Microsecond-lived quantum states in a carbon-based circuit driven by cavity photons

B. Neukelmance,1, 2, ∗ B. Hue,1, 2, ∗ Q. Schaeverbeke,2 L. Jarjat,1 A. Théry,1 J. Craquelin,1

W. Legrand,1, 2 T. Cubaynes,1 G. Abulizi,2 J. Becdelievre,2 M. El Abbassi,2 A. Larrouy,2 K.F. Ourak,2
D. Stefani,2 J.A. Sulpizio,2 A. Cottet,1, 3 M.M. Desjardins,2 T. Kontos,1, 3, † and M.R. Delbecq1, 3, 4, †
1
Laboratoire de Physique de l’École normale supérieure, ENS, Université PSL,
CNRS, Sorbonne Université, Université Paris Cité, Paris, France
2
C12 Quantum Electronics, Paris, France
3
Laboratoire de Physique et d’Etude des Matériaux, ESPCI Paris,
Université PSL, CNRS, Sorbonne Université, Paris, France
4
Institut universitaire de France (IUF)
arXiv:2410.19477v1 [cond-mat.mes-hall] 25 Oct 2024

Semiconductor quantum dots are an attractive platform for the realisation of quantum processors.
To achieve long-range coupling between them, quantum dots have been integrated into microwave
cavities. However, it has been shown that their coherence is then reduced compared to their cavity-
free implementations. Here, we manipulate the quantum states of a suspended carbon nanotube
double quantum dot with ferromagnetic contacts embedded in a microwave cavity. By performing
quantum manipulations via the cavity photons, we demonstrate coherence times of the order of
1.3 µs, two orders of magnitude larger than those measured so far in any carbon quantum circuit
and one order of magnitude larger than silicon-based quantum dots in comparable environment. This
holds promise for carbon as a host material for spin qubits in circuit quantum electrodynamics.

Spins isolated in nano-scale devices are a well- serve as an alternative to existing platforms. Our setup
established platform for solid-state quantum information displays several features which compare favorably with
processing [1, 2]. Although elementary multi-qubit sys- the state-of-the-art. First, we demonstrate coherent con-
tems have been implemented, scaling spin qubit archi- trol of quantum states in a carbon nanotube circuit with
tectures remains an outstanding challenge despite the a coherence time longer by about 2 orders of magnitude
progress made in recent years [3–5]. The hardware prob- than previously observed [17, 18]. Second, we are able
lem at stake in this endeavor constantly triggers the to perform all the basic operations of our carbon-based
search for new materials and new architectures [6]. circuit at zero external magnetic field and elevated tem-
One promising architecture, borrowed from supercon- perature T ≈ 300 mK compared to the driving frequency
ducting qubits, combines isolated spins and microwave fd ≈ 9 GHz. Third, the quantum states are solely con-
cavity photons in a circuit quantum electrodynamics trolled by cavity photons, which fully exploits the possi-
(cQED) setup [7–14]. The methods used in this context bilities of cQED. Finally, the coherence time of approxi-
are very appealing for upscaling spin qubit processors. mately 1.3 µs is the largest reported for quantum dots in
Recent breakthroughs towards this goal include the use of cavities for any material.
high-impedance cavities to enhance the spin-photon cou- In order to address long-lived quantum states in a
pling [9, 13], which has led to the first implementation of carbon-based circuit via cQED techniques, a spin-photon
a two-qubit gate between distant spins [14]. However, the coupling is engineered [7]. A suspended carbon nan-
coherence times observed for spins in cavities are essen- otube is connected to ferromagnetic electrodes with non-
tially two orders of magnitude smaller than their cavity- collinear magnetizations and coupled electrostatically to
free implementations, limiting the single- and two-qubit five gate electrodes. One of those is directly connected
gate fidelities. to the central conductor of a coplanar waveguide (CPW)
Among the many potential solutions, carbon nan- microwave resonator to induce charge-photon coupling.
otubes have shown great promise [15, 16]. In principle, Additionally, the ferromagnetic electrodes polarize lo-
carbon nanotube-based circuits display many attractive cally the spectrum of the carbon nanotube [19] which
features: they can form well-defined and tunable quan- induces a synthetic spin-orbit interaction enabling spin-
tum dots [16], they can be suspended, reducing impact of photon coupling [7]. This is illustrated in the panel a of
stray charges, and they can be grown with pure 12 C, of- figure 1. A typical device layout is shown in panels b,
fering a nuclear spin-free environment for electronic spin c and d. The CPW cavity, visible in panel b, is made
qubits. However, so far they have only been shown to out of a 100 nm layer of Nb. Its fundamental resonance
host quantum states with limited coherence of the order frequency at 300 mK is fc = 6.975 GHz with a quality
of 10 ns [17, 18]. factor Q = 4853. A close-up on the location of the dou-
In this work, we show that carbon nanotube-based ble quantum dot is shown in panel c. Large trenches
quantum devices finally hold to their promise and can are made for the final integration of the carbon nan-
otube [11, 20]. The zig-zag shaped ferromagnetic elec-
trodes are designed for creating the non-collinear mag-
∗ These two authors contributed equally to this work. netization and can be seen in panel d on either side of
† These two authors co-supervised the project. the gates. The ferromagnetic electrodes are made with
 2

