Superconductor Science and Technology

PAPER • OPEN ACCESS Related content

- Transition Edge Sensors and Kinetic
Flexible superconducting Nb transmission lines on Inductance Detectors in Astronomical
Instruments
thin film polyimide for quantum computing P. D. Mauskopf

- Thin-film ferroelectric microwave devices

applications M J Lancaster, J Powell and A Porch

- Design, fabrication and characterization of

To cite this article: David B Tuckerman et al 2016 Supercond. Sci. Technol. 29 084007 a 64 pixel metallic magnetic calorimeter
array with integrated, on-chip microwave
SQUID multiplexer
S Kempf, M Wegner, L Deeg et al.

View the article online for updates and enhancements.

Recent citations
- Thin-Film Nb/Polyimide Superconducting
Stripline Flexible Cables
Vaibhav Gupta et al

- Polyimide for silicon solar cells with

double-sided textured pyramids
Ngwe Zin et al

- Recent research trends for high coherence

quantum circuits
M Steffen et al

This content was downloaded from IP address 190.17.53.139 on 17/04/2019 at 04:12

 Superconductor Science and Technology

Supercond. Sci. Technol. 29 (2016) 084007 (12pp) doi:10.1088/0953-2048/29/8/084007

Flexible superconducting Nb transmission

lines on thin film polyimide for quantum
computing applications
David B Tuckerman1, Michael C Hamilton2, David J Reilly3, Rujun Bai2,
George A Hernandez2, John M Hornibrook3, John A Sellers2 and
Charles D Ellis2
1
Microsoft Research, Redmond, WA 98052, USA
2
Alabama Micro/Nano Science and Technology Center, Electrical and Computer Engineering Department,
Auburn University, Auburn, AL 36849, USA
3
ARC Centre of Excellence for Engineered Quantum Systems and Station Q Sydney, School of Physics,
The University of Sydney, Sydney, NSW 2006, Australia

E-mail: v-datuck@microsoft.com

Received 30 March 2016, revised 1 June 2016

Accepted for publication 8 June 2016
Published 11 July 2016

Abstract
We describe progress and initial results achieved towards the goal of developing integrated
multi-conductor arrays of shielded controlled-impedance flexible superconducting transmission
lines with ultra-miniature cross sections and wide bandwidths (dc to >10 GHz) over meter-scale
lengths. Intended primarily for use in future scaled-up quantum computing systems, such flexible
thin-film niobium/polyimide ribbon cables could provide a physically compact and ultra-low
thermal conductance alternative to the rapidly increasing number of discrete coaxial cables that
are currently used by quantum computing experimentalists to transmit signals between the
several low-temperature stages (from ∼4 K down to ∼20 mK) of a dilution refrigerator. We have
concluded that these structures are technically feasible to fabricate, and so far they have
exhibited acceptable thermo-mechanical reliability. S-parameter results are presented for
individual 2-metal layer Nb microstrip structures having 50 Ω characteristic impedance; lengths
ranging from 50 to 550 mm were successfully fabricated. Solderable pads at the end terminations
allowed testing using conventional rf connectors. Weakly coupled open-circuit microstrip
resonators provided a sensitive measure of the overall transmission line loss as a function of
frequency, temperature, and power. Two common microelectronic-grade polyimide dielectrics,
one conventional and the other photo-definable (PI-2611 and HD-4100, respectively) were
compared. Our most striking result, not previously reported to our knowledge, was that the
dielectric loss tangents of both polyimides, over frequencies from 1 to 20 GHz, are remarkably
low at deep cryogenic temperatures, typically 100× smaller than corresponding room
temperature values. This enables fairly long-distance (meter-scale) transmission of microwave
signals without excessive attenuation, and also permits usefully high rf power levels to be
transmitted without creating excessive dielectric heating. We observed loss tangents as low as
2.2×10−5 at 20 mK, although losses increased somewhat at very low rf power levels, similar to
the well-known behavior of amorphous inorganic dielectrics such as SiO2. Our fabrication
techniques could be extended to more complex structures such as multiconductor cables,

Original content from this work may be used under the terms
of the Creative Commons Attribution 3.0 licence. Any
further distribution of this work must maintain attribution to the author(s) and
the title of the work, journal citation and DOI.

0953-2048/16/084007+12$33.00 1 © 2016 IOP Publishing Ltd Printed in the UK

Supercond. Sci. Technol. 29 (2016) 084007 D B Tuckerman et al

embedded microstrip, 3-metal layer stripline or rectangular coax, and integrated attenuators and
thermalization structures.
Keywords: polyimide, superconducting, resonator, dielectric loss, flexible, transmission lines,
quantum computing

(Some figures may appear in colour only in the online journal)

