Micromachined integrated quantum circuit containing a superconducting qubit

T. Brecht,∗ Y. Chu, C. Axline, W. Pfaff, J. Z. Blumoff, K. Chou, L. Krayzman, L. Frunzio, and R. J. Schoelkopf
Department of Applied Physics, Yale University, New Haven, Connecticut 06511, USA and
Yale Quantum Institute, Yale University, New Haven, Connecticut 06520, USA
(Dated: November 14, 2016)
We present a device demonstrating a lithographically patterned transmon integrated with a mi-
cromachined cavity resonator. Our two-cavity, one-qubit device is a multilayer microwave integrated
quantum circuit (MMIQC), comprising a basic unit capable of performing circuit-QED (cQED) oper-
ations. We describe the qubit-cavity coupling mechanism of a specialized geometry using an electric
field picture and a circuit model, and finally obtain specific system parameters using simulations.
Fabrication of the MMIQC includes lithography, etching, and metallic bonding of silicon wafers.
Superconducting wafer bonding is a critical capability that is demonstrated by a micromachined
storage cavity lifetime 34.3 µs, corresponding to a quality factor of 2 million at single-photon ener-
arXiv:1611.02166v1 [quant-ph] 7 Nov 2016

gies. The transmon coherence times are T1 = 6.4 µs, and T2Echo = 11.7 µs. We measure qubit-cavity
dispersive coupling with rate χqµ /2π = −1.17 MHz, constituting a Jaynes-Cummings system with
an interaction strength g/2π = 49 MHz. With these parameters we are able to demonstrate cQED
operations in the strong dispersive regime with ease. Finally, we highlight several improvements
and anticipated extensions of the technology to complex MMIQCs.

I. INTRODUCTION II. QUBIT-RESONATOR COUPLING

Quantum circuits are soon reaching size and complex- Coupling between an electromagnetic resonator and a
ity that puts extreme demands on input/output connec- transmon occurs via shared electric and magnetic fields
tions as well as selective isolation among internal ele- of their respective modes. Both planar circuits and 3D
ments. Continued progress will require 3D integration qubits use a simple dipole antenna structure aligned with
and RF packaging techniques[1, 2] that allow for scaling. the electric field of a transmission line, waveguide or
Indeed, there are numerous developed technologies wait- cavity.[10] These common schemes are diagrammed in
ing to see fruitful implementation in the field of circuit- Fig. 1(a)-(b). It would be impractical to use a similar
QED (cQED), both from room temperature microwave scheme for coupling qubits to micromachined cavities be-
devices[3, 4] and complex superconducting circuits[5–7]. cause of the extreme aspect ratio imposed by the wafer
To address this opportunity and the associated challenges height. Instead, we desire to achieve the same coupling
for quantum coherence, we recently proposed the mul- while limiting ourselves to planar fabrication and wafer
tilayer microwave integrated quantum circuit (MMIQC) stacking. A circuit can be patterned on one of the cavity
architecture[8], which adapts many existing circuit design walls such that the electromagnetic fields couple to those
and fabrication techniques to cQED. A crucial step to- of the cavity, as shown in Fig. 1(c).
wards this vision is the demonstration of superconducting The coupling of an “aperture transmon” to a resonator
micromachined cavities[4], which can be used as quantum using fields that are out of the plane of transmon fabrica-
memories or as shielding enclosures to prevent cross-talk tion is described in Ref. [11]. In the device of the present
in more complex quantum computing devices. However, work, the coupling can be understood by analyzing the
integrating transmons into these micromachined cavities overlap between the electric fields of the transmon mode
has yet to be discussed, and is not a trivial matter of and those of the adjacent cavity mode(s), and also by
replicating the common methods in either existing pla- an equivalent circuit model. Translation of the aperture
nar or 3D cQED circuits. Fortunately, the flexibility and transmon away from center of the cavity wall results in a
durability of MMIQC hardware affords many possibilities mixture of electric (charge accumulation) and magnetic
for qubit integration. (current flow) coupling. However, the aperture trans-
In this work, we demonstrate one such possiblity mon’s central location maximizes total coupling.
through the design, fabrication, and characterization of a The schematic circuit diagram is depicted in Fig. 1(d).
quantum device containing a transmon qubit coupled to a A single Josephson junction connects the central island
superconducting micromachined cavity. It forms a simple to the rest of the cavity wall. It is accompanied by a
MMIQC capable of performing cQED operations. The junction capacitance (Cj ),which is small compared to the
techniques shown here can be improved and extended to other capacitors in the system: First, there is a capac-
realize more complex quantum circuitry. itance across the open annulus between the island and
the rest of the lower cavity wall (Cp ). Second, there is
a capacitance across the gap between the island and the
opposite wall of the cavity (Cg ). Lastly, there is capac-
itance C associated with the walls of the cavity, which
∗ teresa.brecht@yale.edu combines with an effective inductance L to create the
 2

(a) (b) including an arbitrary number of modes. Each mode

has transition frequency ωi between its first two levels
and an anharmonicity αi , which is greatest for the trans-
mon. Each pair of modes interacts via a dispersive shift
of strength χij .