RFIN VPR RFOUT

a tc b
GND plane

Trenches
εδ + εAC VPL

VΓL Vt VΓR

c d
Cavity field
K K'
θ
Trench Trench

Source Drain

ce

Drai
Sour
VΓL VPL Vt VPR VΓR

n
Gates

FIG. 1. Schematic and images of the device. a. Schematic of the device. A carbon nanotube is suspended above 5 DC
gates and connected to ferromagnetic contacts. The gates allow to form a double well potential in order to trap an electron. ϵδ
represents the depth asymmetry between the two wells. The electron can tunnel from one well to the other at a rate tc , thus
its wave function is delocalized over the two wells as represented by the filling of each circle. The non-collinear ferromagnetic
electrodes impose a different spin quantization axis in each well as shown by the orange arrows. In addition, the electron can be
in either valley K (red) or K ′ (blue). Finally the circuit is coupled to a microwave cavity via the gate VPR which is connected
to the central track of the coplanar waveguide resonator, thus inducing a modulation ϵAC on the energy detuning between the
two wells. b. Optical image of the sample. On the upper part, the coplanar waveguide resonator is connected to two ports.
The IN port is used to both fill the cavity and drive the nanotube circuit. The OUT port is used for reading-out the cavity.
In the middle, there are the trenches used to deposit the carbon nanotube. On the bottom part, there are the DC lines used
to form the double well potential. The bar is 1 mm. c. Optical image of a nanotube transfer region showing the large scale
view of the electrodes. The scale bar is 50 µm. d. Scanning electron microscope image of a typical device with a nanotube
transferred onto the ferromagnetic electrodes and suspended over the gates. The bar is 1 µm and θ = π/6.

a Ti(185 nm)/Ni80 Pd20 (35 nm)/Pd(4 nm) stack whereas ages, in the region of interest in this work. We observe the
the gates are made with a Ti(5 nm)/Pd(50 nm) stack. microwave signal characteristic of the transition between
The gate electrodes having a conducting surface ensure two adjacent charge states. In panel b, the DC current
that we can check electrically on the final device that the as a function of the magnetic field applied perpendicu-
nanotube is suspended and well isolated from the gates. larly to the nanotube axis is shown for different working
At the very last step of the nano-fabrication process, a points on the closest bias triangles where significant cur-
nano-scale transfer of a nanotube grown on a comb via rent is measurable. A gate dependent hysteretic signal in
a CH4 based chemical vapor deposition process is per- the current is observed. It validates spin injection [7, 24]
formed. It leads to device resistances between 0.5 MΩ and therefore the presence of interface exchange fields.
and 20 MΩ at room temperature. The particular device Turning to two-tone spectroscopy, we present in panel
studied here had a room temperature resistance of about c the phase contrast probed at fc as a function of the
8 MΩ. continuous drive frequency fd . The drive tone is applied
via the cavity at an input power Pd = −54 dBm (uncal-
All the measurements presented in this paper have ibrated). A resonance is found at fr = 9.0987 GHz with
been made at 300 mK. Note that this is an order of mag- a linewidth of 589 kHz extracted from the Lorentzian fit
nitude larger than conventional qubit operating temper- shown as an orange line in panel c. Furthermore, we ob-
atures, which has recently been shown to be favorable serve that the resonance frequency varies with respect to
for mitigating heating from microwave drive [21], how- the energy detuning ϵδ between the left and right plunger
ever we are not in the “hot” qubit regime [22, 23]. The gates, as shown in panel d, confirming that it is a transi-
main features of our device can be first characterized via tion of the double quantum dot spectrum. The observed
DC transport measurements and continuous wave spec- dispersion is strikingly small, a few MHz over hundreds of
troscopy. The corresponding measurements are shown in GHz for ϵδ , yet it is well captured by the theory as shown
figure 2. Panel a shows the phase shift of the microwave by the white dashed line (Supplementary Information) .
tone sent at fc as a function of the two plunger gate volt-
 3

-0.8 -0.4 0.0

a (°) b Second, there is a horizontal stripe with almost no os-
cillation contrast very close to the resonance. Neverthe-
18
0.50 less, the cut at fd = 9.0984 GHz shown in figure 3b dis-
plays well-defined damped oscillations which we fit with
VPR (V)

I (pA)
14 an exponentially decaying cosine. We extract a Rabi
frequency ΩR /(2π) = 2.51 MHz and a Rabi decay time
0.48 TRabi = (0.59 ± 0.05) µs.
10
The Rabi chevrons pattern is very sensitive to the spec-
0.79
VPL (V)
0.77 400 0
B (mT)
400 trum of the probed system. In particular, our unconven-
tional chevrons pattern tends to indicate a multi-level
0.0 0.25 0.5
c d (°) dynamics [25]. Furthermore, we observe that the Rabi
2/3
9.098 frequency ΩR scales as Ad with Ad the drive amplitude
0.10
(figure 3c). This scaling strongly points to the dynamics
associated with a quasi-harmonic spectrum [26]. In the
fd (GHz)
(°)

