Microresonators fabricated from high-kinetic- Aluminum films

Wenyuan Zhang,∗ K. Kalashnikov,∗ Wen-Sen Lu,∗ P. Kamenov, T. DiNapoli, and M.E. Gershenson Department of Physics and Astronomy, Rutgers University, 136 Frelinghuysen Rd., Piscataway, NJ 08854, USA

We have studied superconducting coplanar- (CPW) fabricated from disor- dered (granular) films of Aluminum. Very high of these films, inherent to disor- dered materials, allows us to implement ultra-short (200 µm at a 5GHz ) and high-impedance (up to 5 kΩ) half-wavelength resonators. We have shown that the intrinsic losses in these resonators at temperatures . 250 mK are limited by to two-level systems in the environment. The demonstrated internal quality factors are comparable with those for CPW res- onators made of conventional superconductors. High kinetic inductance and well-understood losses make these disordered Aluminum resonators promising for a wide range of applications which include kinetic inductance photon detectors and superconducting quantum circuits.

2 I. INTRODUCTION quantum RQ = h/(2e) [14–16]), and the development of superinductor-based protected qubits [17]. We have The development of novel quantum circuits for infor- fabricated resonators with an impedance Z as high as mation processing requires the implementation of ultra- 5 kΩ, ultra-small dimensions and relatively low losses. low-loss microwave resonators with small dimensions The study of the temperature dependences of the res- [1]. Superconducting resonators have become ubiquitous onance frequency fr and intrinsic quality factor Qi at parts of high-performance superconducting qubits [2, 3] different MW excitation levels allowed us to identify res- and kinetic-inductance photon detectors [4]. An impor- onator coupling to two-level systems in the environment tant resource for resonator miniaturization is the kinetic as the primary dissipation mechanism at T . 250 mK; at higher temperatures the losses can be attributed to inductance of superconductors, LK , which can exceed the magnetic (”geometrical”) inductance by orders of magni- thermally excited quasiparticles. tude in narrow and thin superconducting films [4]. High kinetic inductance translates into a high impedance Z of the microwave (MW) elements, slow propagation of II. EXPERIMENTAL DETAILS electromagnetic waves, and small dimensions of the MW resonators. Ultra-narrow wires and thin films of Nb and NbN [4, 6], TiN [7], InOx[8, 9], and granular Al [10] were The standard method for the fabrication of disordered Al films is the deposition of Al at a reduced oxygen studied recently as candidates for high-LK applications. pressure [18, 19]. Such films consist of nanoscale grains Research in high-LK elements also has an important fundamental aspect. According to the Mattis-Bardeen (3 − 4 nm in diameter) partially covered by AlOx. We (MB) theory [8], the kinetic inductance of a thin super- have fabricated the films by DC magnetron sputtering of an Al target in the atmosphere of Ar and O2. Typically, conducting film LK (T = 0) is proportional to the re- −3 the partial pressures of Ar and O2 were 5 × 10 mbar sistance of the film in the normal state, RN , and thus −5 increases with disorder. This theory, however, cannot and (3÷7)×10 mbar, respectively (the fabrication de- be directly applied to strongly disordered superconduc- tails are provided in the Supplementary Materials [20]). tors near the disorder-driven superconductor-to- The films were deposited onto the intrinsic Si substrates transition (SIT). Recent theories predict a rapid decrease at room temperature. By controlling the deposition rate and O2 pressure, the resistivity of the studied films can of the superfluid density near the SIT and the emer- −4 −1 gence of sub-gap delocalized modes that would result in be tuned between 10 Ω·cm and 10 Ω·cm; the pa- enhanced dissipation at microwave [12, 13]. rameters of several representative samples are listed in Thus, the study of the electrodynamics of strongly dis- Table I. ordered superconductors may also contribute to a better The hybrid microcircuits containing the CPW half- understanding of the disorder-driven SIT. wavelength resonators coupled to a CPW transmission arXiv:1807.00210v1 [cond-mat.supr-con] 30 Jun 2018 In this Letter, we present a detailed characteriza- line (TL) have been fabricated using e-beam lithogra- tion of the half-wavelength microwave resonators fabri- phy. As the first step, the 50-Ω TL was fabricated by cated from disordered Aluminum films. Our interest in the e-gun deposition of a 140-nm-thick film of pure Al high-LK films was stimulated by the possibility of fabri- on a pre-patterned substrate and successive lift-off. The cation of superinductors (dissipationless elements with use of pure Al facilitated the with microwave impedance greatly exceeding the resistance the MW set-up and reduced the number of spurious reso- nances (a large number of these is observed if high-Lk films are used for both the TL and resonator fab- rication). After the e-beam lithography, several ∗ These authors contributed equally to this work. half-wavelength disordered Al resonators were fabricated 2 in the openings in the plane. Before each metal ately disordered films (resonators #2 − 4), both the dis- deposition, reactive ion etching was used to remove the sipation and dispersion at T < 0.25 K can be attributed e-beam resist residue from the substrate surface. The to the resonator coupling to the two-level systems (TLS) width of the central strip of the resonators varied be- in the environment, whereas at higher temperatures they tween 0.5 µm and 10 µm, and the strip-ground distance are controlled by the T dependence of the complex con- was fixed at 4 µm. ductivity of superconductors, σ(T ) = σ1(T ) − iσ2(T ) [8]. For the resonator characterization at ultra-low tem- peratures, we used a microwave setup developed for TABLE I. Summary of the measured parameters of AlOx res- the study of superconducting qubits [16, 20]. The res- onators onators were designed with the resonance frequencies w, l, fr, ρ, Tc, LK , Z, fr ≈ 2 − 4 GHz, which allowed us to probe the first three # . of the resonators within the setup frequency µm µm GHz mΩ·cm K nH  kΩ range (2÷12) GHz. Different resonance frequencies of the 1 11.0 1090 2.42 19.2 1.4 2.0 0.6 resonators enabled multiplexing in the transmission mea- 2 7.4 765 4.05 4.2 1.7 1.2 1.1 surements. In order to ensure accurate extraction of the 3 1.4 445 3.69 4.2 1.7 1.2 2.9 4 0.7 265 3.88 9.9 1.75 2.0 5.0 internal quality factor Qi, the resonators were designed with a coupling quality factor Qc of the same order of magnitude as Qi.