III. DEVICE IMPLEMENTATION

(c) (d) The first MMIQC prototype is designed to have coher-

ent quantum modes that have sufficient coupling rates
L Cg between them, and allow for manipulation and measure-
ment with microwave pulses. The device featured in this
work consists of three quantum objects. An aperture
C Lj transmon couples simultaneously to two cavity modes. In
Cp + C j
addition to a micromachined cavity, a second cavity made
by traditional metal machining is incorporated to com-
pose a two-cavity/one-qubit MMIQC device. This cavity
allows readout of the transmon and micromachined cav-
ity states through two pins leading to coaxial cables for
FIG. 1. Illustrations of dipole coupling in cQED. Electric microwave access to the system, and is hereafter referred
dipole moment orientations for typical transmons are fabri- to as the “readout cavity”. The device, shown in Fig. 2,
cated to align with (a) the electric field of a planar transmis- displays a hybrid multilayer construction, including sil-
sion line resonator or (b) the electric field of an encapsulating icon wafers and conventionally machined metals united
3D cavity. Blue arrows show electric field lines of the each by indium bonding on flat surfaces. The integration of
resonator’s fundamental mode, and red arrows show electric the machined 3D cavity demonstrates the aperture trans-
field lines of the transmon mode. (c) The aperture transmon mon’s bipartite coupling and provides a convenient way
fields couple to the fundamental mode of the micromachined of connectorizing the MMIQC.
cavity in the device discussed in this work. For clarity, the dia-
gram shows an exploded cross-sectional view of two substrate
Next, we must choose design parameters to realize the
wafers, and it is not to scale. (d) Schematic circuit diagram MMIQC. We also impose that the qubit is in the trans-
of the aperture transmon and electromagnetic resonator. The mon regime, with suppressed charge dispersion[1, 15].
red coloring corresponds to the central island. The shape and position of the aperture transmon affect
properties of the system between which tradeoffs are con-
sidered. For example, the size of the inner island must
LC-resonator characteristic of the cavity’s fundamental be large enough to create a measurable g by the capac-
mode at frequency ωµ /2π. itance contribution Cg . However, if the inner island is
The system of qubit excitations and resonator pho- too large, the anharmonicity is reduced, limiting speed of
tons displays a Jaynes-Cummings interaction: ~g(a† σ − + manipulation pulses. Scaling trends of g changing with
aσ + ), where a† (a) creates (annihilates) a photon and respect to several relevant geometrical parameters are in-
σ + (σ − ) creates (annihilates) an excitation of the qubit. cluded in supplementary material[13]. The coupling be-
The coupling rate g = eV0 β/~ is a function tween the qubit and the micromachined cavity, χqµ , and
p of the capaci-
tances, β = Cg /(Cg +Cp +Cj ), and V0 = ~ωµ /2C. This that of the qubit and the readout cavity, χqr , are also
circuit picture using capacitances reveals the relation- adjusted by choice of heights of each cavity and thick-
ships between device geometry and coupling strength, ness of qubit substrate. We perform simulations in order
analogous to other cQED hardware designs.[12] See sup- to confirm our understanding of the qubit-cavity cou-
plementary material for details.[13] In this work, we oper- pling and to aid geometry optimization more precisely.
ate such a system in the the strong dispersive limit, where We model the entire system using a full 3D electromag-
the frequency detuning between resonator and qubit is netic simulation using a finite element solver followed by
much greater than the interaction rate (|∆|  g) and blackbox quantization analysis[16]. For the design fea-
the interaction rate is much greater than the decay rates tured in this work, the anharmonicity is designed to be
of the qubit or cavity (g  γ, κ). αq = −Ec = −204 MHz, and the Josephson energy is
In this strong dispersive limit, we approximate the ap- EJ /h = Φ20 LJ /2πh = 39 GHz. (EJ /EC = 193.)
plicable Hamiltonian as We now briefly describe how the device is constructed.
The multilayer device is fabricated as three separate parts
(see Fig. 2(a)) that are finally bonded together with a
H X αi †2 metal that superconducts. The micromachined cavity
ωi a†i ai − χij a†i ai a†j aj −
X X
= ai a2i (1) chip is created by wet etching a rectangular pit to depth
~ i i
2
i6=j of 300 µm in silicon, followed by metalization with 10 µm
 3

(a) (b) (c)

superconducting micromachined
bonds cavity
(22 x 24 x 0.3 mm)

Si
aperture
In transmon

Cu
readout cavity 1 mm
(28 x 30 x 3 mm)
with I/O couplers 1 cm

FIG. 2. (a) Sketch of device. For clarity, the image shows an exploded cross-sectional view that is not to scale. The annular
structure has electric dipole moment components in two opposing directions, both perpendicular to the plane of fabrication.
Red arrows show electric field lines of the transmon mode, and the transmon chip is shown semi-transparent. (b) Photograph
of the micromachined cavity chip (top) and transmon chip (bottom). (c) False colored SEM image of the aperture transmon,
with silicon in grey, aluminum in purple, and indium in blue. The shape of the electrodes is described in the supplementary
material[13]. An ‘X’ indicates the Josephson Junction position, interrupting a 50 µm wide lead connecting the inner island to
the remainder of the cavity wall.

indium.[4, 13] The transmon chip requires three metaliza-

TABLE I. Measured device parameters. The cross-Kerr inter-
tion steps on a 325 µm thick silicon wafer: a patterning of
action with the qubit mode is denoted χq , and anharmonicity
gold by liftoff, ebeam lithography and shadow angle evap- is α. Simulated parameters are in italics, and all other param-
oration of the aluminum Josephson junction, and mask- eters are measured except the anharmonicities of the cavities,
ing this junction before electroplating 10 µm of indium which are calculated by α = χ2q /4αq .[16]
onto the gold. The readout cavity is milled out of OFHC
Readout µ-Machined
copper and electroplated with 30 µm indium. The three
components are bonded together between parallel plates Mode cavity Transmon cavity
at 120° C in two steps. Once assembled, coaxial pin cou- Frequency (MHz) 6973.4 7351.4 9377.2
plers are added to the readout cavity and device is ther- [simulated] 6945.1 7322.0 9258.0
mally anchored to the baseplate of a dilution refrigerator αi /2π (MHz) -0.012 -209.8 -0.002
reaching a base temperature of 15 mK. See supplemen-
[simulated] -0.004 -204.3 -0.002
tary material[13] for additional fabrication and bonding
details. χqi /2π (MHz) -3.84 - -1.17
[simulated] -3.22 - -1.25
χri /2π (MHz) - - -0.020
IV. EXPERIMENTAL RESULTS [simulated] - - -0.004
T1 (µs) 1.0 6.4 34.3
Successful cQED operation in this new hardware is T2R (µs) - 9.5 -
demonstrated with measurements of coherence times
T2Echo (µs) - 11.7 -
and interactions between each quantum object of the
MMIQC. Measurements of the relevant coherence times
in the device are shown in Fig. 3(a)-(b) and summarized
in Table I. The qubit T1 = 6.4 µs is on the order of sponding to interaction strength g/2π = 38 MHz. In
other 3D transmons recently produced on silicon sub- spectroscopy, we observe both resolved photon number
strate in the same facility with similar methods[18] and splitting of the qubit (Fig. 3(a)) and a qubit-state de-
T2Echo ≈ 1.8T1 . The thermal population of the qubit ex- pendent shift of the micromachined cavity from a dis-
cited state is < 3 percent. The micromachined cavity has persive coupling rate χqµ /2π = −1.17 MHz. At detun-
a lifetime of 34.7 µs, which corresponds to a total quality ing of (ωq − ωµ )/2π = −2.03 GHz, this corresponds to
factor Q = 2 million at single-photon energies. g/2π = 49 MHz.
Next, we find interaction strengths sufficiently large in A final measurable parameter is the cross-Kerr interac-
relation to these coherences by showcasing some stan- tion between the two cavities. The cavity cross-Kerr χrµ
dard cQED functions. These measurements are shown in is measured by relative comparison of χqr and χrµ .[19]
Fig. 3(c)-(e). The dispersive coupling rate of the qubit A microwave pulse detuned 3 MHz above the readout
to the readout cavity is χqr /2π = −3.84 MHz, corre- cavity induces a Stark shift, which we use to precede
 4