1. Introduction number of control signals to those qubits in order to drive

simultaneous gate operations, and how to receive a large
There is currently great interest in building quantum com- number of simultaneous measurement signals from the qubit
puters that could perform certain types of useful computations array, particularly when these signals need to span a large
(e.g., quantum chemistry) faster and more cost-effectively temperature difference (e.g., different stages of a dilution
than any existing digital computer technology. The error rates refrigerator). Our work is intended to address this specific
of certain types of qubit technologies have now improved to interconnection problem.
the point that it seems that such a feat of engineering may With the exception of optically excited qubits that can
become possible in the next decade, albeit very challenging operate at room temperature, most scalable solid-state qubit
and costly [1]. To be economically compelling, a quantum technologies (e.g., superconducting qubits, spin qubits) are
computer should probably contain the equivalent of at least principally controlled by precisely shaped microwave pulses;
several hundred sufficiently error-free qubits, or ‘logical these frequencies are most commonly in the 5–15 GHz range.
qubits’. Unfortunately, physically realizable qubits (‘physical These pulses may be used to drive transitions between two
qubits’) are prone to occasional errors due to relaxation and/ states of a qubit (e.g., between the primary ∣0ñ and ∣1ñ states),
or phase decoherence, quantum logic gate operations are not to create ‘cross-resonance’ conditions [13], or to induce
perfectly accurate, and the measurement of quantum states sideband transitions [14]. Microwaves are also commonly
can return erroneous results as well as cause back-action on used to measure the state of a resonator-coupled qubit (i.e.,
those states. Quantum error correction has recently been ‘dispersive readout’) [15]. Furthermore, microwaves are also
experimentally demonstrated to extend the lifetime of used as pump tones for low-noise superconducting parametric
quantum information [2–5], but to achieve usefully long amplifiers in conjunction with qubit readout [16].
coherence times, the physical ‘overhead’ (ratio of the number The use of 5–15 GHz control signals necessarily implies
of physical qubits required to emulate a given number of that the qubits be cooled to temperatures in the 10 s of mK to
logical qubits) is extraordinarily large: factors of thousands or avoid thermal population of the ∣1ñ state; the qubits also need
even millions may well be needed, depending on one’s to be well-shielded from higher-temperature blackbody
assumptions about error rates [6]. Recent progress towards radiation, which could act as electromagnetic interference or
creating ‘topologically protected’ solid-state qubits (e.g., sources of non-equilibrium quasiparticles [17–19]. Fre-
based on Majorana zero modes) offers the hope that qubit quencies much above 15 GHz are seldom used because of the
error rates could be made exponentially small, producing a difficulty of suppressing parasitic modes, as well as the dra-
near-perfect ‘quantum memory’ [7, 8]. Schemes have also matically higher cost of components and equipment. In
been proposed by which certain quantum gate operations (the addition to the microwave signals, most qubit technologies
Clifford group) could also be implemented in a topologically also have need for relatively slow-changing or even dc signals
protected manner (specifically, by ‘braiding’ Majorana zero (typically currents for superconducting qubits; voltages for
modes, which are predicted to behave as non-Abelian anyons) semiconductor-based qubits), as well as fast (ns-scale) non-
[9]. However, the Clifford gates are insufficient by them- sinusoidal pulses for other control operations such as acti-
selves to realize a universal quantum computer, and therefore, vating tunable couplers to create precise 2-qubit interac-
some substantial level of physical overhead for quantum error tions [20].
correction (e.g., to perform ‘state distillation’) would still be The limited cooling capacities of dilution refrigerators
required [10]. (generally 1 mW at the 20–30 mK temperatures of interest)
It seems, therefore, that even in the most optimistic and the need to prevent contamination of qubits by any type
scenario wherein extremely low physical error rates are of non-equilibrium excitations seem to require that the
achieved, a useful quantum computer will require many majority of control signals in a quantum computer be gen-
thousands of physical qubits, and less optimistic scenarios erated in a higher-temperature environment, where much
will require millions. This is now well understood by qubit greater cooling capacity is available and dissipative circuits
device researchers, and there have been a number of recent will not disrupt nearby qubits. Such signals are then trans-
proposals as to how one might architect systems that are mitted via shielded impedance-controlled GHz-capable
scalable to large numbers of qubits [11, 12]. Most of these transmission lines to the qubit environment. Depending on the
proposals deal with how qubits could be integrated and architecture, the number of required transmission lines may
interconnected in two-dimensional arrays, using compact someday number into the millions. The use of this many high-
wafer-like configurations. There has been limited attention performance discrete coaxial cables would seem to be
given to the physical problem of how to transmit a large impractical, considering their collective physical size, thermal

2
Supercond. Sci. Technol. 29 (2016) 084007 D B Tuckerman et al

Figure 1. Conceptual cross-sectional diagram of a future ‘rectangular coax’ ribbon cable that could be used to communicate between different
temperature stages of a dilution refrigerator. The structures reported in this paper are much simpler, but the experimental results provide
confidence that much more complex cables, such as illustrated here, could be fabricated and be usable in quantum computing experiments.

leakage, and cost. Various multiplexing schemes are being fabricated. Owing to its small cross sectional area, the axial
pursued [21, 22], which could reduce the number of required heat leakage along such a ribbon cable would be orders of
lines by ∼10×, but the numbers would still be daunting. magnitude less than for an equivalent number of discrete
The 4 K temperature stage that exists in the upper portion commercially available coaxial cables. The cable terminations
of typical dilution refrigerators provides a particularly could be bonded to a chip using a solder (Pb or In), either in a
attractive location for a layer of control electronics (including closely spaced linear array for contact near the edge of a chip,
basic quantum error correction functions) [23, 24]. Heat dis- or in a less densely spaced two-dimensional array, which
sipation capabilities of multiple Watts are currently available would mate to the face of a chip or to a connector.
from pulse-tube systems, and it is easy to imagine how the In this work we have fabricated and characterized
cooling capacity could be further increased using an exter- somewhat simpler structures than proposed in figure 1, spe-
nally supplied flow cryostat. Large numbers of ultralow- cifically individual microstrip transmission lines, as a step-
power superconducting electronics can coexist with moderate ping stone towards our ultimate goal. Figure 2(a) is a
quantities of semiconductor circuits at 4 K. One can imagine photograph of one such assembled flexible cable, which is a
that sufficient electronics could exist at the 4 K stage so that functional superconducting Nb microwave transmission line
the bandwidth needed to connect to the room-temperature built on a thin-film polyimide dielectric. The performance of
environment is greatly reduced. In this case, the main inter- these cables, as described herein, gives us confidence that
connection problem becomes the links from 4 K down more sophisticated highly parallel cables are indeed feasible.
to 20 mK. This paper is organized as follows. In section 2 we
We believe that thin-film flexible superconducting describe prior work and estimate thermal performance of the
transmission lines, fabricated by adapting techniques from the superconducting thin-film cables. Section 3 presents concepts
microelectronics industry, can help address this interconnec- and details for the design of the microstrip test structures. In
tion challenge. Owing to their very low rf surface impedance, section 4, we describe relevant fabrication and measurement
superconducting conductors can be scaled to quite small details. Section 5 covers results for transmission lines, reso-
dimensions. Excellent microwave signal propagation can be nators, non-idealities and dielectric loss. Finally, conclusions
achieved with conductor cross sections that are orders of and closing comments are presented in section 6.
magnitude smaller than commercially available coaxial
cables. In principle, as illustrated in figure 1, one could
construct many parallel thin-film transmission lines within a 2. Relationship to prior work and thermal projections
single ribbon cable, with superconducting shield walls (or at
least multiple closely spaced vias between upper and lower A number of researchers have reported making low-frequency
ground planes) surrounding each signal conductor to form (dc to low MHz) superconducting thin film flexible cables
‘rectangular coax’ cables having negligible crosstalk (an [25–30], but none have presented microwave (GHz range)
important consideration in many qubit experiments). Typical performance data. Manufacturer’s data sheets for thin-film
dimensions of the conductors can be 20 μm wide and polyimides generally only provide low-frequency (kHz or
∼250 nm thick on a pitch as fine as 50 μm, which would yield MHz) room-temperature loss tangent (tanδ) data, which is not
a very high interconnect density (up to 200 signals in a 10 mm useful for our purposes. Some cryogenic loss measurements
wide cable), excellent mechanical flexibility, and very low have been reported for polyimide, but again only at very low
axial heat leakage when in the superconducting state. Total frequencies [31–33]. Ponchak and Downey [34] have mea-
dielectric (e.g., polyimide) thickness could be 50 μm or less. sured room-temperature microwave properties of thin-film
Nb (Tc ∼ 9.2 K) would be an acceptable superconductor for copper/polyimide microstrip lines on silicon substrates over
many applications, although alternative materials could be the 1–110 GHz frequency range and reported an average tanδ
used if needed for high-magnetic field environments. Either of 0.006 for PI-2611. Harris et al [35] measured copper
conventional single-ended or differential pairs could be microstrip lines on free-standing Kapton polyimide and