A. The resonance frequency analysis III. MICROWAVE CHARACTERIZATION We start the data analysis with the temperature de- The resonators were characterized using a wide range pendence of the relative shift of the resonance frequency δfr(T )/fr0 ≡ [fr(T ) − fr(25mK)]/fr(25mK). Figure of MW power PMW , two-tone (pump-probe) measure- ments, and domain measurements. The resonator 1(a) shows the dependences δfr(T )/fr0 measured for three resonators (#2 − 4) with different width w. The parameters fr, Qi, and Qc were found from the simulta- neous measurements of the and the phase of low-temperature part of δfr(T )/fr0 is governed by the T -dependent TLS contribution to the imaginary part of the transmitted signal S21(f) using the procedure de- scribed in Refs. [2, 3] and Supplementary Materials the complex (T ) = 1(T )+i2(T ). It should be noted that, in contrast to the TLS-related [20]. The kinetic inductance LK of the central conduc- T LS tor of the resonators, which exceeded the magnetic in- losses, the frequency shift δfr is expected to be weakly ductance by several orders of magnitude, was calculated power-dependent [9]. Indeed, the temperature depen- 2 dences measured for the different values of PMW almost as LK = 1/4fr C (the C between the res- onator strip and the ground was obtained in the Sonnet coincide; this simplifies the analysis and reduces the num- ber of fitting parameters. The low-temperature part of simulations). The parameters of several representative T LS resonators are listed in Table I. δfr (T ) is well described by the following equation [4]: " # The measured sheet kinetic inductance LK ≈ 2 nH/ T LS       δfr (T ) Vf δ0 1 1 hfr hfr is similar to that reported for granular Al films in Ref. [23] = Ψ< + − ln . f π 2 2πi k T k T and TiN in Ref. [24], and exceeds by a factor-of-2 L r0 B B K realized for ultra-thin disordered films of InOx [8, 25]. (1) For the disordered Al films with ρ < 10 mΩ·cm, L is K Here Ψ<(x) is the real part of the complex digamma func- in good agreement with the result of the MB theory [8], tion, the TLS participation ratio Vf is the stored L (T = 0) = R /π∆(0), where ∆(0) is the BCS en- K ~  in the TLS-occupied volume normalized by the total en- ergy gap at T = 0 K. Deviations from this behavior, ob- ergy in the resonator, and the loss tangent δ0 charac- served for highly disordered film (e.g., resonator #1), will terizes the TLS-induced microwave loss in weak electric be discussed in a seperate paper [26]. Very large values of fields at low temperatures kBT  hfr. The product L allowed us to realize the K Vf δ0 is the only fitting parameter, its values are listed Z = pL /C as high as 5 kΩ for the resonators with K in Table II. The obtained values of Vf δ0 are close to narrow (w = 0.7 µm) central strips. The speed of prop- that found for Al-based [9] and AlOx-based resonators agation of the electromagnetic waves in such resonators [24, 28]. Note that resonator #4 demonstrates the most does not exceed 1% of the speed of in free space; ac- pronounced increase of fr(T ) with temperature due to cordingly, their length is two orders of magnitude smaller the stronger electric fields and a larger participation ra- than that for the conventional CPW resonators with the tio characteristic of the high-Z resonators [29]. impedance Z = 50 Ω. At T > 0.25 K, fr rapidly drops due to the decrease of To identify the physical mechanisms of losses in the res- the superfluid density. The dependences δfr(T ) over the onators, we measured the dependences of fr and Qi on whole studied T range can be described as the temperature (T = 25 ÷ 450 mK) and the microwave δf (T )/f = δf T LS(T )/f + δf MB(T )/f (2) power PMW . Below we show that in the case of moder- r r0 r r0 r r0 3

T LS FIG. 1. The temperature dependences of resonance frequency shift δfr (T )/fr0 (a) and the internal quality factor Qi (b) for the resonators #2 − 4 measured atn ¯ ≈ 1(5) andn ¯  1(4). The fitting curves correspond to Eq. (2) and Eq. (7), respectively.

p where is observed for the microwave currents√ I∗ = 2P∗/Z which scale approximately as I / Q [33], where I δf MB(T ) 1σ (T ) − σ (25mK) dp l dp r = 2 2 (3) is the Ginzburg-Landau depairing current in the central fr0 2 σ2(25mK) strip [20]. The power-dependent intrinsic losses can be attributed is the resonance shift due to the T -induced break of to the resonator coupling to the TLS with the Lorentzian- Cooper pairs and subsequent increase of the kinetic in- shaped distribution ductance, calculated in the thin film limit [9]. The only MB free parameter in δf (T )/fr0 is the gap energy ∆(0), 1 r g(E ) ∼ , (4) which can be found by fitting of the high-T portion of T LS 2 2 (ET LS − hfr) + (~/τ2) δfr(T )/fr0 [Eq. (2)]; the measured ratio ∆(0)/Tc is about 10% greater than the BCS value of 1.76kB, which where ET LS is the energy of TLS and τ2 is its dephasing is consistent with previously reported data [30]. time [34]. Once the MW power PMW reaches some char- acteristic level P and the Rabi frequency of the driven √ c √ TLS ΩR ∼ PMW exceeds the relaxation rate 1/ τ1τ2, B. The quality factor analysis the population of the excited TLS increases, and the amount of energy that the TLS with fT LS ≈ fr can ab- We now proceed with the analyses of losses. We sorb from the resonator decreases. Thus, the high PMW observed the enhancement of the internal quality fac- ”burns the hole” in the density of states (DoS) of dis- tor Qi with increasing the average number of photons sipative TLS. The width of the ”hole” is κ/2πτ2, the 2 2 power-dependent factor can be written as in the resonators,n ¯ = 2PMW Ql /(Qchfr ) [31], where −1 Ql = (1/Qi + 1/QC ) is the loaded quality factor. The s dependences Q (¯n) for three resonators with different w  n¯ β i κ = 1 + , (5) measured at the base temperature ≈ 25 mK are shown in nc Fig. 2. Similar behavior of Qi(¯n) have been observed for many types of CPW superconducting resonators (see, e.g. wheren ¯ and nc correspond to PMW and Pc, respectively. [4, 32] and references therein), including the resonators Note that the exponent β is known to be dependent on based on disordered Al films [10, 23]. Note that the in- the electric field distribution in a resonator [35], and the crease of Qi with the input MW power PMW is limited characteristic power nc increases with temperature by or- by the resonance by bifurcation at PMW > P∗. ders of magnitude due to a strong T -dependence of τ1 and 4 For the resonators with Ql & 10 the onset of bifurcation τ2 [36]. Taking into account the TLS at high 4