(a) (c) (d)

20 30 ( )

µ-cavity − ∆ f (MHz)
1.0

RO
D1 40

RO
∆t

qubit − ∆ f (MHz)
signal (a.u.)
10
signal (a.u.)

20 2 2

RO
∆t 20 0.5
20
10 ∆t ∆t
2 2

RO
20 2 2

10 0 0.0
10 −5 0 5 0 300 600
0 10 20 30 40 qubit ∆ f (MHz) Stark tone power (mW)
(b) ∆ t (µs) (e)
50
20
D3

RO
2 2
∆t

signal (a.u.)
signal (a.u.)

40
10 n=0

RO
∆t
30
0
0 100 200 300 400 0 2 4 6 8
∆ t (µs) ∆ t (µs)

FIG. 3. (a) Qubit energy relaxation is fit to a single exponential (red line) with T1 = 6.4 µs. Ramsey dephasing time T2R =
9.5 µs, measured here using 400 kHz detuning from the qubit frequency. Using a Hahn echo sequence, we find T2echo = 11.7 µs,
measured here using 300 kHz detuning from the qubit frequency. (b) Energy decay of the micromachined cavity is measured by
applying a large displacement to this cavity, followed by a variable delay, followed by a spectrally narrow selective π-rotation
of the qubit conditioned on there being no photons in the readout cavity (n = 0). Using a Poissonian decay fit (red line), we
find T1 = 34.7 µs. At 9.4 GHz, this decay time corresponds to quality factor Q = 2 million. (c) We observe number splitting of
the qubit in spectroscopy after displacing the micromachined cavity by one photon. The spacing between the peaks indicates
χqµ /2π = 1.17 MHz (d) A tone detuned 3 MHz above the readout cavity induces a Stark shift that affects both the qubit and
micromachined cavity frequencies. We use the ratio of these slopes χqr /χrµ to determine χrµ . (e) In Ramsey interferometery
following a displacement of the micromachined (storage) cavity, we observe revivals of the qubit state occurring at integer
multiples of 2π/χqµ = 0.855 µs.[17]

single-side-band spectroscopy of both the qubit and mi- V. DISCUSSION OF LOSS MECHANISMS
cromachined cavity peaks. Both shift downward in fre-
quency with increasing power of the Stark pulse. The
slopes of this response are proportional to χqr and χrµ We assess several potential loss mechanisms that could
respectively (Fig. 3(b)). We have independently deter- be limiting the coherence times in our device. All quan-
mined χqr = −3.84 MHz by readout cavity spectroscopy tum circuits are subject to sources of loss associated with
with and without a preceding qubit π-pulse. Finally, we packaging and assembly that become more severe as com-
find cross-Kerr χrµ /2π = −20 kHz, compared to a simu- plexity increases.[8] For example, loss occurs at seams
lated value of −4.4 kHz. where there is finite conductance, gseam , and non-zero
i
admittance to surface currents, yseam , which may limit
i
a mode i’s coherence time to T1 = gseam /yseam ωi . In
the multilayer architecture, these seams are present in
the bonds between layers and interfaces between differ-
ent materials.
As a further demonstration of the micromachined cav- In this device, there are two types of seams that could
ity’s utility as a quantum memory, we perform Ramsey contribute to loss. The first consists of In/In bonds at
interferometry following a displacement that initializes the perimeter of the cavities. Using simulated surface
the micromachined cavity to a coherent state of |βi with currents, we calculate the admittance in the microma-
µ
an average of three photons. (Fig. 3(c)) In this experi- chined cavity mode to be yIn/In = 16.0 /Ωm. For the
ment, we prepare an initial state |βiµ ⊗ {|gi + |ei}, which q
qubit mode, yIn/In = 0.02 /Ωm, which is smaller be-
†
precesses according to eiχqµ ta a|eihe| .[17] Qubit state re- cuase the surface currents are localized away from the
vivals occur at time intervals 2π/χqµ , consistent with our In/In bond. Using the technique developed in [4], we
spectroscopic measurements of χqµ . are able to achieve an In/In bond conductance in our de-
 5