3
Supercond. Sci. Technol. 29 (2016) 084007 D B Tuckerman et al

Figure 2. (a) Assembled thin film PI-2611 (meandered) microstrip transmission line with edge launch SMA connectors and support substrate,
(b) Released thin film resonator with signal trace, prior to deposition of ground metallization and (c) Cross sectional structure of the
simplified structures tested in this work, which were individual superconducting microstrip transmission lines and resonators fabricated on
20 μm thick polyimide, with lengths ranging from 50 to 550 mm. Based on the measured relative dielectric constant of the polyimide at low
temperature (er ) of 3.2, we chose a line width of 47.4 μm to yield a 50 Ω characteristic impedance. Our early test structures were straight
50 mm long lines, and we have since successfully tested serpentine transmission lines and resonators having total lengths of up to 550 mm.

reported a room-temperature tanδ of 0.013 at 13 GHz, which conductive, e.g. only 0.6 nW of heat leakage from 800 mK to
fell to 0.007 when cooled to 77 K. We will demonstrate in this lower temperatures.
work that, remarkably, polyimide microwave dielectric losses In contrast to the flex cables in [29], where the thermal
are reduced by roughly two orders of magnitude when cooled conductivity of their Nb traces were deemed negligible based
to deep cryogenic temperatures (4 K and below). on a 2000:1 polyimide/Nb area ratio, we use ground planes in
Using thin-film geometries similar to those in our work, our devices, for shielding and precise impedance control at
van Weers et al [29] have measured the axial thermal con- microwave frequencies. Our polyimide/Nb ratio for micro-
ductivity of a released layer of thin-film PI-2611 to be strip is in the range between 40:1 and 80:1 depending on the
1.6 ´ 10-3T1.17 W m−1 K−1 over the temperature range signal density, and so we cannot neglect the Nb; in fact it may
150 mK to 10 K. In the absence of convective or radiative be the dominant contributor to axial heat leakage. The
heat transfer, it can be shown by integrating the one-dimen- reported thermal conductivity of Nb with an RRR of 40 from
sional Fourier law that whenever a cable’s axial thermal T=4 K down to T=1.5 K ranged from 9 to 3 W m−1 K−1,
conductivity has that type of power-law dependence on respectively [36]. Lower-temperature measurements of bulk
absolute temperature, i.e., of the form k = aT n , then the Nb with RRR=26 have shown that the thermal conductivity
steady-state heat Q̇ conducting along a cable of uniform drops monotonically starting from ∼2 W m−1 K−1 at 0.7 K,
cross-sectional area A and length L having a ‘hot’ end at and diminishing to ∼0.002 W m−1 K−1 at 40 mK, following
temperature Thot and a ‘cold’ end at Tcold will be fairly close to the theoretical T 3 dependence (a T 2.4 depend-
Q˙ = (khot Thot - k cold Tcold ) A [(n + 1) L ], where khot and kcold ence looks like a best fit) [37]; kNb∼0.028 W m−1 K−1 at
are the thermal conductivities at the hot and cold ends. When 120 mK using this fit. In contrast to these bulk samples, our
Thot Tcold  1, as is the case in our contemplated applications, Nb thin films have an RRR of ∼3, so we might expect both
this simplifies to Q˙ = khot Thot A [(n + 1) L ]. We can thus electronic and phononic contributions to thermal conductivity
predict that the heat conduction of a 10 mm wide, 20 μm to be reduced by an order of magnitude, but to be very
thick polyimide ribbon spanning a 150 mm distance from the conservative we will assume the bulk values here. The heat
4 K stage of a dilution refrigerator to the next lower stage leaks of greatest concern in a dilution refrigerator are nor-
would be only 20 nW, which is insignificant. Cables with a mally those loading the lowest temperature stage (mixing
‘hot’ end at a lower temperature stage would be even less chamber), where one wants to maintain T∼20 mK. The next

4
Supercond. Sci. Technol. 29 (2016) 084007 D B Tuckerman et al

wavelength, series microstrip transmission line resonators

using 250 nm thick Nb on 20 μm thick PI-2611 or HD-4100
films. A linewidth of 47.4 μm was used to provide a nominal
50 Ω characteristic impedance. In order to make reliable
microwave connections, we used a 120 μm (Ws)×1200 μm
solder pad as shown in figure 3, which was designed to mate
Figure 3. Layout of end section of Nb resonator. The length of the to Southwest Microwave edge launch SMA connectors. The
resonator region is 46.1 mm, Ls is 1200 μm, Ws is 120 μm, LfNb is transmission line lengths ranged from 50 mm (straight) to
100 μm, WCNb is 20 μm, WrNb is 47.4 μm, and LCNb ranges from 300 550 mm (meandered). For the resonators, the solder pads
to 500 μm.
were connected to 47.4 μm (WrNb)×100 μm (LfNb) traces as
feeding structures at both ends of the resonator. The resonator
temperature stage would often be at approximately 120 mK, line length was 46.1 mm, corresponding to a fundamental
and so we can assume that cabling is thermally tied at that resonance frequency ( f0) of approximately 2 GHz. The reso-
stage. Using the same calculation method as for the polyimide nator coupling structure is shown in figure 3. The gap
case above, but with the n=2.4 power law, we estimate that between the coupling lines is fixed at 20 μm (WCNb) and the
the ground plane of a 10 mm wide, 150 mm long microstrip coupling strength is determined by choosing the overlap
cable would conduct only 16 pW of heat; again, this is quite length LCNb. We simulated and fabricated resonators with
negligible, particularly since such a cable could contain as different coupling strengths (LCNb=300, 400, 500 μm). We
many as 100 parallel transmission lines. note that LCNb=300 μm was the weakest coupling in the
present designs (an effective series coupling capacitance of
2.76 fF at each end), which turned out to be the preferred
3. Design of microstrip transmission line structures choice in view of the unexpectedly low losses of our
on thin film polyimide resonators.