TABLE II. Summary of the fitting parameters −4 −3 # ∆(0)/kB Tc β Vf δ0 · 10 nc(0)·10 2 1.96 0.60 1.4 50 3 1.98 0.55 4.8 1.6 4 1.88 0.38 6.7 0.23

when we monitored the second and applied the pump signal at the first harmonic. Since the resonator coupling to the pump signal varies by several orders of magnitude within the detuning range 0÷10 MHz, it is more informative to analyze Qi as a func- tion of the average number of the ”pump” photons in the 2 2 2 resonator,n ¯p = Pp(1−|S21(fp)| −|S11(fp)| )/hfp , where S21 and S11 = 1−S21 are the transmission and reflection FIG. 2. The dependences Qi(¯n) at T ≈ 25 mK for the res- at the pump frequency, respectively. The de- onators with different widths. Solid curves represent the the- pendence Q on the detuning ∆f for a fixedn ¯ ≈ 1000 is oretical fits of the quality factor governed by TLS losses [Eq. i p (5), see the text for details]. depicted in Fig. 3(b). The resonance behavior of Qi(∆f) is expected since only a narrow TLS band [Eq. (4)] con- tributes to dissipation: the ”hole” extension in the DoS temperature, the power dependence of the TLS-related is limited by ∼ κ/τ2 around the pump frequency. Indeed, part of the loss tangent can be expressed as follows [29]: using the approach developed in [37], one can obtain the following expression:   Vf δ0 hfr δT LS(¯n, T ) = tanh . (6) κ 2kBT  2  (κ/2πτ2) Qi(∆) = Q0 1 + 2 2 , (8) By fitting the experimental data with Eq. (6) we found ∆f + κ(1/2πτ2) β and nc, the obtained parameters are listed in Table II. We found that larger values of β correspond to wide where Q0 is the off-resonance quality factor, and in- troduced by Eq. (5) factor κ might be calculated as strips, and the extracted nc scales as the square of the electric field on the surface of the resonator. The details κ = Qmax/Q0. The dephasing time is the only fitting of the fitting procedure can be found in Supplementary parameter and it is found to be τ2 ≈ 60 ns. This re- Materials [20]. sult agrees with the measurements of the dephasing time The experimental dependences Q (T ) measured for for individual TLS in amorphous Al2O3 tunnel barrier in i Josephson junctions [38]. resonators #2 − 4 atn ¯ w 1 andn ¯  1 [Fig. 1(b)] are well described by the sum of the TLS contribution [Eq. By application of the MW pulses at the pump fre- quency, we observed that the characteristic time at which (6)] and the MB term δMB = σ1(T )/σ2(T ) [9]: Qi varies with Pp does not exceed 36 ms (see Supplemen- −1 Qi(T ) = {δT LS(T, β, nc,Vf δ0) + δMB[T, ∆(0)]} . (7) tary Materials [20] for details). For several resonators we have observed the telegraph in the resonance fre- The agreement of measured Qi with the prediction of Eq. quency on the time scale of 1 − 10 s. This noise can be (7) over the whole measured temperature range proves attributed to interactions of the resonators with a small that the losses in the developed resonators are limited by number of strongly coupled TLS. the sum of TLS and MB terms.

IV. SUMMARY C. The two-tone and time-domain measurements In conclusion, we have fabricated CPW half- We obtained an additional information on the TLS- wavelength resonators made of strongly disordered Al related dissipation by performing the pump-probe exper- films. Because of the very high kinetic inductance of iments in which Qi was measured at a low-power (n w 1) these films, we were able to significantly reduce the length probe signal while the power Pp of the pump signal at the of these resonators, down to ∼ 1% of that of conven- frequency fp was varied over a wide range. Figure 3(a) tional CPW resonators with a 50 Ω impedance. Due to shows the dependences Qi(Pp) measured at different de- ultra-small dimensions and relatively low losses at mK tuning values ∆f = fp −fr = 0, ±1 MHz, and ±10 MHz. temperatures, these resonators are promising for the use Note that we have not observed any changes in Qi when in quantum superconducting circuits operating at ultra- the pump signal was applied at the second and third har- low temperatures, especially for the applications that re- monics of the resonator. Also, Qi was Pp-independent quire numerous resonators, such as multi-pixel MKIDs 5

FIG. 3. (a) The dependences of Qi for resonator #1 on the pump tone power Pp for several values of detuning ∆f between resonance and pump frequencies. (b) The values of Qi measured versus detuning ∆f at a fixed number of the pump tone photons in the resonatorn ¯p ≈ 1000.