0.05 9.090 simplest picture of a single particle Hamiltonian with two

orbital states |L⟩ and |R⟩ for left and right dot occupancy,
0.00 two spin states |↑⟩ and |↓⟩, and the two valley states |K⟩
9.082
and |K ′ ⟩ inherited from the carbon nanotube band struc-
9.0967 9.0987
fd (GHz)
9.1007 200 100 0
/ h (GHz)
100 200
ture, the double quantum dot spectrum is made up of
8 levels from which two gapped 4-levels quasi-harmonic
FIG. 2. Spectroscopy and magnetic hysteresis of the ladder can be formed, as illustrated in figure 3e. We find
circuit. a. Continuous wave spectroscopy of the interdot indeed that with experimentally reasonable parameters,
transition studied in the paper. The measured signal is the we can account for both the small dispersion with detun-
phase of the microwave signal at fc as a function of the left ing ϵδ of the transition, as shown by the white dotted
and right plunger gate voltages. The arrow indicates the ϵδ line in figure 2d, and reproduce very well the peculiar
axis presented in panel d. The red dot displays the work- Rabi chevrons, as shown in figure 3d (Supplementary In-
ing point ϵδ = 0 at which the continuous spectroscopy and formation) . The corresponding spectrum is shown in
the time domain measurements are performed. b. Transport
figures 3e and 3f. It is calculated with the angle θ be-
measurements as a function of the magnetic field showing the
characteristic hysteretic magnetoresistance. The arrows indi-
tween the ferromagnetic contacts set to its lithographi-
cate which way the magnetic field was swept for each curve. cally designed value of π/6, intervalley couplings of the
The three pairs of curves correspond to three different plunger order of 4.5 GHz, exchange fields EL/RZ
/h on the left and
gates settings. c. Two-tone microwave spectroscopy as a func- the right dots between 9 and 10 GHz, a large tunnel cou-
tion of fd showing a resonance at fr = 9.0987 GHz. The pling tc /h ≈ 31 GHz and a bare charge-photon coupling
Lorentzian fit yields a linewidth of 589 kHz. The grey arrows of g0 /(2π) ≈ 27 MHz. The small intervalley couplings
indicate the full width at half maximum. d. Dispersion of indicate that our nanotube is very clean [16]. Having the
the resonance frequency with respect to the energy detuning intervalley couplings smaller than the exchange fields is
ϵδ . The white dashed line is the dispersion predicted by our
essential to obtain the small dispersion in detuning. With
model.
these parameters, all eigenstates of the spectrum are fully
hybridized between charge, spin and valley at zero detun-
ing. Despite this, the transitions between states |0⟩ and
This small dispersion indicates that low charge noise de- |1⟩ and between states |2⟩ and |3⟩ are valley transitions.
phasing can be achieved. Overall, these measurements It is confirmed by an almost flat dispersion of the transi-
validate that we have indeed implemented the scheme tion with external magnetic field applied perpendicularly
presented in figure 1a. to the nanotube axis both in experiment and the simula-
Going to the time domain is crucial to further under- tion (Supplementary Information) . All other transitions
stand the spectrum and the dynamics of our system. We would be spin-valley transitions similar to previously ma-
first perform Rabi oscillations. For this purpose, we ex- nipulated spin-valley qubits [17, 18]. In addition, this
cite the system with a frequency fd close to the resonance model and parametrization predicts the coupling of each
found in figure 2c and for a burst time tburst up to 2 µs, transition to the cavity mode. By applying a common
by steps of 20 ns. Right after, we read-out the result- drive amplitude coefficient, we can thus match the ob-
ing system state using a second pulse at fc (Supplemen- served Rabi frequencies Ω01 , Ω12 and Ω23 .
tary Information) . The corresponding measurement is Despite the multilevel dynamics, the observation of
shown in figure 3a. We observe fringes signaling quantum well-defined Rabi oscillations at a given drive frequency
oscillations. These fringes resemble conventional Rabi enables to measure accurately the relaxation time T1 and
chevrons observed in many qubit experiments and ex- decoherence time T2∗ of the system. The latter is achieved
tend up to almost 2 µs. However, there are two main by Ramsey interferometry, that is, performing two π/2-
differences here. First, the chevrons for fd > 9.0970 GHz pulses separated by a wait time τ and measuring the
oscillate slightly slower than those for fd < 9.0970 GHz. resulting state. In our case, a π/2-pulse is achieved in
 4

0.0 0.08 0.16

a (°) b c
3
0.15
9.107

Rabi frequency (MHz)

0.10 2
fd (GHz)

(°)
9.097
0.05 1

9.087
0.00
0
0.0 0.5 1.0 1.5 2.0 0.0 0.5 1.0 1.5 2.0 0.0 0.1 0.2 0.3 0.4
tburst (µs) tburst (µs) Ad (a. u)

0.0 0.08 0.16

d (°) e f
150
20
3
9.107 100

50 2
E / h (GHz)

E / h (GHz)
30
fd (GHz)

9.097 0 2tc
1
50 40

9.087 100 0
50
150
0.0 0.5 1.0 1.5 2.0 200 100 0 100 200 40 20 0 20 40
tburst (µs) / h (GHz) / h (GHz)

FIG. 3. Rabi oscillations. a. Rabi chevrons pattern of the carbon nanotube device. It is observed through the oscillations
of the phase contrast ∆φ as a function of the drive frequency fd and drive time tburst . b. Cut in panel a at fd = 9.0984 GHz
(white dashed line) showing Rabi oscillations as a function of time fitted by an exponentially decaying cosine. c. Dependence
2/3
of the Rabi frequency as a function of the drive amplitude Ad (filled circles). The orange line is a fit to a power law ΩR ∝ Ad .
d. Simulation of the Rabi chevrons pattern of panel a. e. Energy spectrum used to model the Rabi chevrons. f. Close-up on
the first four levels involved in the Rabi oscillations dynamics.