The structures fabricated and tested in this work are micro-

strip transmission line elements on thin-film dielectrics that 4. Fabrication and measurement details
were spin-deposited onto Si substrate wafers and later
released to produce free-standing flexible films. Both PI-2611 Si wafers were used as handle wafers and temporary support
and HD-4100 polyimides from HD MicroSystems were substrates for fabrication of the thin-film transmission lines
characterized in this work. In our experience, superconductors and resonators. To start, a Cr/Al layer was deposited onto the
deposited on these thin-film dielectrics exhibited higher Si wafers for use as a release layer later in the fabrication
quality than when we use commercially supplied free-stand- flow. Polyimide (PI-2611 or HD-4100 from HD Micro-
ing Kapton films, because the surface quality is smoother and Systems) was deposited by spin-coating to an approximate
has fewer defects [38]. PI-2611, while not photodefineable, is thickness of 10 μm. The films were then cured, using standard
expansion-matched to silicon, exhibits excellent insulation curing profiles, in a nitrogen ambient oven at 350 °C. A
characteristics and high material flexibility, and is widely second polyimide layer of the same type was then deposited
used as a microwave device packaging and substrate material and cured on top of the first layer to achieve a total dielectric
[39]. Photodefineable HD-4100 is also widely used in device thickness of 20 μm. The resonator traces were defined
packaging since it can be directly patterned using photo- through a standard photolithographic lift-off process. In our
lithography. Our initial resonator designs were based on the initial samples, we first sputter-deposited a 10 nm thick Ti
dielectric properties listed on the manufacturer’s data sheets, layer to promote adhesion and potentially improve the Nb
which report PI-2611 to have a relative dielectric constant film quality, as this had been necessary in earlier work on
(er )=2.9 at 1 kHz at room temperature [40]; HD-4100 is Kapton films [38]; we later found this extra Ti layer to be
reported to have er =3.36 at 1 MHz [41]. In order to better unnecessary when depositing Nb onto thin-film polyimides,
determine er at temperatures and frequencies of interest and and so it was omitted in later samples. Instead, prior to the Nb
avoid confounding effects due to superconductor kinetic deposition, a brief Ti getter-pumping step was performed, in
inductance, which can cause a shift in resonant frequency as which the wafers were stationary behind a shield, while Ti
the temperature changes, Cu/polyimide resonators were was evaporated into the chamber at a rate of 2 Å s−1 for 250 s.
fabricated and their S-parameters were measured at 6.5 K. We After this gettering process, wafers were then rotated at 22
used Keysight Advanced Design System (ADS 2015) for RPM while 250 nm of Nb was sputter-deposited (1 kW dc, 30
simulation of the resonator microwave response and fit the minutes, 6.1 mTorr of Ar). Vacuum base pressures of
measured data by adjusting the dielectric constant in the ∼3×10−7 were attained prior to depositing the Nb, which
simulation. The results indicate the dielectric constant is quite gave fairly repeatable properties. In order to improve the
constant over a frequency range from 1 to 10GHz, and was adhesion between the Nb and the polyimide, an in situ ion
determined to be er = 3.2. HD-4100 was assumed to have a milling process was performed for 120 s on the surface of the
similar value for er and this was later confirmed by super- polyimide before the sputtering process. Si wafer pieces with
conducting resonator measurements. Based on our fit results, a SiO2 surface were included in the same Nb deposition runs
we designed a series of microstrip transmission lines and half- and used as witness samples for characterization of the sheet

5
Supercond. Sci. Technol. 29 (2016) 084007 D B Tuckerman et al

resistance and superconducting critical transition temperature sections are corrected for system attenuation/amplification,
(Tc) of the deposited Nb. Typical sheet resistances of our unless otherwise indicated.
250 nm thick Nb films ranged from 1 Ω/square to 2.1 Ω/ In all our measurement set-ups, no intentional magnetic
square at room temperature. Typical Tc values for the Nb shielding was used, so it is quite possible that flux vortices
layers used in this work ranged from 8.7 to 9.1 K, with an threading the Nb thin films reduced our resonator Q values
RRR (R280 K R10 K ) between 2.5 and 3.5. below what would be observed in an ideal field-free
After the Nb deposition, a lift-off process was performed environment [43].
to complete the definition of the signal traces. The samples
were then patterned for solder contact pad areas (i.e., under
bump metallization, UBM). We used electron beam physical
5. Results and discussion
vapor deposition to deposit Ti (50 nm), then Cu (500 nm),
onto the samples. The samples were then protected with
In this section, we describe and discuss results for super-
photoresist and released in a NaCl solution by anodic dis-
conducting microstrip transmission lines and resonators fab-
solution [42]. A released and cleaned structure is shown in
ricated on thin-film polyimide. These structures are all free-
figure 2(b). After removing the protective photoresist, we
standing films, supported only at the ends by the edge-launch
mounted the free-standing film onto another Si support wafer connectors. We also discuss non-ideal low power and high
with the non-metallized back side of the film exposed. The power effects, as well as extraction of upper bounds on tanδ
sample was then loaded into the deposition system for back- for the polyimide at various temperatures.
side Nb metallization, with the same thickness as the signal
layer. A schematic cross-section of the transmission line
structure is shown in figure 2(c). 5.1. Performance of superconducting thin-film polyimide
In order to facilitate suitable microwave connections to transmission lines
the flexible transmission lines and resonators, we used edge
launch SMA connectors. Due to the small dimensions of the S-parameters were measured for superconducting Nb trans-
mission lines at frequencies up to approximately 14 GHz and
signal solder pad, we used Southwest Microwave connectors
at an incident microwave power of −20 dbm, while immersed
with the smallest available pin size (0.005″, i.e. 127 μm). In
in LHe (4.2 K). SOLR calibration was used, requiring several
order to achieve reliable connections for use at cryogenic
immersion cycles [44]. Representative S21 results for a 50 mm
temperatures, we soldered the signal pin from the SMA
long line are shown in figure 4(a). The observed oscillations
connector to the solder pad on the flex using high purity In
are due to alternating constructive and destructive interference
solder. Additionally, we prevented the hefty SMA connectors
from impedance discontinuities at the two ends of the near-
from straining and damaging the thin flexible structures by
lossless transmission line. Specifically, the connector pin
mounting the connectors and the sample onto a support board,
attachment points required a wide soldering pad (see figure 3)
as shown in figure 2(a).
and additional solder, resulting in parasitic capacitance to
Measurements above ∼1 K were carried out in either a
ground. To confirm this explanation, Keysight ADS was used
LHe dewar (4.2 K), or a Cryo Industries of America pulse- to simulate the response of the transmission lines. We used
tube based cryostat with stainless steel cryogenic rf coaxial dielectric loss tangent values derived from measurements of
cables. The pulse-tube cryostat with the rf cables stabilizes at superconducting resonators (described below) fabricated on
approximately 3 K, and can temporarily cool to ∼1 K by the same substrate as the transmission lines. Excellent
pumping on the sample space that has been back-filled with matching of simulation to measurement was obtained and is
He. Thermometry was performed at the sample holder near shown as simulation results ‘with pads’. The variables in
the top of the sample. The sample was primarily thermalized these simulations were the kinetic inductance of the super-
through the He exchange gas. A performance network ana- conductor and the two values of the extra capacitance (one at
lyzer (Keysight N5227A PNA) was used for measurement of each end of the line, and not necessarily symmetric due to
the scattering parameters. variability in solder quantity). These capacitive pads and
The milliKelvin measurements were performed in a 3He/ accompanying extra solder, which result from the current
4He dilution refrigerator (Leiden Cryogenics CF450) with method of making contact to the structure, are expected to be
base temperature of 20 mK. High frequency coax lines were eliminated with a more refined connection method, such as
attenuated at each temperature stage to reduce noise and to flip-chip attachment. Therefore, we also performed simula-
thermalize the inner conductors. The Southwest Microwave tions without these capacitive instances and show these
SMA connectors at each end of the flex cable were therma- results as simulation ‘without pads’ in figure 4(a); these
lized to the mixing chamber via SMA feedthroughs (for the simulation results closely coincide with the constructively
grounds) and cryogenic attenuators (for the center con- interfering peaks (top envelope) of the S21 response, as
ductors). Transmission measurements were performed using a expected. These results demonstrate that, with ideal signal
vector network analyzer (Keysight N5245A), with room launchers, insertion losses less than approximately 0.1 dB up
temperature amplification (Low Noise Factory LNF- to at least 14 GHz can be achieved for a 50 mm long super-
LNR1_15A) and cryogenic amplification on some samples conducting microstrip transmission line constructed from Nb
(Caltech CITLF). Measurement results in the following on flexible polyimide substrates.