vices of gIn/In ≈ 108 /Ωm. If this were the only source MMIQC designs are being developed that contain differ-
of loss, it would limit the micromachined cavity to life- ent seams and minimize the use of normal metals like
time 100 µs. The second type of seam is a Al/Au/In gold. Extended qubit lifetimes can be achieved by re-
transition in a 3 × 3 mm square shape around the Al moval of silicon substrate in the junction area and more
aperture transmon region. Using simulated surface cur- directed surface cleaning.[18] Furthermore, a wide range
rents, we calculate the admittances in the micromachined of coupling rates can be accessed by geometry modifica-
µ
cavity and qubit mode: yAl/Au/In = 0.17 /Ωm and tions, some of which would require precise alignment and
q leveling control during wafer bonding.
yAl/Au/In = 0.52 /Ωm. Independent measurements of
stripline resonators fabricated with like procedures show We have shown a proof-of-principle MMIQC that
that gAl/Au/In ≈ 4.2 × 105 /Ωm.[13] The resulting limita- demonstrates the engineering of qubit-cavity coupling.
q
tion on the qubit mode lifetime is T1 < gseam /yseam ωq ≈ There are numerous possible next steps using the design
20 µs. Limitation on the cavity mode lifetime due to this strategies and fabrication tools described in this work.
seam is T1 ≈ 40 µs. For instance, the micromachined cavity and qubit can
be addressed using microstrips on the side of the wafer
We also verified that qubit and micromachined cavity
opposite to the micromachined cavity wall and qubit fab-
lifetimes are not limited by the Purcell effect due to the
rication, eliminating the machined readout cavity. The
overcoupled readout cavity. We simulated that the upper
microstrips would be positioned above apertures in the
bounds to the qubit and micromachined cavity lifetimes
metal of the cavity wall sized to create desired coupling
due to this effect are 200 and 500 µs respectively. In
rates. They can compose planar readout resonators and
design, the Purcell limit of the micromachined cavity is
incorporate Purcell filtering[20]. As a further example,
mitigated by minimizing the area of the annular opening
the addition of a second junction and flux-bias-line would
created by the aperture transmon between cavities.
constitute a frequency tunable device inspired by the
Also present here are surface dielectric and conductor concentric transmon in Ref. [21]. More sophisticated
loss mechanisms that are broadly studied in supercon- on-chip input/output circuitry, such as quantum lim-
ducting circuits. The particular shapes of the electrodes ited amplifiers[22–24], circulators[25, 26], and switching
in Fig. 2(c) are designed to minimize dielectric loss near elements[27, 28], will also be required for practical quan-
the surface of the electrodes. The aperture transmon has tum information processing. We anticipate that the tech-
smooth edges that are easily parameterized for optimiza- niques demonstrated here can be successfully employed
tion that includes consideration of surface participation toward integrating these elements into increasingly com-
ratios. See supplementary material[13]. plex MMIQCs.

VII. ACKNOWLEDGMENTS
VI. OUTLOOK

We thank Zlatko Minev, Kyle Serniak, Ioan Pop,

The device presented here demonstrates the integra- Michel H. Devoret, Hanhee Paik and David Pappas for
tion of a transmon with a superconducting microma- useful conversations, Subhajyoti Chaudhuri for assis-
chined cavity, forming the first actual MMIQC. The co- tance with simulations, Jan Schroers, and Emily Kinser
herence times and coupling rates are in the strong dis- for assistance with wafer bonding. This research was
persive regime of cQED, enabling many quantum manip- supported by the U.S. Army Research Office under grant
ulations that are the precursor to large scale quantum W911NF-14-1-0011. W.P. was supported by NSF grant
information processing. PHY1309996 and by a fellowship instituted with a Max
We remark that this achievement was made without Planck Research Award from the Alexander von Hum-
extensive fabrication optimization using industrial scale boldt Foundation. C.A. acknowledges support from the
tools, indicating process robustness and potential for im- NSF Graduate Research Fellowship under Grant No.
provement. For example, it is expected that surface DGE-1122492. Facilities use was supported by the Yale
cleaning will improve gAl/Au/In , a seam conductance rel- SEAS cleanroom, YINQE, and NSF MRSEC DMR-
evant to both the qubit and cavity modes. Alternative 1119826.

[1] Katehi, L.P.B, J. F. Harvey, and K. J. Herrick, IEEE [3] A. R. Brown, P. Blondy, and G. M. Rebeiz, Int. J. RF
Microwave Magazine 2, 30 (2001). Microw. Comp. Aid. Eng. 9, 326 (1999).
[2] J. H. Béjanin, T. G. McConkey, J. R. Rinehart, C. T. [4] L. P. Katehi, High Efficiency Micromachined Anten-
Earnest, C. R. H. McRae, D. Shiri, J. D. Bateman, Y. Ro- nas: Micromachined Antennas for Microwave and Mm-
hanizadegan, B. Penava, P. Breul, S. Royak, M. Zapatka, Wave Applications, Office of Naval Research Final Report
A. G. Fowler, and M. Mariantoni, Phys. Rev. Applied 6, 033095-1-F (Radiation Laboratory, University of Michi-
044010 (2016). gan, Ann Arbor, Michigan, 1997).
 6