100 ns. This yields the phase contrast as a function of which hints at very slow dephasing noise.
the drive frequency fd and the wait time τ shown in fig- Given that we now have all the figures of merit of our
ure 4a. The observed fringes are reproduced in figure 4b system, we can make its decoherence “budget”. Among
using the same model and parameters as the one used for the processes which can contribute to decoherence, we
the Rabi chevrons. Moreover, a cut in these fringes at have radiative processes with the photon bath of the cav-
fd = 9.0984 GHz is fitted to a Gaussian decaying cosine ity (essentially the Purcell effect since the cavity ther-
which yields T2∗ = (1.27 ± 0.15) µs (Supplementary In- mal mean occupation is about 1 photon at 300 mK)
formation). This is the highest reported coherence time and the phonon bath of the suspended nanotube, charge
for a state in a quantum dot in a cavity [14]. We note noise, nuclear spin noise and cotunneling arising from
that despite the multilevel dynamics of our system, the virtual tunneling on and off of the dots levels to the
Ramsey fringes are very similar to the ones of a qubit. leads. The Purcell rate reads: ΓPurcell = κg 2 /(ωc − ωq )2 ,
This is further confirmed by looking at the Fourier trans- where κ = ωc /Q is the mode damping rate, ωc/q /(2π)
form (Supplementary Information) where we clearly see are the frequencies of the mode and the driven tran-
that the oscillations frequency depends linearly on the sitions respectively and g is the coupling strength be-
drive frequency detuning. In order to characterize further tween the transition and the mode. The model gives
the dephasing mechanisms in our system, it is instructive g/(2π) ≤ 2.5 MHz for all transitions involved, leading to
to perform a Hahn-echo measurement for which one in- ΓPurcell /(2π) ≤ 1.7 × 10−6 MHz for the cavity, which is 6
terleaves a π-pulse in between the two π/2-pulses of the orders of magnitude below the rates which we measure.
Ramsey sequence. This is shown in figure 4c. It allows us Such a small value is expected given the large frequency
to extend the coherence time up to T2E = (2.02±0.21) µs, detuning between the cavity and the transitions. The
which can be considered as our T2 time. Finally, we get effect of charge noise can be inferred from the slope of
a complete picture of the decoherence by performing a the transition with respect to detuning which is of the
relaxation measurement using a π-pulse and waiting a order ofp 4 × 10−3 MHz /µeV. Taking detuning fluctua-
time τ to measure. It yields T1 = (1.12 ± 0.06) µs (see tions of ⟨σϵ2 ⟩ ≈ 10 µeV, typical for CNT devices in cav-
figure 4d). The T2 time is thus not far from the 2 T1 limit ities [11, 27, 28], we obtain Γϕ,charge /(2π) ≈ 4×10−2 MHz,
 5

0.0 0.1 0.2 0.0 0.08 0.16

a (°) b (°) was very well isolated from the leads. However, the fact
that we observed T1 ≈ T2∗ is consistent with cotunneling
9.107 9.107
being the dominant source of decoherence.
By studying the peculiar Rabi chevrons pattern and
fd (GHz)

fd (GHz)
9.097 9.097 the weak dispersion of the transitions’ frequencies, we
were able to determine the spectrum of our system. In
9.087 9.087 particular, we have found that the regime with a very
large tunnel coupling is favorable to strongly reduce the
0.0 0.5 1.0 1.5 2.0 0.0 0.5 1.0 1.5 2.0
charge noise through a very weak dependence of the
0.0 0.1 0.2
c (°) d transitions’ frequencies with ϵδ and tc . We have thus
0.10
demonstrated microsecond-lived quantum states in a
9.107 carbon-based circuit in cQED, an order of magnitude
larger than with equivalent Si-based circuits. The
fd (GHz)

0.05
(°)

9.097 coherence is currently limited by T1 . Extending T1 by

reducing the coupling to the leads could further boost
9.087
0.00
coherence in our system to reach the pure dephasing
time Tφ = (2.9 ± 0.8) µs that we estimate and which
0.0 0.5 1.0
(µs)
1.5 2.0 0 1 2 3
(µs)
4 5 is slightly larger than in natural Si spin qubits [32].
Furthermore, all states having a spin content, it is
FIG. 4. Decoherence characterization. a. Ramsey expected that istopically purified 12 C nanotubes would
fringes of the carbon nanotube device. It is observed through yield even larger coherence times. Finally, combining
the oscillations of the phase contrast ∆φ as a function of the the achieved regime with a high-impedance resonator
drive frequency fd and time delay τ between the two π/2- to enhance the bare charge-photon coupling should, in
pulses of duration τπ/2 = 100 ns. b. Simulation of the Ram- future work, enable high-fidelity single- and two-qubit
sey fringes pattern of panel a using the level structure dis- gates far from the Purcell limit.
played in figure 3f and using the same parameters as in the
figures 3d and 3e. c. Hahn-echo fringes. d. Relaxation mea-
Acknowledgements: We acknowledge fruitful dis-
surement yielding T1 = 1.1 µs.
cussions with Z. Leghtas, L. Bretheau and J. D. Pillet.
This work is supported by the ANR MITIQ and ANR
about one order of magnitude smaller than our obser- JCJC STOIC, the Emergence project MIGHTY of ville
vations. Hyperfine interaction-mediated spin dephas- de Paris and the BPI project QUARBONE.
ing is also expected to yield small contribution with Data availability: The data that support the find-
Γϕ,hf /(2π) ≈ 4 × 10−2 MHz. This value is slightly larger ings of this work are available from the corresponding
than previous estimates because they were based on for- authors upon reasonable request.
mulae widely used in the literature, e.g. in ref. [11, 18], Author contributions: T.K. and M.R.D. designed
which led to an underestimation of the contribution of and supervised the experiment. B.N. and B.H. performed
nuclear spins to spin dephasing in carbon nanotubes. the measurements. B.N., B.H., G.A. and J.B. fabri-
This issue has been addressed in ref. [29]. For relaxation cated the device with inputs from A.L. for the design.
with phonons, as the CNT is suspended, we considered T.C., W.L., M.E., K.F.O., D.S., J.A.S., M.M.D., T.K.
stretching modes which have a fundamental frequency and M.R.D. contributed to the development of the CNT
ν0 close to the transitions given by ν0 L = 12.3 GHz ·µm nano-assembly technique. L.J. and A.T. developed the
with L the length of the CNT in µm [30]. It gives ν0 acquisition protocols. B.N. B.H., Q.S., A.C., T.K. and
between 8.2 and 6.15 GHz for L between 1.5 and 2 µm M.R.D. analyzed the data and did the theory. B.H., Q.S
as designed. The Rabi chevrons and Ramsey fringes and M.R.D. did the numerical simulations. B.N., B.H.,
are well reproduced considering this relaxation process Q.S., L.J., A.T., J.C., M.M.D., T.K. and M.R.D. con-
with a bare electron-phonon coupling of 500 MHz and a tributed to the discussion, interpretation and presenta-
damping rate of 2.5 GHz for L = 1.5 µm (Supplemen- tion of the results. B.N. B.H., T.K. and M.R.D. wrote
tary Information) . It however leads to a too small an the manuscript with inputs from all the authors.
estimation of T1 ≈ 0.2 µs indicating that it is not the Corresponding authors: Correspondence should
main relaxation process. There remains cotunneling as be addressed to B. Hue (benjamin.hue@phys.ens.fr) and
a decoherence mechanism. Interestingly, it is known to M.R. Delbecq (matthieu.delbecq@phys.ens.fr).
contribute both to relaxation and decoherence in essen- Competing interests: Authors affiliated with C12
tially the same way and amplitude, scaling as Γ2lead /U Quantum Electronics have financial interest in the com-
with Γlead the coupling to the leads and U the on-site pany. T.K. and M.R.D. declare equity interest in C12
Coulomb interaction [31]. We lack transport measure- Quantum Electronics. The remaining authors declare no
ments to properly characterize cotunneling as our device competing interest.
 6