6
Supercond. Sci. Technol. 29 (2016) 084007 D B Tuckerman et al

Figure 4. Measurement and simulation data of: (a) S21 for a 50 mm long flexible superconducting microstrip transmission line (Ti/Nb on PI-
2611). Simulation results are shown both with and without the terminal solder pads, to demonstrate the effects of those parasitic capacitive
structures. (b) S21 for a 5× longer (i.e., 250 mm), otherwise nominally identical microstrip line. These samples were immersed in liquid He
(T=4.2 K) and tested using −20 dBm incident power; results at higher powers were identical up to 0 dBm, beyond which 3rd-order
nonlinearities started to become discernible in the power spectrum.

We also fabricated and tested a 5× longer transmission

line (in the form of a 250 mm serpentine). S21 for this trans-
mission line is shown in figure 4(b), along with a simulation
that includes the extra capacitance. Excellent matching
between measurement and simulation up to 12 GHz is evident
for this structure, as well; the discrepancies above 12 GHz are
believed to be imperfections in the calibration procedure.
These results show the feasibility of fabricating functional,
low-loss flexible superconducting transmission lines with
lengths sufficient to span adjacent temperature stages in a
typical dilution refrigerator.
The dc critical current (Ic) of a representative microstrip
signal conductor was measured as 11 mA at 4.2 K. Although
this is well below the critical current density of high-quality
Nb films, we have evidence that the reduced Ic is at least
partly due to various defects in the Nb film, so we can pre-
sumably increase Ic in the future by making process
Figure 5. Fundamental (∼2 GHz) resonance of Nb on HD-4100
improvements, if it becomes necessary. Ic=11 mA would microstrip resonator at 1.2 K and 20 mK, showing measured S21 and
seem to imply that the transmitted rf power is limited to fits to a Lorentzian function. Data has been plotted normalized to
∼5 dBm, however in the GHz band we were able to transmit center frequency at each temperature. Center frequency, 3 dB
at least 10 dBm with negligible loss, although a small amount bandwidth, and resultant loaded Q-factor are also provided for each
of nonlinearity in the form of a 3rd harmonic began to appear temperature.
as the incident power was increased above 0 dBm.
S21 and S12 were measured up to 20 GHz at various tem-
peratures (4.2, 3.6, 3 K, ∼1 K and 20 mK). Care was taken to
5.2. Performance of superconducting thin-film polyimide
obtain sufficient measurement points in a narrow frequency
resonators
range around each resonance in order to achieve a reasonable
Due to the low loss of superconducting transmission lines of fit to a Lorentzian line shape, which was characteristic of the
manageable lengths, detailed characterization of the loss resonances in these structures and allowed extraction of the
mechanisms is challenging. Instead, we measured the quality loaded Q-factor. Figure 5 shows a Nb/HD-4100 resonator
factors (‘Q’) of structurally similar resonators, which provide with loaded Q > 17 000 at 1.2 K and Q > 42 800 at 20 mK.
a sensitive probe of the combined dielectric and super- In each of these measurements, the resonance is for a signal
conductor losses [45]. For each of the fabricated resonators, power level that provides the highest Q. Power levels