p [4, 33]. The high impedance Z = LK /C of the nar- tion, such as substrate trenching (see [41] and references row resonators can be used for effective coupling of spin within) and increasing the gap between the center con- qubits [39, 40]. The high resonator impedance imposes ductor and the ground plane [35]. The evidence for that limitations on the strength of resonator coupling to the was provided by the results of Ref. [23] obtained for the . For the studied CPW resonators with modified three-dimensional structures based Z ∼ 5 kΩ, the strongest realized coupling (when half of on disordered Al films. It is also worth mentioning that the resonator length was used as the element of capac- the losses can be reduced using TLS saturation by the itive coupling to the transmission line) corresponded to microwave signal outside of the resonator bandwidth but 4 Qc ∼ 10 . On the other hand, for many applications, within the TLS spectral diffusion range. A fundamental such as large MKID arrays that require a high loaded Q issue pertinent to all strongly disordered superconduc- factor, this should not be a limitation. tors is the development of a better understanding of the We have shown that the main source of losses in these impedance of superconductors near the disorder-driven resonators at T  Tc is the coupling to the resonant two- SIT. This issue requires further research, and the mi- level systems. A comparison of our results with those of crowave experiments with the resonators made of disor- the other groups shows that the obtained Qi values, in- dered Al and other disordered materials demonstrating creasing from (1÷2)×104 in the single-photon regime to the SIT may shed light on the nature of this quantum 3×105 at high microwave power, are typical for the CPW transition. superconducting resonators with similar TLS participa- tion ratios. This implies that the disorder in Al films does not introduce any additional, anomalous losses. Most likely, the relevant TLS are located near the edges of the ACKNOWLEDGMENTS central resonator strip either in the native oxide on the Si substrate surface or in the amorphous oxide covering the films. Further increase of Qi can be achieved by This work was supported by the NSF award 1708954 the methods aimed at the reduction of surface participa- and the ARO award W911NF-17-C-0024.

[1] M. H. Devoret and R. J. Schoelkopf, “Superconducting frey, Y. Chen, Y. Yin, B. Chiaro, J. Mutus, C. Neill, circuits for quantum information: An outlook,” Science P. O’Malley, P. Roushan, J. Wenner, T. C. White, A. N. 339, 1169–1174 (2013). Cleland, and John M. Martinis, “Coherent josephson [2] H. Paik, D. I. Schuster, L. S. Bishop, G. Kirchmair, qubit suitable for scalable quantum integrated circuits,” G. Catelani, A. P. Sears, B. R. Johnson, M. J. Reagor, Phys. Rev. Lett. 111, 080502 (2013). L. Frunzio, L. I. Glazman, S. M. Girvin, M. H. De- [4] J. Zmuidzinas, “Superconducting Microresonators: voret, and R. J. Schoelkopf, “Observation of High Co- Physics and Applications,” Annu. Rev. Condens. Matter herence in Josephson Junction Qubits Measured in a Phys. 3, 169–214 (2012). Three-Dimensional Circuit QED Architecture,” Phys. [4] M. Tinkham, Introduction to superconductivity (Dover Rev. Lett. 107, 240501 (2011). Publications, 2004). [3] R. Barends, J. Kelly, A. Megrant, D. Sank, E. Jef- [6] D. Niepce, J. Burnett, and J. Bylander, “High Kinetic 6