[5] S. K. Tolpygo, Low Temperature Physics 42, 361 (2016). Science 350, 307 (2015).
[6] S. K. Tolpygo, V. Bolkhovsky, T. Weir, L. M. Johnson, [25] K. M. Sliwa, M. Hatridge, A. Narla, S. Shankar, L. Frun-
W. D. Oliver, and M. A. Gouker, Superconductor Science zio, R. J. Schoelkopf, and M. H. Devoret, Phys. Rev. X
and Technology 27, 025016 (2014). 5, 041020 (2015).
[7] D. Bintley, M. J. MacIntosh, W. S. Holland, J. T. [26] J. Kerckhoff, K. Lalumière, B. J. Chapman, A. Blais,
Dempsey, P. Friberg, J. T. Kuroda, E. G. Starman, H. S. and K. W. Lehnert, Phys. Rev. Applied 4, 034002 (2015).
Thomas, C. Walther, X. Gao, P. A. R. Ade, R. V. Sudi- [27] M. Pechal, J.-C. Besse, M. Mondal, M. Oppliger, S. Gas-
wala, C. Dunare, W. Parkes, A. J. Walton, K. D. Ir- parinetti, and A. Wallraff, Phys. Rev. Applied 6, 024009
win, G. C. Hilton, M. Niemack, M. Amiri, V. Asboth, (2016).
B. Burger, E. L. Chapin, M. Halpern, M. Hasselfield, [28] B. J. Chapman, B. A. Moores, E. I. Rosenthal, J. Kerck-
and A. L. Woodcraft, in SPIE Astronomical Telescopes hoff, and K. W. Lehnert, Appl. Phys. Lett. 108, 222602
+ Instrumentation, edited by W. S. Holland (Proc. SPIE, (2016).
2012) p. 845208.
[8] T. Brecht, W. Pfaff, C. Wang, Y. Chu, L. Frunzio, M. H.
Devoret, and R. J. Schoelkopf, npj Quantum Information
2, 16002 EP (2016).
[4] T. Brecht, M. Reagor, Y. Chu, W. Pfaff, C. Wang, L. Frun-
zio, M. H. Devoret, and R. J. Schoelkopf, Appl. Phys.
Lett. 107, 192603 (2015).
[10] H. Paik, D. I. Schuster, L. S. Bishop, G. Kirchmair,
G. Catelani, A. P. Sears, B. R. Johnson, M. J. Reagor,
L. Frunzio, L. I. Glazman, S. M. Girvin, M. H. Devoret,
and R. J. Schoelkopf, Phys. Rev. Lett. 107, 240501 (2011).
[11] Z. K. Minev, K. Serniak, I. M. Pop, Z. Leghtas, K. Sliwa,
L. Frunzio, R. J. Schoelkopf, and M. H. Devoret, Phys.
Rev. Applied 5, 044021 (2016).
[12] A. Blais, R.-S. Huang, A. Wallraff, S. Girvin, and
R. Schoelkopf, Phys. Rev. A 69, 062320 (2004).
[13] See Supplemental Material for additional details pertain-
ing to coupling trends, fabrication, transmon geometry,
and loss analysis.
[1] J. Koch, T. M. Yu, J. Gambetta, A. A. Houck, D. I. Schus-
ter, J. Majer, A. Blais, M. H. Devoret, S. M. Girvin, and
R. J. Schoelkopf, Phys. Rev. A 76, 042319 (2007).
[15] J. Schreier, A. Houck, J. Koch, D. Schuster, B. Johnson,
J. Chow, J. Gambetta, J. Majer, L. Frunzio, and D. M,
Phys. Rev. B 77, 180502(R) (2008).
[16] S. E. Nigg, H. Paik, B. Vlastakis, G. Kirchmair,
S. Shankar, L. Frunzio, M. H. Devoret, R. J. Schoelkopf,
and S. M. Girvin, Phys. Rev. Lett. 108, 240502 (2012).
[17] B. Vlastakis, G. Kirchmair, Z. Leghtas, S. E. Nigg,
L. Frunzio, S. M. Girvin, M. Mirrahimi, M. H. Devoret,
and R. J. Schoelkopf, Science 342, 607 (2013).
[18] Y. Chu, C. Axline, C. Wang, T. Brecht, Y. Y. Gao,
L. Frunzio, and R. J. Schoelkopf, Appl. Phys. Lett. 109,
112601 (2016).
[19] Z. Leghtas, S. Touzard, I. M. Pop, A. Kou, B. Vlastakis,
A. Petrenko, K. M. Sliwa, A. Narla, S. Shankar, M. J. Ha-
tridge, M. Reagor, L. Frunzio, R. J. Schoelkopf, M. Mir-
rahimi, and M. H. Devoret, Science 347, 853 (2015).
[20] M. D. Reed, B. R. Johnson, A. A. Houck, L. DiCarlo,
J. M. Chow, D. I. Schuster, L. Frunzio, and R. J.
Schoelkopf, Appl. Phys. Lett. 96, 203110 (2010).
[21] J. Braumüller, M. Sandberg, M. R. Vissers, A. Schnei-
der, S. Schlör, L. Grünhaupt, H. Rotzinger, M. Marthaler,
A. Lukashenko, A. Dieter, A. V. Ustinov, M. Weides, and
D. P. Pappas, Appl. Phys. Lett. 108, 032601 (2016).
[22] M. A. Castellanos-Beltran and K. W. Lehnert, Appl.
Phys. Lett. 91, 083509 (2007).
[23] N. Bergeal, F. Schackert, M. Metcalfe, R. Vijay, V. E.
Manucharyan, L. Frunzio, D. E. Prober, R. J. Schoelkopf,
S. M. Girvin, and M. H. Devoret, Nature 465, 64 (2010).
[24] C. Macklin, K. O’Brien, D. Hover, M. E. Schwartz,
V. Bolkhovsky, X. Zhang, W. D. Oliver, and I. Siddiqi,
 1

Supplemental Material:

Micromachined integrated quantum circuit containing a superconducting qubit

I. COUPLING TRENDS

To get a rough estimate of the coupling, we can approach the problem in a similar way to the derivation found in
Ref. [S1], which describes coupling of a dipole moment to the fields of a transmission line resonator. First, we seek to
obtain an expression for the dipole moment of the annulus structure in terms of the inner radius ri and outer radius
ro .
Here we describe an analytic approach in two limiting cases. Begin with a large conductor plane with an annular
ring removed, and assume that a voltage V0 is applied across the annulus. There will there be an electric field across
the gap as well as an electric field pattern out of the plane. The total field has dipole components pointing in opposite
directions on either side of the plane.
In the limiting case of an infinitely thin annulus, ri ≈ ro ≡ r, the field is that of two opposing dipoles of identical
magnitude pointing out of the annular plane in both directions. The magnitude of each is calculated in Ref. [S2]:
4πo
|p| = V0 r 2 . (S1)
3
In the limit of no central island at all (ri = 0), we have a circular aperture. If ro < λ, we can make the following
dipole approximation. In the presence of an incident E-field Eo , the field on the other side of the conducting plane
can be approximated as a dipole with r3 dependence, [S3]
2 3
|p| = r Eo . (S2)
3 o
At the aperture of a resonant cavity, there is an oscillating dipole moment peiωt which radiates power. Consider
the average power radiated from a dipole into half-space: [S2]