7
Supercond. Sci. Technol. 29 (2016) 084007 D B Tuckerman et al

since the Q of a resonator varies inversely with the wave-

length, this nets a linear dependence between 1/Q and fre-
quency, which is what we generally observe. (We note that
the 2nd harmonic (∼4 GHz) response is sometimes an outlier,
showing higher loss than the trend line. This effect has been
intermittently observed on some samples, and may be caused
by surface contamination of the exposed Nb signal trace, and/
or by unintended resonant interactions with the sample
chamber. These anomalies are being investigated further.)
The BCS losses should fall exponentially as T/Tc is
reduced; this is consistent with figure 7(a), which shows a
substantial reduction in slope as temperature reduces from 4.2
to 3 K. The ∼1 K data shows a slope of nearly zero; at this
temperature, the Nb has minimal BCS loss and the Q-factor is
presumed to be almost entirely due to dielectric losses,
although we cannot rule out the possibility of other residual
Figure 6. S21 of fundamental resonance versus frequency at multiple losses that might be associated with the superconductor, with
temperatures for a Nb on HD-4100 resonator. various interfaces, or with unintended electromagnetic inter-
actions with the environment.
Assuming that the lowest-temperature (1.2 K and 20 mK)
significantly below this value resulted in lower Q values at Q values are indeed primarily due to polyimide loss, the near-
20 mK but not at 1 K or above, while much higher power zero slopes in 7(a) are indicative of a near-constant loss
levels caused various nonlinear effects and distortions of the tangent value across the 2–12 GHz frequency range. This was
resonance peak at any temperature. Further details of these not unexpected, since  r was almost constant at 3.2 across this
effects are described later in this paper. frequency range (as evidenced by the resonator harmonics
In figure 6, we show the impact of sample temperature on occurring at almost exact integer multiples of fo). Significant
the Q-factor and center frequency f0 of the fundamental changes in loss tangent versus frequency over a large fre-
resonance. Reduction of the sample temperature causes a shift quency range normally coincide with noticeable variations in
in the center frequency towards a higher frequency and an  r , a consequence of causality (i.e., the Kramers–Kronig
increase in the Q-factor. The center frequency shifts can be relations).
explained by a reduction of the superconductor kinetic Extrapolating the linear fits in figure 7(a) to zero fre-
inductance with decrease in sample temperature. Furthermore, quency provides an estimate of actual dielectric loss by
as the sample is cooled further below the Tc, there is a removing the BCS loss contribution (provided that the loss
reduction of quasiparticles (un-paired electrons), as expected tangent is in fact nearly frequency independent). It is evident
from BCS theory, which leads to a reduction in attenuation that this intercept point falls with decreasing temperature,
from surface resistance, as well as further reduction in di- suggesting that the loss tangent is continuing to fall even at
electric loss tangent, both of which increase the Q-factor. We these extremely low temperatures. To provide additional
note that in figure 6, the 3.6 K data and 3 K data are quite insight into the temperature dependence of the dielectric loss
similar. A possible explanation for this is the existence of an of the polyimide, figure 7(b) plots 1/Q for the fundamental
interfacial region between the dielectric and Nb super- frequency as a function of temperature for each type of
conductor that has different superconducting properties, such resonator. This data suggests an approximately linear rela-
as Tc. Another possibility is that the overlap results from a tionship between dielectric loss and absolute temperature,
limitation of the temperature stability of the cryogenic system although again we cannot rule out the possibility that the Nb
used for these measurements at temperatures that are not superconductors are also contributing some residual loss,
inherently stable in this system. We estimate that the temp- even though the basic BCS theory predicts negligible losses at
erature stability of the pulse-tube based cryostat is within and below 1.2 K for Nb.
0.5 K at temperatures below 4.2 K.
In order to determine the temperature below which
5.3. Impact of signal power on Q-factor
superconductor loss becomes negligible, we generated 1/Q
versus frequency plots, as shown in figure 7(a). In order to Q-factors of these resonators are power dependent, as shown
consider measurement error and system temperature drift, in figures 8 and 9. These effects differ significantly depending
each of the measured Q-factors is the average of ten indivi- on the temperature regime. Figure 8 illustrates typical beha-
dual measurement results (for the same sample cool-down) vior at 3 K. As the incident power is increased above
with a 3σ error bar. Based on superconducting transmission −25 dBm, the resonant frequency trends lower, the Q
line resonator theory, a 1/Q versus resonant frequency plot decreases, and the resonance exhibits nonlinearity. This can
should yield a straight line, with a slope proportional to the be explained by the current concentration at the edges of the
magnitude of quasiparticle-induced losses (‘BCS losses’). Nb signal trace, which causes locally increased kinetic
BCS losses actually increase as the square of frequency, but inductance and ultimately loss of superconductivity. For the

8
Supercond. Sci. Technol. 29 (2016) 084007 D B Tuckerman et al

Figure 7. (a) 1/Q (loaded) for multiple harmonics of a Nb/HD-4100 microstrip resonator at different temperatures. Dashed lines are linear
fits. (b) 1/Q (loaded) as a function of temperature for the fundamental (∼2 GHz) resonance of several PI-2611 and HD-4100 resonators.

nonlinear behavior that reduces Q, although the shapes of the

resonances suggest that different loss mechanisms are
operative at these low temperatures. For example figure 9(b)
shows a flat-top profile that may indicate switching of a weak
link within the resonator when the local rf current exceeds a
critical value. (We observed that after driving the resonator
normal with high-power rf and then removing the high power,
the resonator would return to its high-Q value after about 5 s,
indicating that the thermal anchoring was effective.)
Figure 9(c) suggests a Kerr nonlinearity associated with
partially saturated two level states [46]. We also observe a
new effect in figure 9 that was not observed at higher tem-
peratures, which is a reduction in Q at very low power levels;
this reduction levels off at sufficiently low power. This
behavior is likely due to unsaturated two level systems in the
polyimide and/or surface oxides on the Nb, which also occur
in amorphous dielectric materials such as SiO2 in this milli-
Figure 8. Fundamental harmonic response of Nb on PI-2611
resonator at 3 K for different power levels incident at the sample. Kelvin temperature regime [47–51].
The range of loaded Q values is shown. Due to the obvious To compare the different materials tested in this work, we
nonlinearity at higher powers, Q was calculated from the measured plotted the high power and low power Q values at each
3 dB bandwidths, rather than fitting to a Lorentzian function. harmonic to get an idea of the frequency-dependence of the
loss tangent. In figure 10(a) we plot the asymptotic low-power
data in figure 7, we used drive power levels just below the Q values, and in figure 10(b) we plot the moderate-power Q
onset of this nonlinearity. values, i.e., just below the onset of nonlinearity. We note that
Figure 9 illustrates representative behavior observed at above 10 GHz, the incident power was limited by cabling
20 mK for our resonators. Again, high powers induce losses and attenuators in the dilution refrigerator, so we were

9
Supercond. Sci. Technol. 29 (2016) 084007 D B Tuckerman et al

Figure 9. (a) Loaded Q versus incident power at the sample for the first four harmonics (2–8 GHz) of a Ti/Nb on PI-2611 resonator at 20 mK,
(b) fundamental resonance showing high power effects of high Q-factor and resonance peak clipping, and (c) fundamental frequency
resonances showing Duffing-type nonlinearity due to high power effects.