Inductance NbN Nanowire Superinductors,” (2018), scattering data under noise in microwave resonators,” arXiv:1802.01723. Rev. Sci. Instrum. 86, 024706 (2015). [7] P. C. J. J. Coumou, E. F. C. Driessen, J. Bueno, [23] L. Gr¨unhaupt,N. Maleeva, S. T. Skacel, M. Calvo, F. C. Chapelier, and T. M. Klapwijk, “Electrodynamic re- Levy-Bertrand, A. V. Ustinov, H. Rotzinger, A. Monfar- sponse and local tunneling spectroscopy of strongly dis- dini, G.i Catelani, and I. M. Pop, “Quasiparticle dynam- ordered superconducting TiN films,” Phys. Rev. B 88, ics in granular aluminum close to the superconductor to 180505 (2013). insulator transition,” (2018), arXiv:1802.01858. [8] O. Dupr´e,A. Benoˆıt,M. Calvo, A. Catalano, J. Goupy, [24] J. T. Peltonen, P. C. J. J. Coumou, Z. H. Peng, T. M. C. Hoarau, T. Klein, K. Le Calvez, B. Sac´ep´e,A. Monfar- Klapwijk, J. S. Tsai, and O. V. Astafiev, “Hybrid rf dini, and F. Levy-Bertrand, “Tunable sub-gap radiation SQUID qubit based on high kinetic inductance,” (2017), detection with superconducting resonators,” Supercond. arXiv:1709.09720. Sci. Technol. 30, 045007 (2017). [25] O. V. Astafiev, L. B. Ioffe, S. Kafanov, Yu. A. Pashkin, [9] S. E. de Graaf, S. T. Skacel, T. H¨onigl-Decrinis, K. Yu. Arutyunov, D. Shahar, O. Cohen, and J. S. Tsai, R. Shaikhaidarov, H. Rotzinger, S. Linzen, M. Ziegler, “Coherent quantum phase slip,” Nature 484, 355–358 U. H¨ubner,H.-G. Meyer, V. Antonov, E. Il’ichev, A. V. (2012). Ustinov, A. Ya. Tzalenchuk, and O. V. Astafiev, “Charge [26] K. Kalashnikov, Wenyuan Zhang, Wen-Sen Lu, P. Ka- quantum interference device,” Nat. Phys. 14, 590–594 menov, T. DiNapoli, and M. E. Gershenson, (unpub- (2018). lished) . [10] H. Rotzinger, S. T. Skacel, M. Pfirrmann, J. N. Voss, [9] J. Gao, The physics of superconducting microwave res- J. M¨unzberg, S. Probst, P. Bushev, M. P. Weides, A. V. onators, Ph.D. thesis (2008). Ustinov, and J. E. Mooij, “Aluminium-oxide wires for [28] J. Wenner, R. Barends, R. C. Bialczak, Y. Chen, J. Kelly, superconducting high kinetic inductance circuits,” Super- E. Lucero, M. Mariantoni, A. Megrant, P. J.J. O’Malley, cond. Sci. Technol. 30, 025002 (2017). D. Sank, A. Vainsencher, H. Wang, T. C. White, Y. Yin, [8] D. C. Mattis and J. Bardeen, “Theory of the Anoma- J. Zhao, A. N. Cleland, and J. M. Martinis, “Surface lous Skin Effect in Normal and Superconducting Metals,” loss simulations of superconducting Phys. Rev. 111, 412–417 (1958). resonators,” Appl. Phys. Lett. 99, 99–101 (2011). [12] M. V. Feigel’man and L. B. Ioffe, “Microwave Proper- [29] J. M. Sage, V. Bolkhovsky, W. D. Oliver, B. Turek, and ties of Superconductors Close to the Superconductor- P. B. Welander, “Study of loss in superconducting copla- Insulator Transition,” Phys. Rev. Lett. 120, 037004 nar waveguide resonators,” J. Appl. Phys. 109, 063915 (2018). (2011). [13] M. Swanson, Y. L. Loh, M. Randeria, and N. Trivedi, [30] U. S. Pracht, N. Bachar, L. Benfatto, G. Deutscher, “Dynamical Conductivity across the Disorder-Tuned E. Farber, M. Dressel, and M. Scheffler, “Enhanced Superconductor-Insulator Transition,” Phys. Rev. X 4, Cooper pairing versus suppressed phase coherence shap- 021007 (2014). ing the superconducting dome in coupled aluminum [14] A. J. Annunziata, D. F. Santavicca, L. Frunzio, G. Cate- nanograins,” Phys. Rev. B 93, 100503 (2016). lani, M. J. Rooks, A. Frydman, and Daniel E Prober, [31] A. Bruno, G. de Lange, S. Asaad, K. L. van der Enden, “Tunable superconducting nanoinductors,” Nanotechnol- N. K. Langford, and L. DiCarlo, “Reducing intrinsic loss ogy 21, 445202 (2010). in superconducting resonators by surface treatment and [15] V. E. Manucharyan, J. Koch, L. I. Glazman, and M. H. deep etching of silicon substrates,” Appl. Phys. Lett. 106 Devoret, “Fluxonium: single cooper-pair circuit free of (2015). charge offsets.” Science 326, 113–6 (2009). [32] A. Megrant, C. Neill, R. Barends, B. Chiaro, Y. Chen, [16] M. T. Bell, I. A. Sadovskyy, L. B. Ioffe, A. Yu. Ki- L. Feigl, J. Kelly, E. Lucero, M. Mariantoni, P. J. J. taev, and M. E. Gershenson, “Quantum Superinduc- O’Malley, D. Sank, A. Vainsencher, J. Wenner, T. C. tor with Tunable Nonlinearity,” Phys. Rev. Lett. 109, White, Y. Yin, J. Zhao, C. J. Palmstrøm, J. M. Martinis, 137003 (2012). and A. N. Cleland, “Planar superconducting resonators [17] M. T. Bell, W. Zhang, L. B. Ioffe, and M. E. Gershenson, with internal quality factors above one million,” Appl. “Spectroscopic Evidence of the Aharonov-Casher Effect Phys. Lett. 100, 113510 (2012). in a Cooper Pair Box,” Phys. Rev. Lett. 116, 107002 [33] L. J. Swenson, P. K. Day, B. H. Eom, H. G. Leduc, (2016). N. Llombart, C. M. McKenney, O. Noroozian, and [18] G. Deutscher, H. Fenichel, M. Gershenson, E. Gr?nbaum, J. Zmuidzinas, “Operation of a titanium nitride su- and Z. Ovadyahu, “Transition to zero dimensionality perconducting microresonator detector in the nonlinear in granular aluminum superconducting films,” J. Low regime,” J. Appl. Phys. 113, 104501 (2013). Temp. Phys. 10, 231–243 (1973). [34] D. P. Pappas, M. R. Vissers, D. S. Wisbey, J. S. Kline, [19] K. C. Mui, P. Lindenfeld, and W. L. McLean, “Lo- and J. Gao, “Two Level System Loss in Superconducting calization and -interaction contributions to the Microwave Resonators,” IEEE Trans. Appl. Supercond. magnetoresistance in three-dimensional metallic granu- 21, 871–874 (2011). lar aluminum,” Phys. Rev. B 30, 2951–2954 (1984). [35] H. Wang, M. Hofheinz, J. Wenner, M. Ansmann, R. C. [20] “Supplementary Materials,”. Bialczak, M. Lenander, E. Lucero, M. Neeley, A. D. [2] M. S. Khalil, M. J. A. Stoutimore, F. C. Wellstood, and O’Connell, D. Sank, M. Weides, A. N. Cleland, and K. D. Osborn, “An analysis method for asymmetric res- J. M. Martinis, “Improving the coherence time of super- onator transmission applied to superconducting devices,” conducting coplanar resonators,” Appl. Phys. Lett. 95, J. Appl. Phys. 111, 054510 (2012). 233508 (2009). [3] S. Probst, F. B. Song, P. A. Bushev, A. V. Ustinov, [36] J. Goetz, F. Deppe, M. Haeberlein, F. Wulschner, C. W. and M. Weides, “Efficient and robust analysis of complex Zollitsch, S. Meier, M. Fischer, P. Eder, E. Xie, K. G. 7

Fedorov, E. P. Menzel, A. Marx, and R. Gross, “Loss The reactive DC sputtering of disordered Al was then mechanisms in superconducting thin film microwave res- initiated by introducing 1 sccm O2 and 115 sccm Ar gas onators,” J. Appl. Phys. 119, 015304 (2016). mixture from two independent feedback-controlled mass [37] T. Capelle, E. Flurin, E. Ivanov, J. Palomo, M. Rosticher, flow meters (MicroTrakTM and SmartTrakTM). During S. Chua, T. Briant, P.-F. Cohadon, A. Heidmann, T. the sputtering process, typical partial pressures of Ar and Jacqmin, and S. Deleglise, “Energy relaxation properties O were 5 × 10−3 mbar and (3 ÷ 7) × 10−5 mbar, respec- of a microwave resonator coupled to a pumped two-level 2 system bath,” (2018), arXiv:1805.04397. tively. After the second lift-off, the chip was installed in [38] Y. Shalibo, Y. Rofe, D. Shwa, F. Zeides, M. Neeley, J. M. the sample holder by wire bonding. Martinis, and N. Katz, “Lifetime and Coherence of Two- Level Defects in a Josephson Junction,” Phys. Rev. Lett. 105, 177001 (2010). [39] N. Samkharadze, A. Bruno, P. Scarlino, G. Zheng, D. P. DiVincenzo, L. DiCarlo, and L. M. K. Vandersypen, “High Kinetic Inductance Superconducting Nanowire Resonators for Circuit QED in a ,” Phys. Rev. Appl. 5, 044004 (2016). [40] A. Stockklauser, P. Scarlino, J. V. Koski, S. Gasparinetti, C. K. Andersen, C. Reichl, W. Wegscheider, T. Ihn, K. Ensslin, and A. Wallraff, “Strong Coupling Cavity QED with Gate-Defined Double Quantum Dots Enabled by a High Impedance Resonator,” Phys. Rev. X 7, 011030 (2017). FIG. S1. (a) Microphotograph of a portion of the half- [41] G. Calusine, A. Melville, W. Woods, R. Das, C. Stull, wavelength resonator capacitively coupled to the coplanar V. Bolkhovsky, D. Braje, D. Hover, D. K. Kim, waveguide transmission line. Light green - Al ground plane X. Miloshi, D. Rosenberg, A. Sevi, J. L. Yoder, E. Dauler, and the central conductor of the transmission line, green - sil- and W. D. Oliver, “Analysis and mitigation of interface icon substrate, black - the central strip of the resonator made losses in trenched superconducting coplanar waveguide of strongly disordered Al. (b) Several resonators with differ- resonators,” Appl. Phys. Lett. 112, 062601 (2018). ent resonance frequencies coupled to the transmission line.