1 1 |p|2 ω 4
Prad = .
2 4πo 3c3
As the radius of the inner island goes from 0 to ro , there must be some smooth correspondence between the fields
in the two limits described above. The actual system to which we want to apply this understanding is complicated
by several effects. First is the presence of a nearby conductor: The distance from the annulus island to the wall of

(a) ri = 500 µm, ro= 800 µm (b) h = 300 µm, ro = 800 µm (c) h = 300 µm, ri = 500 µm
0.3 200 0.3 200 0.3 200

h ri ro
0.2 0.2 0.2
g (MHz)

g (MHz)

g (MHz)

100 100 100

β

0.1 0.1 0.1

0.0 0 0.0 0 0.0 0

0 300 600 0 400 800 500 700 900
µ− cavity height (µm) inner radius (µm) outer radius (µm)

FIG. S1. Relationships of the capacitance ratio β (black) and coupling rate g (red) versus three geometrical variations. These
results were obtained using ANSYS® Maxwell® to model the capacitance matrix of a set of three conductors while including
the silicon dielectric. The capacitance ratio β = Cg /(Cg + Cp + Cj ) is plotted in black and the coupling rate g is plotted in
red. (a) Coupling increases with reduced micromachined cavity height by way of increasing gap capacitance Cg . (b) Coupling
increases with smaller ri , but with weakening dependence as ri becomes smaller than h. (c) Coupling increases with larger ro ,
and the relationship is linear when ro > 2ri .
 2

the micromachined cavity is on the same order as the annulus dimensions. Furthermore, the fields are affected by the
presence of the dielectric substrate. Finally, we have eventually chosen a shape that differs from the circular annulus.
In practice, an analytic expression for the fields or coupling strength is difficult and not the best avenue for design.
Instead, we do an electrostatic simulation of the relevant capacitances to investigate relationships to geometry.
Some such relationships are shown in Fig. S1. We p plot the coupling rate g = eV0 β/~, which is a function of the
capacitances, β = Cg /(Cg + Cp + Cj ), and V0 = ~ωµ /2C, as they are defined in the main text. These trends are
useful guides to design. Finally, we do a full driven modal finite element analysis simulation of the fields of the total
design using ANSYS® HFSSTM .

II. DEVICE FABRICATION

The device is fabricated as three separate parts that are finally bonded together (see Fig. 2(a) of the main text).
The process for the transmon chip begins with a 325 µm thick (100)-orientation double-side-polished silicon wafer
with resistivity ρ > 10 kΩcm. First, a gold pattern is defined on what will be one of the cavity walls using optical
lithography and liftoff. The deposition is 10 nm Ti followed by 100 nm Au by e-beam evaporation. Second, the
qubit is fabricated by e-beam lithography and a double-angle shadow e-beam evaporation of an aluminum Josephson
junction, followed by liftoff. Third, the aluminum is masked with photoresist and the gold pattern on the wafer is
electroplated with indium to a thickness of 10 µm.
The fabrication process for the micromachined cavity chip is described in Ref. [S4]. It begins with a 1 mm thick
(100)-orientation silicon wafer with resistivity ρ > 10 kΩcm. A 22 × 24 mm rectangular pit is lithographically defined
using a nitride mask. The pit is then etched to depth 300 µm using a KOH etch bath at 85 °C. The concentration of
KOH is 30 percent by weight in water and the etch bath also includes 1 percent isopropanol added as a surfactant.
The wet etch is anisotropic with selectivity to silicon’s (100) : (111) planes, resulting in a rectangular recess with
sidewalls determined by the crystal planes and RMS surface roughness of Rq = 20 − 40 nm. The nitride mask is
removed using BOE, which also removes salts left over from the etch bath. Then the wafer is coated in 100 nm
evaporated gold before electroplating with indium to a thickness of 10 µm.
Electroplating of the above described chips is done in an indium sulfamate plating bath solution purchased from
Indium Corporation is operated at room temperature. In a wafer plating system purchased from Wafer Power
Technology, a rotating Pt-Ti mesh serves as the insoluble anode and the wafer serves as the cathode onto which
indium precipitates from the bath. A DC current of 300 mA is used to deposit indium at at rate of 150 nm/min to
a final thickness of 10 µm on the full area of the 4 inch diameter substrate. The deposited material is known to be
> 99.9% pure indium.
The readout cavity is machined in OFHC copper and electroplated to a thickness of 30 µm by an external vendor.
The surface indium thickness is greater than that of the chip components in order to ensure complete coverage of the
interior corners of the cavity.
The three components are bonded together between parallel plates of an Instron (5969) at 120°C in two steps. The
components are sized such that the readout cavity completely surrounds the transmon chip and intersects the cavity
chip. The two chips are bonded together first with 1 kN, followed by the bond of the readout cavity part to the cavity
chip using 5 kN. For each bond, the force is ramped from 0 to the target over 1 minute and then held at target for 1
minute. Immediately prior to bonding, indium oxide is etched away with a solution of 10 percent hydrochloric acid
in for 5 minutes, followed by DI water, acetone and methanol washing. (However, the transmon chip is not given a
pre-bond etch because the hydrochloric acid damages the aluminum.) Once it is bonded, a cover piece made of copper
(not pictured in Fig. 2) protects the stack and allows mounting to a thermalization bracket.
Once assembled, coaxial pin couplers to the readout cavity are tuned to provide Qin = 430, 000 and Qout = 42, 000.
The device is thermally anchored to the baseplate of a dilution refrigerator reaching a base temperature of 10 mK. A
cryoperm shield protects the device from magnetic fields.
Note that in this assembly the readout cavity is larger than the micromachined cavity in all dimensions, and it
is loaded with the bulk dielectric of the transmon chip. The readout is therefore lower frequency and a junction
inductance Lj is chosen such that the transmon frequency lies between those of the readout and micromachined
cavities: ωr < ωq < ωµ .
 3

III. TRANSMON GEOMETRY

With all of our design considerations in mind, we chose the following transmon geometry. The shape of the aperture
cut from the outer conductor is a cardiod described by

x = 2r (2 cos t − cos 2t + 21 )

outer r , 0 ≤ t < 2π. (S3)
y = 2 (2 sin t − sin 2t)

The inner island is a piriform curve described by

x = b( 12 − 2 sin t)

inner island , 0 ≤ t < 2π. (S4)
y = b cos t(1 + sin t)

These shapes were chosen in order to easily parameterize simulation variations. They may not necessarily be the
optimal shapes for maximizing coupling strengths and minimizing surface losses. Nevertheless, these shapes produce
a satisfactory design with parameter choices r = 750 µm and b = 100 µm. This shape is seen in Fig. 2(c) of the main
text.