still partially in the low-power regime, which artificially From Qd, we can determine tanδ from:
depresses those Q values.
1 ⎛ 1 - q⎞
Qd = ⎜1 + ⎟, (3 )
5.4. Dielectric loss tangent extraction tand ⎝ qer ⎠

To allow more precise modeling and performance predictions where q is the dielectric filling factor, which is not unity since
for the flexible superconducting transmission lines, knowl- this is a non-embedded microstrip structure. From ADS
edge of the dielectric tanδ for the various polyimide materials simulation results, with er =3.2 and for the case of the
is needed. As shown previously, we extracted Q-factors at microstrip transmission line stack-up used in this work, we
multiple resonant frequencies and multiple temperatures. The find q=0.703. Based on 3, we determine the tanδ of these
extracted Q is the loaded quality factor (Ql) and can be two types of polyimide films at various frequencies and
described by: temperatures. The loaded Q-factor and corresponding tanδ
(corrected for coupling loss) at ∼1.2 K (for both types of
1
=
1
+
1
+
1
+
1
(1 ) polyimide) and 20 mK (for HD-4100), are shown in table 1.
Ql Qc Qd Qr Qcoup We emphasize again that the existence of other parasitic
losses (i.e., not associated with the polyimide dielectric)
where Qc, Qd, Qr and Qcoup are the Q-factors associated with cannot be ruled out, and so these calculated loss tangents
conductor loss, dielectric loss, radiation loss and coupling should be viewed as worst-case upper bounds for the actual
loss, respectively [52]. Radiation losses are assumed to be dielectric loss tangent.
negligible for these resonators. Given the previously
presented 1/Q results (see figure 7), we assume that the
conductor (BCS-related) losses are negligible at temperatures 6. Conclusion and future work
below ∼1.2 K. Furthermore, ADS simulation results for
resonators with no conductor loss and tanδ=0 provide Flexible superconducting transmission line cables, with small
values for the coupling loss at each resonant frequency. These physical and thermal cross sections, can be an enabling
values are provided in table 1. With this in mind, we technology for future computing technologies, such as
mathematically correct for the coupling loss to get Qd quantum computing. Our results so far give us confidence that
according to: more sophisticated structures can be built. Future work is
1 1 1 planned to characterize crosstalk and create embedded and
= - . (2 ) more highly shielded structures. Crosstalk between parallel
Qd Ql Qcoup
microstrip conductors follows an inverse square law and so

10
Supercond. Sci. Technol. 29 (2016) 084007 D B Tuckerman et al

Figure 10. Internal Q-factors (corrected for coupling loss) versus frequency for PI-2611 and HD-4100 resonators at 20 mK measured in the
(a) low power regime and (b) moderate power regime. Results shown here are for five different samples, with the Ti/Nb on PI-2611 being
nominally identical.

Table 1. Measured loaded Q-factor and calculated dielectric loss tangent tanδ at multiple frequencies at ∼1.2 K for PI-2611, and at both 1.2 K
and 20 mK for HD-4100. The tanδ calculation corrected for the coupling Q and the non-unity dielectric filling factor of the microstrip. We
assume no other loss mechanisms at these low temperatures, and so the actual loss tangents may be smaller than shown.
Loaded Q-factor (tanδ)
∼f0 (GHz) Qcoup PI-2611 @ 1.2 K HD-4100 @ 1.2 K HD-4100 @ 20 mK
−5 −5
2 251 000 13 200 (8.21×10 ) 17 300 (6.07×10 ) 42 800 (2.19×10−5)
4 139 000 7 880 (1.35×10−4) 14 100 (7.19×10−5) 7 070(1.52×10−4)
6 106 000 10 500 (9.72×10−5) 13 400 (7.34×10−5) 23 300 (3.78×10−5)
8 88 400 9 170 (1.10×10−4) 12 000 (8.17×10−5) 15 500 (6.00×10−5)
10 77 300 9 660 (1.02×10−4) 10 400 (9.42×10−4) 10 000 (9.83×10−5)
12 75 400 8 870 (1.12×10−4) 9 710 (1.01×10−4) —

can be made arbitrarily small by spacing the conductors suf- options for wider and longer cables. In the long run, we
ficiently far apart, but stripline allows much closer spacing for suggest that these sorts of cables could be most economically
the same amount of crosstalk. Shield vias, as illustrated in fabricated using large glass panel substrates, such as are
figure 1, will reduce crosstalk even further and allow maxi- commonly used to manufacture LCD displays.
mally dense conductor spacing. The principal challenge to Connectors could be integrated or flip-chip attached to
making stripline is the degradation in the superconducting the ends of the cables. Similarly, attenuators and thermali-
properties (reduced Tc and Ic) of the Nb films caused by sub- zation structures could be incorporated in various ways, either
sequent high-temperature curing of polyimide. We have as integral thin-film patterns or as discrete chips that are flip-
recently found that reducing the polyimide curing temperature chip bonded to the cables. A large number of transmission
substantially alleviates this problem, and so are quite optimistic lines could be integrated within a single relatively narrow
about the prospects for building more highly shielded cables. ribbon, with outstanding phase matching owing to the inte-
Using a spiral pattern on 100 mm diameter wafers, we grated construction.
have built 2 mm wide cables which were fully 1 m long after In addition to our flexible cable applications, the extre-
being released. The use of 300 mm wafers would offer more mely low loss exhibited by polyimide thin films at deep