[1] D. Loss and D. P. DiVincenzo, Quantum computation otube quantum dots, Physical Review B 77, 235301
with quantum dots, Physical Review A 57, 120 (1998). (2008).
[2] R. Hanson, L. P. Kouwenhoven, J. R. Petta, S. Tarucha, [16] E. A. Laird, F. Kuemmeth, G. A. Steele, K. Grove-
and L. M. K. Vandersypen, Spins in few-electron quan- Rasmussen, J. Nygård, K. Flensberg, and L. P. Kouwen-
tum dots, Reviews of Modern Physics 79, 1217 (2007). hoven, Quantum transport in carbon nanotubes, Reviews
[3] M. Veldhorst, J. C. C. Hwang, C. H. Yang, A. W. Leen- of Modern Physics 87, 703 (2015).
stra, B. de Ronde, J. P. Dehollain, J. T. Muhonen, F. E. [17] E. A. Laird, F. Pei, and L. P. Kouwenhoven, A valley–
Hudson, K. M. Itoh, A. Morello, and A. S. Dzurak, spin qubit in a carbon nanotube, Nature Nanotechnology
An addressable quantum dot qubit with fault-tolerant 8, 565 (2013).
control-fidelity, Nature Nanotechnology 9, 981 (2014). [18] T. Pei, A. Pályi, M. Mergenthaler, N. Ares,
[4] J. Yoneda, K. Takeda, T. Otsuka, T. Nakajima, M. R. A. Mavalankar, J. H. Warner, G. A. D. Briggs, and E. A.
Delbecq, G. Allison, T. Honda, T. Kodera, S. Oda, Laird, Hyperfine and Spin-Orbit Coupling Effects on De-
Y. Hoshi, N. Usami, K. M. Itoh, and S. Tarucha, A cay of Spin-Valley States in a Carbon Nanotube, Physical
quantum-dot spin qubit with coherence limited by charge Review Letters 118, 177701 (2017).
noise and fidelity higher than 99.9%, Nature Nanotech- [19] A. Cottet and T. Kontos, Spin Quantum Bit with Fer-
nology 13, 102 (2018). romagnetic Contacts for Circuit QED, Physical Review
[5] S. G. J. Philips, M. T. Mądzik, S. V. Amitonov, S. L. de Letters 105, 160502 (2010).
Snoo, M. Russ, N. Kalhor, C. Volk, W. I. L. Lawrie, [20] T. Cubaynes, L. C. Contamin, M. C. Dartiailh, M. M.
D. Brousse, L. Tryputen, B. P. Wuetz, A. Sammak, Desjardins, A. Cottet, M. R. Delbecq, and T. Kontos,
M. Veldhorst, G. Scappucci, and L. M. K. Vandersypen, Nanoassembly technique of carbon nanotubes for hy-
Universal control of a six-qubit quantum processor in sil- brid circuit-QED, Applied Physics Letters 117, 114001
icon, Nature 609, 919 (2022). (2020).
[6] G. Burkard, T. D. Ladd, A. Pan, J. M. Nichol, and J. R. [21] B. Undseth, O. Pietx-Casas, E. Raymenants,
Petta, Semiconductor spin qubits, Reviews of Modern M. Mehmandoost, M. T. Mądzik, S. G. J. Philips,
Physics 95, 025003 (2023). S. L. de Snoo, D. J. Michalak, S. V. Amitonov, L. Try-
[7] J. J. Viennot, M. C. Dartiailh, A. Cottet, and T. Kontos, puten, B. P. Wuetz, V. Fezzi, D. D. Esposti, A. Sammak,
Coherent coupling of a single spin to microwave cavity G. Scappucci, and L. M. K. Vandersypen, Hotter is
photons, Science 349, 408 (2015). Easier: Unexpected Temperature Dependence of Spin
[8] X. Mi, M. Benito, S. Putz, D. M. Zajac, J. M. Taylor, Qubit Frequencies, Physical Review X 13, 041015
G. Burkard, and J. R. Petta, A coherent spin–photon (2023).
interface in silicon, Nature 555, 599 (2018). [22] L. Petit, H. G. J. Eenink, M. Russ, W. I. L. Lawrie,
[9] N. Samkharadze, G. Zheng, N. Kalhor, D. Brousse, N. W. Hendrickx, S. G. J. Philips, J. S. Clarke, L. M. K.
A. Sammak, U. C. Mendes, A. Blais, G. Scappucci, and Vandersypen, and M. Veldhorst, Universal quantum logic
L. M. K. Vandersypen, Strong spin-photon coupling in in hot silicon qubits, Nature 580, 355 (2020).
silicon, Science 359, 1123 (2018). [23] C. H. Yang, R. C. C. Leon, J. C. C. Hwang, A. Saraiva,
[10] A. J. Landig, J. V. Koski, P. Scarlino, U. C. Mendes, T. Tanttu, W. Huang, J. Camirand Lemyre, K. W.
A. Blais, C. Reichl, W. Wegscheider, A. Wallraff, K. En- Chan, K. Y. Tan, F. E. Hudson, K. M. Itoh, A. Morello,
sslin, and T. Ihn, Coherent spin–photon coupling using a M. Pioro-Ladrière, A. Laucht, and A. S. Dzurak, Oper-
resonant exchange qubit, Nature 560, 179 (2018). ation of a silicon quantum processor unit cell above one
[11] T. Cubaynes, M. R. Delbecq, M. C. Dartiailh, R. As- kelvin, Nature 580, 350 (2020).
souly, M. M. Desjardins, L. C. Contamin, L. E. Bruhat, [24] S. Sahoo, T. Kontos, J. Furer, C. Hoffmann, M. Gräber,
Z. Leghtas, F. Mallet, A. Cottet, and T. Kontos, Highly A. Cottet, and C. Schönenberger, Electric field control of
coherent spin states in carbon nanotubes coupled to cav- spin transport, Nature Physics 1, 99 (2005).
ity photons, npj Quantum Information 5, 1 (2019). [25] A. Théry, B. Neukelmance, B. Hue, W. Legrand, L. Jar-
[12] P. Harvey-Collard, J. Dijkema, G. Zheng, A. Sammak, jat, J. Craquelin, M. Villiers, A. Cottet, M. R. Delbecq,
G. Scappucci, and L. M. K. Vandersypen, Coherent Spin- and T. Kontos, Observation of quantum oscillations in
Spin Coupling Mediated by Virtual Microwave Photons, the extreme weak anharmonic limit, Physical Review B
Physical Review X 12, 021026 (2022). 109, 064505 (2024).
[13] C. X. Yu, S. Zihlmann, J. C. Abadillo-Uriel, V. P. Michal, [26] J. Claudon, A. Zazunov, F. W. J. Hekking, and O. Buis-
N. Rambal, H. Niebojewski, T. Bedecarrats, M. Vinet, son, Rabi-like oscillations of an anharmonic oscillator:
É. Dumur, M. Filippone, B. Bertrand, S. De Franceschi, Classical versus quantum interpretation, Physical Review
Y.-M. Niquet, and R. Maurand, Strong coupling between B 78, 184503 (2008).
a photon and a hole spin in silicon, Nature Nanotechnol- [27] J. J. Viennot, M. R. Delbecq, M. C. Dartiailh, A. Cottet,
ogy 18, 741 (2023). and T. Kontos, Out-of-equilibrium charge dynamics in
[14] J. Dijkema, X. Xue, P. Harvey-Collard, M. Rimbach- a hybrid circuit quantum electrodynamics architecture,
Russ, S. L. de Snoo, G. Zheng, A. Sammak, G. Scappucci, Physical Review B 89, 165404 (2014).
and L. M. K. Vandersypen, Two-qubit logic between dis- [28] L. E. Bruhat, T. Cubaynes, J. J. Viennot, M. C. Darti-
tant spins in silicon (2023), arxiv:2310.16805 [cond-mat, ailh, M. M. Desjardins, A. Cottet, and T. Kontos, Cir-
physics:quant-ph]. cuit QED with a quantum-dot charge qubit dressed by
[15] D. V. Bulaev, B. Trauzettel, and D. Loss, Spin-orbit in- Cooper pairs, Physical Review B 98, 155313 (2018).
teraction and anomalous spin relaxation in carbon nan- [29] M. Delbecq, Spin Quantum Dynamics in Hybrid Circuits,
 7