IV. SURFACE LOSS

One design consideration is the effect of surface dielectric loss on qubit lifetime. The two-step simulation method
described in Ref. [S5] was performed on several representative design variations. For the chosen geometry, the
simulated metal-substrate surface participation is 6.08 × 10−4 , substrate-air participation is 3.24 × 10−4 , and metal-
air participation is 3.56 × 10−5 .

V. SEAM LOSS ACCOUNTING

We include in this supplementary note more detailed information on the In/Au/Al seam mentioned in the main
text. We apply the model introduced in Ref. [S4], which treats a seam as a distributed port around a path ~l. Surface
currents J~s that pass across the seam and dissipate power
Z
1
Pdis = L |J~s × ˆl|2 dl, (S5)
2Gseam seam

where L is the total length of the seam and Gseam is the total conductance.
If it is damped solely by this power dissipation at the seam, a cavity mode of frequency ω and total energy Etot
has a quality factor Qi given by
" R #
1 1 Pdis 1 L seam |J~s × ˆl|2 dl yseam
= = = , (S6)
Qi ω Etot Gseam ωµo ~ gseam
R
2
|H| dV
tot

where the field H~ is integrated over the volume V of the mode and µo is the magnetic permeability. We identify the
expression in square brackets as the admittance, Yseam , of the cavity presented to the seam. This admittance is zero
when the seam is placed such that there are no perpendicular surface currents.
In order to compare intrinsic seam properties in different cavity constructions, assume uniform conductance and
introduce a conductance per unit length gseam = Gseam /L. Then we can isolate the conductance per unit length and
identify the remainder of the expression as an admittance of the cavity as seen by the seam per unit length of the
seam:

|J~s × ˆl|2 dl
R
yseam = seamR . (S7)
ωµo ~ 2 dV
|H|
tot

Using this model, we can associate yseam with the seam location and cavity fields and gseam with materials properties
in the seam region. In the next section, we analyze the placement of Al/Au/In seam in the device of the main text.
Then, in the following section, we describe an experiment in which we have measured the conductance and conclude
with some remarks on the materials.
 4

A. Admittance control

We claim in the main text that one important design choice is the path of the seams. Several possible paths for the
seam near the transmon are drawn as white dotted lines in Fig. S1. The corresponding seam admittances for both the
qubit and cavity modes are numerically calculated in Table S1. The table also calculates limits on qubit and cavity
lifetimes using a conductance measured by an experiment described in the next section. We conclude that this seam
may spoil the lifetime of the qubit or the cavity if it is not placed with intention to minimize its admittance. The
path used for the actual device is the 3×3 mm square.
TABLE S1. Limits on qubit and storage cavity life-
times derived from seam admittances. The qubit mode yseam
2 values are found from HFSS simulation of the design fea-
tured in this work. Storage mode yseam is calculated ana-
lytically by integrating surface currents of rectangular cav-
1 ity’s TE101 mode, and checked against simulation. The in-
ferred lifetime limits are T1q,µ < gseam /yseam ωq,µ , assuming
(mm)

0 gAl/Au/In = 4.2 × 105 /Ωm, and using ωq /2π = 7.3 GHz, and
ωµ /2π = 9.25 GHz. *The last line computes limits imposed
by the indium-to-indium bond around the perimeter of the
−1 micromachined cavity using gIn/In = 1 × 108 /Ωm.
q
yseam µ
yseam max T1q max T1µ
−2 seam (/Ωm) (/Ωm) (µs) (µs)
In/Au/Al circle, r=1.00 mm 7.955 0.0114 1.1 640
−3 −2 −1 0 1 2 3 In/Au/Al circle, r=1.25 mm 1.565 0.0467 5.8 160
(mm)
In/Au/Al circle, r=1.75 mm 0.382 0.161 24 57
In/Au/Al square, 3×3 mm 0.524 0.172 17 42
FIG. S1. White dotted lines show seam paths that were con-
sidered for the In/Au/Al transition. The patchmon geometry In/Au/Al square, 4×4 mm 0.187 0.398 49 18
(black lines) is fabricated in Al (purple) on Si (grey), defined In/Au/Al square, 5×5 mm 0.096 0.756 96 9.6
by curves following equations 3 and 4 with r = 750 µm and In/In, cavity perimeter* 0.0239 15.96 91000 108
b = 100 µm. The purple area inside the white dotted line
is Al evaporated in a double-angle fashion simultaneous to
the Josephson-junction. The purple area outside the white
dotted line is In, and extends to form one wall of the mi-
cromachined cavity. This seam region is conceptually drawn
in cross-section in Fig. S2. The corresponding seam admit-
tances for both the qubit and cavity modes are numerically
calculated in Table S1.
 5

10 µm
100 nm Au/ 3 µm PR
10 nm Ti 70 nm Al

100 µm

wafer

NOT TO SCALE
10 µm In

Si
Ti + Au
Al
Photoresist
In

FIG. S2. Conceptual drawings of the Al/Au/In seam re-

gion shown in cross-section during two intermediate fabrica-
tion steps (top) and the final product (bottom). The drawing
portrays the rugged surface of the indium (blue). Note that
the In overlaps the Al such that Au is not seen when imaged
from above, as in Fig. S3. Nevertheless, current transiting
this seam must pass from superconducting In through 100 µm
normal Au to superconducting Al.