11
Supercond. Sci. Technol. 29 (2016) 084007 D B Tuckerman et al

cryogenic temperatures suggests that polyimides may also be [24] Homulle H, Visser S, Patra B, Ferrari G, Prati E,
viable dielectrics for signal distribution on rigid substrates, Sebastiano F and Charbon E 2016 A reconfigurable
e.g., for fabricating quantum integrated circuits. cryogenic platform for the classical control of scalable
quantum computers arXiv:1602.05786
[25] Yung C S and Moeckly B H 2011 IEEE Trans. Appl.
Supercond. 21 107–10
Acknowledgments [26] Cheng M Y, Park W T, Yu A, Xue R F, Tan K L, Yu D,
Lee S H, Gan C L and Je M 2013 Microsyst. Technol. 19
1111–8
We gratefully acknowledge funding and technical guidance [27] Tighe T, Akerling G and Smith A 1999 IEEE Trans. Appl.
from Microsoft Research for this work, including valuable Supercond. 9 3173–6
discussions with Burton J Smith. We thank Y Cao of [28] Chervenak J and Mateo J 2013 Flexible microstrip circuits for
AMNSTC for assistance with sample fabrication. superconducting electronics Technical Report GSC-16718-1
(Washington DC: NASA)
[29] van Weers H, Kunkel G, Lindeman M and Leeman M 2013
Cryogenics 55–56 1–4
References [30] Bruijn M P, van der Linden A J, Ridder M L and
van Weers H J 2015 J. Low Temp. Phys. 184 369–3
[31] Simon N 1994 Cryogenic properties of inorganic materials for
[1] Barends R et al 2014 Nature 508 500–3 iter magnets: a review Technical Report NISTIR 5030
[2] Reed M D, DiCarlo L, Nigg S E, Sun L, Frunzio L, (Gaithersburg, MA: NIST)
Girvin S M and Schoelkopf R J 2012 Nature 482 382–5 [32] Weedy B and Swingler S 1984 Cryogenics 24 367–70
[3] Kelly J et al 2015 Nature 519 66–9 [33] Yamaoka H, Miyata K and Yano O 1995 Cryogenics 35 787–9
[4] Corcoles A D, Magesan E, Srinivasan S J, Cross A W, [34] Ponchak G E and Downey A N 1998 IEEE Trans. Compon.
Steffen M, Gambetta J M and Chow J M 2015 Nat. Packag. Manuf. Technol. B 21 171–6
Commun. 6 6979 [35] Harris A, Sieth M, Lau J, Church S, Samoska L and Cleary K
[5] Ofek N et al 2016 arXiv:1602.04768 2012 Rev. Sci. Instrum. 83 086105
[6] Fowler A G, Mariantoni M, Martinis J M and Cleland A N [36] Koechlin F and Bonin B 1999 Supercond. Sci. Technol. 9
2012 Phys. Rev. A 86 032324 453–60
[7] Mourik V, Zuo K, Frolov S M, Plissard S R, [37] Anderson A C, Satterthwaite C B and Smith S C 1971 Phys.
Bakkers E P A M and Kouwenhoven L P 2012 Science 336 Rev. B 3 3762–4
1003–7 [38] Bai R, Hernandez G A, Cao Y, Sellers A J, Ellis C D,
[8] Albrecht S M, Higginbotham A P, Madsen M, Kuemmeth F, Tuckerman D B and Hamilton M C 2016 Cryogenic
Jespersen T S, Nygård J, Krogstrup P and Marcus C M 2016 microwave characterization of kapton polyimide using
Nature 531 206–9 superconducting resonators International Microwave Symp.
[9] Hyart T, Van Heck B, Fulga I C, Burrello M, (IMS) (San Francisco)
Akhmerov A R and Beenakker C W J 2013 Phys. Rev. B 88 [39] Meyer J U, Stieglitz T, Scholz O, Haberer W and Beutel H
1–7 2001 IEEE Trans. Adv. Packag. 24 366–74
[10] Bravyi S and Kitaev A 2004 Phys. Rev. A 71 022316 [40] HD MicroSystems 2009 PI-2600 Series-Low Stress
[11] Brecht T, Pfaff W, Wang C, Chu Y, Frunzio L, Applications product bulletin (http://www.dupont.com/
Devoret M H and Schoelkopf R J 2016 npj Quant. Inf. 2 content/dam/dupont/products-and-services/electronic-and-
16002 electrical-materials/semiconductor-fabrication-and-
[12] Hornibrook J M et al 2015 Phys. Rev. Appl. 3 024010 packaging-materials/documents/PI-2600_Process
[13] Chow J M, Gambetta J M, Cross A W, Merkel S T, Guide.pdf)
Rigetti C and Steffen M 2013 New J. Phys. 15 115012 [41] HD MicroSystems 2014 HD-4100 Series product bulletin
[14] Leek P J, Filipp S, Maurer P, Baur M, Bianchetti R, Fink J M, (http://www.hdmicrosystems.com/HDMicroSystems/en_
Göppl M, Steffen L and Wallraff A 2009 Phys. Rev. B 79 US/products/photodefineable/HD-4100_negative_
180511 solvent.html)
[15] Wallraff A, Schuster D I, Blais A, Frunzio L, Majer J, [42] Metz S, Bertsch A and Renaud P 2005 J. Microelectromech.
Devoret M H, Girvin S M and Schoelkopf R J 2005 Phys. Syst. 14 383–91
Rev. Lett. 95 1–4 [43] Song C, Heitmann T W, DeFeo M P, Yu K, McDermott R,
[16] Macklin C, O’Brien K, Hover D, Schwartz M E, Neeley M, Martinis J M and Plourde B L T 2009 Phys. Rev.
Bolkhovsky V, Zhang X, Oliver W D and Siddiqi I 2015 B 79 174512
Science 350 307–10 [44] Ferrero A and Pisani U 1992 IEEE Microw. Guid. Wave Lett. 2
[17] Córcoles A D, Chow J M, Gambetta J M, Rigetti C, Rozen J R, 505–7
Keefe G A, Beth Rothwell M, Ketchen M B and Steffen M [45] Megrant A et al 2012 Appl. Phys. Lett. 100 113510
2011 Appl. Phys. Lett. 99 181906 [46] Suh J, Weinstein A and Schwab K 2013 Appl. Phys. Lett. 103
[18] Barends R et al 2011 Appl. Phys. Lett. 99 113507 052604
[19] Wenner J et al 2013 Phys. Rev. Lett. 110 150502 [47] O’Connell A D et al 2008 Appl. Phys. Lett. 92 112903
[20] Chen Y et al 2014 Phys. Rev. Lett. 113 220502 [48] Gao J, Daal M, Martinis J M, Vayonakis A, Zmuidzinas J,
[21] Hornibrook J M, Colless J I, Mahoney A C, Croot X G, Sadoulet B, Mazin B A, Day P K and Leduc H G 2008 Appl.
Blanvillain S, Lu H, Gossard A C and Reilly D J 2014 Appl. Phys. Lett. 92 212504
Phys. Lett. 104 103108 [49] Wang H et al 2009 Appl. Phys. Lett. 95 233508
[22] Chen Y et al 2012 Appl. Phys. Lett. 101 182601 [50] Wenner J et al 2011 Appl. Phys. Lett. 99 113513
[23] Conway Lamb I D, Colless J I, Hornibrook J M, Pauka S J, [51] Neill C et al 2013 Appl. Phys. Lett. 103 072601
Waddy S J, Frechtling M K and Reilly D J 2016 Rev. Sci. [52] Belohoubek E and Denlinger E 1975 IEEE Trans. Microw.
Instrum. 87 014701 Theory Tech. 23 522–6

12