Habilitation thesis, Sorbonne Université (2020).

[30] E. Mariani and F. von Oppen, Electron-vibron coupling
in suspended carbon nanotube quantum dots, Physical
Review B 80, 155411 (2009).
[31] U. Hartmann and F. Wilhelm, Decoherence of Charge
States in Double Quantum Dots Due to Cotunneling,
physica status solidi (b) 233, 385 (2002).
[32] K. Takeda, J. Kamioka, T. Otsuka, J. Yoneda, T. Naka-
jima, M. R. Delbecq, S. Amaha, G. Allison, T. Kodera,
S. Oda, and S. Tarucha, A fault-tolerant addressable spin
qubit in a natural silicon quantum dot, Science Advances
2, e1600694 (2016).
[33] J. J. Viennot, M. C. Dartiailh, A. Cottet, and T. Kontos,
Coherent coupling of a single spin to microwave cavity
photons, Science 349, 408 (2015).
[34] A. Cottet and T. Kontos, Spin Quantum Bit with Fer-
romagnetic Contacts for Circuit QED, Physical Review
Letters 105, 160502 (2010).
[35] A. Cottet, M. C. Dartiailh, M. M. Desjardins,
T. Cubaynes, L. C. Contamin, M. Delbecq, J. J. Viennot,
L. E. Bruhat, B. Douçot, and T. Kontos, Cavity QED
with hybrid nanocircuits: From atomic-like physics to
condensed matter phenomena, Journal of Physics: Con-
densed Matter 29, 433002 (2017).
 1

Supplementary materials for "Microsecond-lived quantum states in a

carbon-based circuit driven by cavity photons"

S1. MATERIALS AND METHODS

A. Wiring

Rubber joint 2 cm
CPW lines
PCB with Pi filters

IN OUT
R R
x24
C1 C2 C1

2 mm

DC lines

Supplementary Figure 1. Circuit chip embedded in the sampler holder. The PCB contains 24 DC lines and 4 CPW
lines for RF signals. Each DC line has a Pi filter: R = 1 kΩ, C1 = 1 nF and C2 = 2.2 nF. A rubber joint ensures the vacuum
tightness when closing the sample holder with its cover inside the stapler.