SUPPLEMENTARY MATERIALS II. MEASUREMENT SETUP I. FABRICATION OF MICROWAVE RESONATORS A. Microwave setup

All microwave (MW) resonators studied in this work All measurements were performed in the BlueForsTM consisted of two parts. First, the 50- coplanar MW BF-SD250 dilution refrigerator with a base temperature transmission line was formed on an intrinsic Si substrate of ∼25mK. To reduce stray magnetic fields, a µ-metal by electron beam deposition of a 140-nm-thick film of shield was installed outside of the cryostat. We used the pure Al through a lift-off mask, which comprised of a microwave measurement setup (Fig. S2) developed for 300-nm-thick e-beam resist (the top layer) and a 150-nm- the research in superconducting qubits; it was described thick copolymer (the bottom layer). After the deposition in our previous publication [S1]. The setup enabled the of the bilayer resist and its patterning with e-beam lithog- resonator testing over a wide range of MW power, includ- raphy, the sample was placed in a reactive ion etching ing the single-photon population regime, the two-tone system and etched with 75 mbar O2 plasma at a power (pump-probe) and time domain measurements. of 30 watts for 30 to remove any resist residue The probe signal at f and the pump signal at fp, gen- from the substrate surface. The use of this pure Al trans- erated by two microwave synthesizers, were coupled to mission line facilitated the impedance matching with the the input of the cryostat through directional couplers. MW tract and eliminated spurious resonances. After the Depending on the experiment performed, the pump sig- second e-beam lithography with alignment precision bet- nal could be pulsed using an internal RF of the ter than 0.5 µm, several half-wavelength disordered Al microwave synthesizer. Attenuators and low-pass filters resonators were fabricated on the same substrate by re- were installed in the microwave input line to prevent leak- active DC magnetron sputtering in a vacuum system with age of thermal radiation into the resonator. The sig- the base pressure of < 1 × 10−6mbar (Fig. S1). The dis- nal, after passing the sample, was amplified by a cryo- ordered films were deposited by sputtering of a 6N-purity genic high-electron mobility transistor (HEMT amplifier Al target onto Si substrates held at room temperature. In Caltech CITCRYO 1-12, 35 dB gain between 1 and 12 order to improve reproducibility, prior to the disordered GHz) and two 30dB room-temperature amplifier. Two Al deposition the target was pre-cleaned in a pure Ar- cryogenic Pamtech isolators (each provides 18dB isola- plasma by sputtering at a rate of 0.6 nm/s for 5 minutes. tion between 3 and 12 GHz) were anchored at the base 8

B. DC setup

On the same resonator chip, we also patterned Hall bars to measure critical currents for the disordered Alu- minum films. The critical currents were measured using an Arbitrary Waveform Generator (Tektronix AFG3252) and HP 34401A (see Fig. S3).

III. THE PROCEDURE OF EXTRACTING THE QUALITY FACTORS

The magnitude and phase of the transmitted signal S21 have been used to extract the quality factors Ql, QC , and Qi and the resonance frequency fr. Typically, an asymmetry in the coupling of a resonator to the input and output ports results in deviation of the resonator re- sponse from a symmetric Lorentzian function [S2]. If the coupling between the resonator and the transmission line is weak, the frequency dependence S21(f) near the reso- nance frequency fr is described by the following equation [S2, S3]:

iφ iα −2iπfτ 1 − (Ql/|Qc|e ) S21(f) = ae e . (9SM) 1 + 2iQl(f/fr − 1)

The phase delay τ can be found from the value of d[Arg(S21)]/df measured over a range of f away from the resonance. All other parameters in Eq. (9SM) have FIG. S2. Schematics of the resonator measurement setup been determined similar to the iteration procedure de- scribed in Ref. [S3]. We first selected the initial val- ues of unknown parameters in Eq. (9SM), and ran a temperature to reduce the 5K noise from the HEMT am- multi-variable nonlinear fitting procedure for the entire plifier. The amplified signal was downconverted to the model. The output of the nonlinear fit was used to ob- intermediate frequency (IF) fIF = |f − fLO| ≈ 30MHz tain the final values of unknown parameters and the error using mixer M1 with the local oscillator signal fLO. bars. The parameter initialization procedure was as fol- The IF signal was digitized using the card AlazarTech lows. After elimination of the phase delay e−2iπfτ , the ATS 9870 at 1GS/s. The magnitude and phase of data S21(f) formed a circle on the IQ-plane [Fig. S4(a)]. iα the signal S21 was obtained by digital demodulation as The prefactor ae corresponds to the center of this cir- q cle. For the normalized circle S∗ = S (f)/aeiα−2iπfτ , a = (ha2(t) sin2(2πft)i + ha2(t) cos2(2πft)i) and φ = 21 21 the angle between the off-resonance points and the I- 2 2 2 2 arctan(ha (t) sin (2πft)i/ha (t) cos (2πft)i) − φ0 (here axis corresponds to φ, and the circle diameter corre- h...i stands for the time averaging over integer number sponds to the ratio of Ql/|Qc| [Fig. S4(b)]. Next we of periods, typically 106). The reference phase φ was ∗ 0 translated S21 so that the circle center coincided with provided by mixer M2. the origin. Ql can then be obtained from fitting the ∗ phase of the translated S21, θ, versus frequency with θ = θ0 + arctan[2Ql(1 − f/fr)] [see Fig. S4(c)]. Fig- ures S4(d,e) show the experimental data with the result of fitting.