B. Conductance measurement

Now we take a closer look at the seam in question. Fig. S2 shows the seam region in cross-section. Current transiting
this seam must pass from superconducting In through 100 µm normal Au to superconducting Al. Superconductor
proximity effect is complicated by the possibility of intermetallics on both interfaces, which we discuss more later.
The conductance of this seam, gIn/Au/Al , was determined by a separate set of devices and an experiment that is
illustrated in Fig. S3. In this experiment, we measured quality factors of stripline resonators in which the seam in
question is the dominant loss mechanism. Microscrip resonators were fabricated on sapphire substrate with the same
series of fabrication steps described for the transmon wafer in the main text. The resonator chips were loaded in a
multiplexed version of the coaxial stripline package described in Ref. [S6] (Fig. S3(e)) and measured in a cryogen-free
dilution refrigerator at 15 mK. The measured internal quality factors of these resonators are plotted against seam
admittance in Fig. S3(d). The blue line is the best fit of the data to Qi = gseam /yseam using linear least-squares
regression in the log-log domain, which yields gseam = (4.2+0.6 5
−0.5 ) × 10 /Ωm. We expect that improvement is possible
using cleaning methods, the exploration of which can be the subject of future study.

C. Conductance discussion

The presence of Au is certainly non-ideal. It exists in this process to serve as the conducting layer onto which In
is electroplated. Most other conductors are not suitable for In electroplating because of either oxides or incompatible
electronegativity. Cu is another appropriate metal for In electroplating. However, we find that DC contact resistance
for Cu/Al is lower than that of Au/Al (at least without experimenting with cleaning procedures before Al deposition).
Therefore, Au is used as the electroplating seed layer. Furthermore, all Al on the wafer must be completely masked
during electroplating to prevent its destruction.
 6

(a) (b) 500 µm (c)

PR/Al

Al In/Au 100 µm
PR/Al/Au

10 - 11.5 mm Al
Sapphire

Au Au PR/Au
In

Sapphire

(d) (e)
4 coupling
10 pins clamp
measured Q i

3
10

5
gseam = 4.2 x 10 /Ωm substrate
stripline
1 2 3
3 mm
10 10 10
waveguide enclosure
yseam ( /Ωm)

FIG. S3. Stripline resonators were used to measure the conductance of the seam in question. (a) Each device is a rectangular
strip of conductor with transitions from Al (purple) to In (blue) that resemble those of the device featured in the main text
as closely as possible. Devices were made with various numbers of transitions, and varying lengths to produce fundamental
half-wave modes in the 7-8 GHz range. (b) Optical microscope image of and end of one device. Three transition regions are
visible. False coloring is applied for clarity. (c) Optical microscope image showing an intermediate fabrication step, which is
featured in cross section in Fig. S2. False coloring is applied for clarity. (e) Measured internal quality factors of the several
devices of varying seam admittances. The blue line is the best fit of the data to Qi = gseam /yseam using linear least-squares
regression in the log-log domain, which yields gseam = (4.2+0.6 5
−0.5 ) × 10 /Ωm. (e) Each device is inserted into a tube-like package
before measurement at 15 mK. Image used with permission from Ref. [S6].

It is worth noting that Au is known to interact with the nearby metals in worrisome ways. In particular, Au will
form intermetallic compounds with both Al and In. Aux Aly intermetallics are known problems in microelectronics
and wire-bonding. Two such legendary intermetallics are “white plague” (Au5 Al2 ), which has low conductivity, and
“purple plague” (AuAl2 ), which leads to voids at the interface. Our process avoids the high temperatures (> 624°C)
that produce these compounds, but it is possible that diffusion takes place to form AuAl2 (occurring at > 400°C,
possibly occurring slowly at lower temperatures), which is also a poor conductor.

Au also diffuses into In, and the intermetallic AuIn2 forms when the temperature is elevated above the melting point
of In: 156°C. This intermetallic is a known problem in indium bump manufacturing methods that involve reflow of the
indium[S7], and diffusion rates of have been measured by Refs. [S7, S8]. Our process does not involve molten In. There
is concern, that any contaminants in the In, Au or other, can reduce its ductility, which has negative consequences
for the conductance of In bonds between layers. We do not know how this may affect the superconductivity of In.

Though evidence of these intermetallics has not been discovered in our samples, they are considered cause for
concern. Therefore, fabrication procedures that eliminate the Au layer altogether are being explored for future use.
In particular, In can be deposited via thermal evaporation instead of electroplating. Alternatively, a thin Ti layer
between Au and In could be incorporated into our process, as it is an effective diffusion barrier at low temperature in
 7

Ref. [S8].

[S1] J. Koch, T. M. Yu, J. Gambetta, A. A. Houck, D. I. Schuster, J. Majer, A. Blais, M. H. Devoret, S. M. Girvin, and R. J.
Schoelkopf, Phys. Rev. A 76, 042319 (2007).
[S2] J. D. Jackson, Classical Electrodynamics, 3rd ed. (Wiley, 1998).
[S3] R. E. Collin, Field Theory of Guided Waves, 2nd ed. (Wiley-IEEE Press, New York, 1990).
[S4] T. Brecht, M. Reagor, Y. Chu, W. Pfaff, C. Wang, L. Frunzio, M. H. Devoret, and R. J. Schoelkopf, Appl. Phys. Lett.
107, 192603 (2015).
[S5] C. Wang, C. Axline, Y. Y. Gao, T. Brecht, Y. Chu, L. Frunzio, M. H. Devoret, and R. J. Schoelkopf, Appl. Phys. Lett.
107, 162601 (2015).
[S6] C. Axline, M. Reagor, R. W. Heeres, P. Reinhold, C. Wang, K. Shain, W. Pfaff, Y. Chu, L. Frunzio, and R. J. Schoelkopf,
Appl. Phys. Lett. 109, 042601 (2016).
[S7] Q. Huang, G. Xu, Y. Yuan, X. Cheng, and L. Luo, J. Micromech. Microeng. 20, 055035 (2010).
[S8] W. Zhang and W. Ruythooren, J. Elec. Materi. 37, 1095 (2008).