DC lines
RF input RF output
(x12)

300 K
CuBe

4K
Stainless
NbTi
Steel
20
1K

20
280 mK
π-filters

Sample holder

Thermalization clamp

20 20 dB attenuator

Supplementary Figure 2. 3 He cryostat wiring. The attenuators thermalize both the outer and inner part of the coaxial
cables and reduce thermal noise. The NbTi line provides very few attenuation to the RF output line.
 2

B. DC measurements

Supplementary Figure 3. DC measurements set-up. The voltage divider reduces the bias voltage VSD by a factor of 103 .
The current going out of the cryostat is amplified by a factor 107 by a low noise I-V converter.

C. RF measurements

1 Digitizer SP Device
ADQ14AC Digitizer
(sampling rate = 1GSPS) 1
A B CLK TRIG
2 Low noise voltage amplifier
FEMTO HVA-200M-40-B
(+20 dB) 2 2
+20 +20
3 Band pass Mini-circuits
155542 SIF-50+ (41-58 MHz)
3 3
4 I/Q mixer Marki ωIF
IQ4509LXP 1828
4 I ωIF
RF Q
5 Splitter Mini-Circuits
LO
ZX10-2-1252-S+
(1.8 - 12.5 GHz) Cavity
-3 drive
6 RF source
6
(19 dBm)
Agilent E8257D 13
5 ωLO
7 Circulator AEROTEK
H54-1FFF (6-8 GHz) 14
+30 7 ωIF
8 Arbitrary Waveform Generator
Tektronix 5014C

9 Single-sideband modulator 13 AWG 8

Polyphase Microwave LO
SSB4080A 9 IF
10 MHz
12 RF
ωIF reference
10 Directional coupler
+32
ωLO+ωIF
KRYTAR 180120 IF
10 RF LO
ωLO 6
11 Single-sideband modulator ωLO-ωIF
Polyphase Microwave
SSB80120A 11 Qubit
3
He cryostat drive
12 RF amplifier MITEQ (16 dBm)
AMF-4F-04001200-15-10P -40
(+32 dB / 15 V / 200 mA)

13 Band pass Mini-circuits

VHF-4600+ (5-11 GHz)

14 RF amplifier MITEQ
AFS3-00100800-45-10P-4
(+30 dB / 15 V / 200 mA)

Supplementary Figure 4. RF measurements set-up.

 3

D. Pulse sequences for the RF time domain measurements

a Rabi measurement b T1 measurement

tburst
Qubit drive Qubit drive
Cavity filling Cavity filling
Cavity readout tmeas Cavity readout
tdelay tpost-meas

c T2* measurement d T2E measurement

Qubit drive Qubit drive

Cavity filling Cavity filling
Cavity readout Cavity readout

Supplementary Figure 5. Pulse sequences.

For each RF time domain measurement presented in the paper, the cavity is measured during tmeas = 600 ns
after a delay time tdelay = 250 ns, such that in total the cavity is filled during tdelay + tmeas = 850 ns. We add a
post-measuring time tpost−meas = 3 µs to let both the cavity and the transitions relax in their ground state after the
measurement. In order to average the signal acquired, each pulse sequence is repeated Navg = 150 000 times. For the
Rabi measurements, tburst is varied from 0 µs to 2 µs by steps of 20 ns. For the T1 measurement τ is varied from 0 µs
to 5 µs by steps of 20 ns. For the T2∗ and T2E measurements τ is varied from 0 µs to 2 µs by steps of 20 ns. A π-pulse
is achieved in 200 ns, so logically a π/2-pulse is achieved in 100 ns.

S2. ADDITIONAL EXPERIMENTAL DATA

A. Cavity resonance

0.25

0.20
Amplitude (V)

0.15

0.10

0.05

0.00
6.965 6.970 6.975 6.980
f (GHz)
Supplementary Figure 6. Cavity resonance. The filled circles are the data and solid orange line is the fit.
 4

The cavity is fitted with | S21 | and S21 is defined as follows:

√ √
κin κout
S21 = + i T eiζ (S1)
i2π(fc − f ) + κ/2

where fc is resonance frequency of the cavity, κ is the linewidth of the cavity, κin and κout are respectively the
couplings to the input and output ports of the cavity, f is driving frequency, and T and ζ are Fano parameters.

Parameters Values
fc 6.975 GHz
κ 1.437 MHz
√
κin κout 174.311 kHz
T 0.00226699
ζ 317.369467

Supplementary Table I. Fit parameters for the cavity resonance

B. Ramsey oscillations

0.15

0.10
(°)

0.05

0.00

0.0 0.5 1.0 1.5 2.0

(µs)
Supplementary Figure 7. Ramsey oscillations

These Ramsey oscillations are obtained by driving at 9.0986 GHz. The blue line represents the data and the solid
yellow line is a fit with a Gaussian decaying cosine. It yields T2∗ = (1.27 ± 0.15) µs.
 5

C. Hahn-echo oscillations

0.10

(°)
0.05

0.00
0 1 2 3 4
(µs)
Supplementary Figure 8. Hahn-echo oscillations

These Hahn-echo oscillations are obtained by driving at 9.1010 GHz. The blue line represents the data and the
solid yellow line is a fit with a Gaussian decaying cosine. It yields T2E = (2.02 ± 0.21) µs.