IV. CRITICAL CURRENTS OF NARROW DISORDERED FILMS

To calculate the Ginzburg-Landau depairing current Idp(0) for strongly disordered Al films at T  TC , we used the equation for the critical supercurrent density FIG. S3. Schematics of the DC setup √ jc = 2/(3 3)(e~ns)/(meξ) [S4]. The concentration of 9

TABLE S3. Summary of predicted and measured depairing currents

4 ρ W I∗ Idp(0) I√∗ # Ql · 10 mΩ·cm µm nA µA Idp/ Ql 1 3.7 19.2 11 90 110.2 0.16 2 1.9 4.2 7.4 240 52.9 0.63 3 3.9 4.2 1.3 30 12.6 0.47 4 8.4 9.9 0.8 20 4.6 1.26

V. DETAILS OF fr(T ) AND Qi(T ) FITTING

To identify the dominant mechanisms of losses in the studied resonators, we have analyzed the experimental dependences fr(T ) and Qi(T ) on the basis of the theory of two-level systems [S7] and the Mattis-Bardeen theory of the complex impedance of superconductors [S8]. The losses due to the real part of the complex impedance of superconductors, σ = σ1 − iσ2, can be esti- mated using the Mattis-Bardeen theory. In the thin film FIG. S4. Fitting procedure. (a) Blue and red points cor- limit [S9]: respond to the transmission measured before and after the phase delay is removed, respectively. After removing the δMB(T ) = σ1(T )/σ2(T ), phase delay, the data form a circle on the IQ plane with the iα ∗ where center at ae . (b) Normalized transmission S21 on the com- ∗ plex plane. The angle between the center of the S circle and ∞ 21 Z the real axis corresponds to φ. (c) The phase versus frequency σn σ1(T ) = [f(E) − f(E + hν)]× (blue points) fitted with θ = θ0 + arctan[2Ql(1 − f/fr)] (red hν curve). (d,e) Measured data (blue points) and the fit with ∆ (11SM) Eq. (9SM) (red curve). E2 + ∆2 + Ehν √ dE, E2 − ∆2p(E + hν)2 − ∆2

Cooper pairs ns can be found either from the measured kinetic inductance per square L , or from the result ∆ K σ Z of the Mattis-Bardeen theory L = m /(2e2n t) = σ (T ) = n [1 − 2f(E + hν)]× K e s 2 hν ( R )/π∆ where t is the film thickness. The supercur- ~  ∆−hν (12SM) rent density is uniform over the cross section of a super- 2 2 2 E + ∆ + Ehν conducting film provided that the film width W < λ /t, √ p dE, where λ is the London penetration length. This condition E2 − ∆2 (E + hν)2 − ∆2 is satisfied for all studied films. Thus, one can estimate f(E) is Fermi-Dirac distribution function, ∆(0) is the I as c energy gap. The temperature-dependent shift of the res- onance frequency fr is given in the main text by Eq. (3). Idp(0) = jc · (W t) Since the frequency shift dfr(T ) does not depend on the 1 π∆ = √ W MW power, the fitting procedure included the following eR ξ(0) (10SM) 3 3  steps: kBTc ≈ 1.07 W. • fitting the dfr(T ) dependence with only two free eR ξ(0)  parameters: ∆(0) (which controls the behavior of MB δfr (T ) term at T > 300 mK), and the product The coherence length ξ(0) can be found from the data of the participation ratio and the material loss tan- on the upper critical magnetic field for granular Al films, gent, Vf δ0 (which governs the rising part of dfr(T )); BC2 ≈ 4T [S5, S6]. This yields an estimate ξ(0) = p • finding the index β from the linear part of Qi(¯n) at Φ0/(2πBC2) ≈ 10 nm. The data in Table S3 show p T = 25mK plotted on the double-log scale. that the values of the microwave current I∗ = 2P∗/Z, which corresponds to the onset of strong nonlinearity of • knowing Vf δ0, β, ∆(0), one can find the low- the resonator response, are of√ the same order of mag- temperature characteristic power (i.e. the number nitude as the current Idp(0)/ Ql corresponding to the of photons nc(0)) using Qi(¯n) measured at the base bifurcation threshold. temperature T = 25mK; 10

• finding nc(T ) by fitting Qi(T ) data for both low as and high values of the input power. 1 1 Z dt 1 Significant change in the population of the ground and α = ξ(a) = p ≡ K(k). excited TLS states due to Rabi is expected at b (1 − t2)(1 − t2k2) b 0 the average number of photons in the resonator n > n . √ c (14SM) The characteristic value nc ∼ 1/ τ1τφ depends on the K(k) is also known as the complete elliptic integral of the TLS relaxation time τ1(T ) and the dephasing time τϕ(T ). first kind, k = a/b. Similarly, the height of the In agreement with Ref. [S10], where the TLS relaxation is ξ time was shown to be τ1,2 ≈ τ(1 + γT ), we found that the extracted characteristic values of nc depend on the 1/k α 1 Z dt 1 p temperature as nc(T ) = nc(0) + µT , α ≈ 2 (Fig. S5). β = ≡ K( 1 − k2). b p(1 − t2)(1 − t2k2) b 0 (15SM) The electric field in the ξ-plane for the given V across the capacitor is uniform and can be easily obtained as V V Eξ = = √ b. (16SM) β K( 1 − k2)

The corresponding field in the w-plane scales with the factor ξ0(w) = dξ/dw which is

1 ξ0(w) = . (17SM) p(a2 − w2)(b2 − a2)

