spatially close to each other, and we recover the BCS result. The slow recombination rate as a function of quasiparticle density, n_{qp} can be expressed as [ 1 ] , \Gamma_{B}(n_{qp})=D\bar{\Gamma}r_{c}^{-3}\left(\frac{3C_{p}}{4\pi r_{c}^{3}\penalty\ n_{qp}}\right)^{\beta/3}e^{-\left(\frac{3C_{p}}{4\pi r_{c}^{3}\penalty\ n_{qp}}\right)^{1/3}} (3) where D\approx 0.06 , \beta=0.41 , and C_{p}=0.605 . The critical quasiparticle density, n_{qp,c}=5\times 10^{-3}\;r_{c}^{-3} , sets the density range for which slow recombination holds. Therefore, recombination is exponentially suppressed when n_{qp}n_{qp,c} , the BCS result is recovered. Density of states fitting — To evaluate the recombination rate of Eq. 3 , we carefully characterize deviations from the BCS density of states. As we can see in Fig. 1 , in real materials there is both a rounding of the BCS singularity and sub-gap states, captured by the energy scales \varepsilon_{g} and \varepsilon_{t} , respectively. Here we outline our fitting routine which combines Abrikosov-Gor’kov theory [ 29 ] to capture the rounding of the BCS singularity with a Lifshitz tail [ 25 ] to capture the sub-gap states. Figure 1 : Black points represent a typical density of states measurement. Dashed blue curve is the calculated BCS density of states with a hard gap at E = \Delta . Non-idealities in the DOS can be expressed as a broadening of the BCS singularity and sub-gap states below the shifted hard gap at E=E_{g} , that is quantified by Abrikosov and Gor’kov [ 29 ] theory (salmon curve) with energy \varepsilon_{g} , and Lifshitz tail [ 25 ] (green curve) with a characteristic energy decay \varepsilon_{t} , respectively. Figure 2 : Measured current-voltage curves for superconducting tunnel junctions (black points) and fits using AG theory (salmon curve) to extract \varepsilon_{g}/\Delta and two Lifshitz tails (formulation is given in Appendix B) to extract \varepsilon_{t}/\Delta for exponents \xi_{MS} (green curve) and \xi_{LO} (blue curve). (a) Data and fits for an Al – AlOx – AlMn NIS junction at 89 mK. Salmon curve includes a voltage-dependent temperature calculated using a model of NIS junction self-cooling [ 31 ] . (b) Data and fits for a Nb – AlOx – Nb SIS junction at 67 mK. Non-idealities in the BCS density of states (DOS) have been theoretically studied and observed. Abrikosov and Gor’kov (AG) described the effect of magnetic impurities by a pair-breaking parameter \eta [ 29 ] . For small values of \eta , the DOS retains a hard gap but the gap edge shifts to E_{g} due to the rounding of the BCS singularity as can be observed in Fig 1 . The scale of the rounding is \varepsilon_{g}=\Delta-E_{g} . Details of the AG DOS formulation are given in Appendix A. Larkin and Ovchinnikov [ 25 ] later showed that the problem of magnetic impurities is the same as considering fluctuations in the pairing potential. At the mean field level the AG result is recovered. Beyond mean field, rare optimal fluctuations (for example, in the presence of disorder) result in sub-gap states described by a Lifshitz tail. The characteristic decay energy of the tail is \varepsilon_{t} as shown in Fig 1 . The terminal quasiparticle energy \varepsilon_{c} is related to \varepsilon_{t} . Here we make a step towards identifying the effects of these fluctuations in our data. To extract the characteristic decay of the Lifshitz tail, we use a stretched exponential of the form, \nu_{L}(E)\propto\exp{\left[-\left(\frac{E_{g}-E}{\varepsilon_{t}}\right)^{\xi}\right]} (4) where \xi is the characteristic exponent. Larkin and Ovchinnikov [ 25 ] derived an exponent \xi_{LO} by studying optimal fluctuations of \Delta in the Usadel equation close to the mean-field gap edge. Details on the functional form of the Lifshitz tail and the resulting \varepsilon_{c} and r_{c} parameters for both the original formulation in [ 1 ] with \xi_{LO} and the consequences of an alternative formulation of the Lifshitz tail with \xi_{MS} given in [ 32 , 33 ] are described in Appendix B. Thus, we have a DOS which we can express as, \nu(E)=\begin{cases}\nu_{AG}(E,\Delta,\eta),&E\geq E_{g}\\ \nu_{L}(E,E_{g},\varepsilon_{t}),&E