Thus, for example, the field strength at the center of FIG. S5. The temperature dependences of nc for different microstrip is resonators. 0 V |Ew(w = 0)| = Eξ · ξ (w = 0) = √ . (18SM) K( 1 − k2)a

VI. SCALING OF Pc(0) Accordingly, the power in the CPW can be written as

V 2 E2 p The ground and excited states of TLS become equally P ∼ ∼ a2K( 1 − k2)2. (19SM) populated when the Rabi frequency Ω = (d · Z Z √ E)/~ exceeds the rate 1/ τ1τφ, or, equivalently, when the electric field in the TLS-occupied volume exceeds the Therefore, we expect that the characteristic power scales 2 0 2 √ as Pc ∼ (a K(k ) )/Z, which is in agreement with the critical value Ec ≈ ~/(d τ1τφ). In order to understand the variation of the observed characteristic power for dif- experimental data. ferent resonators, we considered the dependence of the maximum electric field near the surface on the resonator parameters. The standard way to evaluate the characteristics of CPW resonators is by the Schwarz-Cristoffel (SC) map- ping of the coplanar topology to the trivial -plate capacitor geometry. Let us consider a zero-thickness CPW with a central strip width 2a and a ground-to- ground distance 2b. The transfer function for the map- ping of the upper half-plane to the rectangle is given by

w Z dw0 ξ(w) = A . (13SM) FIG. S6. The time dependence of Re[S21] measured at T = (w0 − a)(w0 + a)(w0 − b)(w0 + b) 25 mK at a fixed frequency on the slope of a resonance dip. 0 The microwave power corresponds to hni ∼ 1000. Each point Here A is an integration constant, chosen to be A = 1. corresponds to the data averaging over 1 sec. The half-width of the equivalent capacitor are calculated 11

kBT ) fluctuators result in the TLS spectral diffusion as well as the flicker noise. The telegraph noise in the reso- nance frequency fr is expected if some of the TLS with f ≈ fr are strongly coupled to a resonator. Typical TLS densities for Al/AlOx junctions are ∼1 (GHz·µm2)−1 [S11]. Interestingly, the number of strongly coupled TLS for our resonators (assuming that the strongly coupled TLS are in the oxide layer of the resonator) is of the or- der of unity [1 (GHz · µm2)−1 × 0.1MHz × 104µm2]. To study the telegraphy noise, we repetitively measured S21 at a fixed frequency on a slope of the resonance dip for a few minutes. Figure S6 shows an example of the mea- sured telegraph noise in Re[S21]. The characteristic time scale of random switching between two Re[S21] levels is 1-10 seconds.

VIII. PUMP-PROBE MEASUREMENTS OF THE TLS RELAXATION TIME FIG. S7. (a) The pulse sequence. (b) The time dependence of |S21| measured at f0 = 2.4258 GHz. The pump pulse at We have performed the time domain measurements of fp = f0 + 1 MHz was applied between t = 0 s and t = 0.5 s. The pump tone power corresponds ton ¯p ≈ 1000. Each data the TLS relaxation time for resonator #1 using the pulse point was averaged over 4000 cycles with the same readout sequence shown in Fig. S7(a). A 0.5 s-long pump pulse delay time. The inset shows CW measurement of S21 versus f was applied to the resonator at the beginning of each duty with (red) and without (blue) the pump signal and indicates cycle. A readout pulse at the single-photon power level the position of f0 used in the relaxation time measurement. lasting for 36 ms followed the pump pulse and was digi- The readout power was at the single photon level for all mea- tized to obtain S21. The readout delay time was varied surements on this plot. between 0 s and 1 s. Figure S7(b) shows the result of the experiment at the readout frequency f0 = 2.4258 GHz and the pump frequency fp = f0 + 1 MHz. The change VII. TELEGRAPH NOISE IN THE in |S21(f0)| at t = 0.5 s is consistent with CW measure- RESONATORS ments at the same readout frequency and power level when a pump tone was turned on and off. This indicates Interactions between the high-frequency (coherent, that an upper limit of the TLS relaxation time for our E > kBT ) TLS with the low-frequency (thermal, E < sample is much less than 36 ms.

[S1] M. T. Bell, L. B. Ioffe, and M. E. Gershenson, “Mi- [S6] R. C. Dynes and J. P. Garno, “Metal-Insulator Transi- crowave spectroscopy of a Cooper-pair transistor cou- tion in Granular Aluminum,” Phys. Rev. Lett. 46, 137– pled to a lumped-element resonator,” Phys. Rev. B 86, 140 (1981). 144512 (2012). [S7] W. A. Phillips, “Two-level states in glasses,” Reports [S2] M. S. Khalil, M. J. A. Stoutimore, F. C. Wellstood, Prog. Phys. 50, 1657–1708 (1987). and K. D. Osborn, “An analysis method for asymmet- [S8] D. C. Mattis and J. Bardeen, “Theory of the Anomalous ric resonator transmission applied to superconducting Skin Effect in Normal and Superconducting Metals,” devices,” J. Appl. Phys. 111, 054510 (2012). Phys. Rev. 111, 412–417 (1958). [S3] S. Probst, F. B. Song, P. A. Bushev, A. V. Ustinov, and [S9] J. Gao, The physics of superconducting microwave res- M. Weides, “Efficient and robust analysis of complex onators, Ph.D. thesis (2008). scattering data under noise in microwave resonators,” [S10] J. Lisenfeld, C. M¨uller, J. H. Cole, P. Bushev, Rev. Sci. Instrum. 86, 024706 (2015). A. Lukashenko, A. Shnirman, and A. V. Ustinov, “Mea- [S4] M. Tinkham, Introduction to superconductivity (Dover suring the Temperature Dependence of Individual Two- Publications, 2004). Level Systems by Direct Coherent Control,” Phys. Rev. [S5] R. W. Cohen and B. Abeles, “Superconductivity in Lett. 105, 230504 (2010). Granular Aluminum Films,” Phys. Rev. 168, 444–450 [S11] C. M¨uller, J. H. Cole, and J. Lisenfeld, “To- (1968). wards understanding two-level-systems in amorphous solids - Insights from quantum devices,” (2017), arXiv:1705.01108.