D. Fourier transform of the Ramsey fringes

9.107
Amplitude
fd (GHz)

9.097

9.087

2 4 6 8 10
f (MHz)
Supplementary Figure 9. Fourier transform of the Ramsey fringes. The white dashed lined plots the following formula :
| f − f0 | with f0 = 9.0974 GHz. We do observe that the oscillations frequency depends linearly on the drive frequency detuning.
 6

E. Fourier transform of the Hahn-echo fringes

9.107

Amplitude
fd (GHz)

9.097

9.087

2 4 6 8 10 12 14 16 18
f (MHz)
Supplementary Figure 10. Fourier transform of the Hahn-echo fringes. The white dashed lined plots | f − f0 | and the
blue dashed lined plots 2× | f − f0 |, both with f0 = 9.0978 GHz.

F. Magnetic field dispersion

0.3
9.115
0.2
fd (GHz)

(°)

0.1
9.110
0
9.105 -0.1

10 8 6 4 2 0 2 4 6
B (mT)
Supplementary Figure 11. Magnetic field dispersion.
 7

G. Working points for magnetoresistance experiment

0.60

15
0.58
VPR (V)
10

I (pA)
0.56
5

0.54 0

0.79 0.77 0.75 0.73 0.71

VPL (V)
Supplementary Figure 12. Working points for magnetoresistance experiment. The red dot shows the working point for
the upper pair of curves in figure 1B, the purple dot shows the working point for the middle pair of curves in figure 1B, and
blue dot shows the working point for the lower pair of curves in figure 1B. The bias triangles are measured at VSD = 10 mV.

S3. SINGLE PARTICLE DQD HAMILTONIAN COUPLED TO A CAVITY MODE

The simulations presented in the main text are based on microscopic modelling of the CNT based double quantum
dot circuit coupled to microwave photons [33].

A. The DQD-cavity Hamiltonian

The Hamiltonian of the coupled system is

Ĥ = Ĥcav + Ĥint + ĤDQD . (S2)
The cavity Hamiltonian is
Ĥcav = ℏωc ↠â (S3)
with ωc /(2π) the cavity frequency and â (↠) the creation (annihilation) operator of a photon in the cavity mode.
We restrict the Hamiltonian ĤDQD of the CNT double quantum dot with non-collinear ferromagnetic contacts to
single particle occupancy. We consider the two valleys K and K ′ in the description. The exchange interaction in
the ferromagnetic leads induce effective Zeeman fields E L/R,K/K′ which can be different in both dots (due to different
⃗Z
contacts) and can differ from K to K . The double quantum dot Hamiltonian reads
′

h i
⃗Z ⃗Z ⃗
1 + τ̂z EL,K (1 + γ̂z ) + EL,K′ (1 − γ̂z ) + gL µB Bext ⃗
ĤDQD = − · σ̂
2 h 2 i
⃗Z ⃗Z ⃗
1 − τ̂z ER,K (1 + γ̂z ) + ER,K′ (1 − γ̂z ) + gR µB Bext ⃗
− · σ̂ (S4)
2 2
1
+ [∆L,KK′ (1 + γ̂z ) + ∆R,KK′ (1 − γ̂z )]
2
1  
⃗ ext · ⃗uz γ̂z τ̂0 σ̂0 ,
+ (tc τ̂x + ϵδ τ̂z ) γ̂0 σ̂0 − gorb µB B
2
 8

with E⃗Z uz + sin(θL/R )⃗ux , where ⃗ux , ⃗uz are units vectors along the x and z direc-
Z
 
L/R,K/K′ = EL/R,K/K′ cos(θL/R )⃗
tions, the z directions being chosen along the CNT axis and θL/R is the angle between the L/R contact magnetization
and the z direction. The parametrization of the angles is described in Fig. 13, giving θL/R = 12 (π − δθ) where δθ is
the angle between the zig-zags of the ferromagnetic contacts, fixed by design to δθ = π/6.
In ĤDQD , σ̂i , γ̂i , τ̂i with i ∈ {0, x, y, z} are the Pauli matrices operators acting in the spin space, the valley space
and the L/R space of the double quantum dot respectively, with index 0 giving the identity matrix in the corresponding
subspace. The Landé g-factors gL/R can in principle be different on both dots, although we fix them equal to 2 in
both dots in the following, µB is the Bohr magneton, B ⃗ ext is the external magnetic field, gorb the orbital g-factor.
We account for valley mixing ∆L/R,KK′ which can differ in both dots due to spatially varying disorder or curvature
of the suspended CNT. Finally the tunnel coupling between the two dots is tc and the energy detuning between the
two levels in each dot is ϵδ .

δθ=π/6 θ0
x Bext

θL
θR z

Supplementary Figure 13. Schematic representation of the device showing the the angle parametrizations. The ferromagnetic
contacts magnetizations are displayed as orange (left contact) and blue (right contact) arrows making an angle θL and θR with
the z-axis respectively. The angle between the contacts zig-zag is δθ = π/6 designed by EB lithography (see figure 1 of the
main text). The external magnetic field B⃗ ext has an angle θ0 with respect to the x-axis to account for possible misalignment.

The coupling Hamiltonian between the cavity and the DQD is

Ĥint = ℏg0 (â + ↠)τ̂z γ̂0 σ̂0 , (S5)

where g0 is the bare electron-photon coupling strength.

To reproduce both the dispersion with ϵδ and the time domain measurements, we used two sets of parameters that
are close to each other (see table II). This is motivated by the fact that over the measurement run the frequencies
slightly shifted. Then we adjust the spectrum to match the transitions frequencies at zero detuning ϵδ = 0 and their
dispersion in detuning ϵδ as well as the couplings of each transition to cavity photon to account for the contrast
of the Rabi chevrons. The couplings to each transition is gij = g0 ⟨i | τ̂z γ̂0 σ̂0 | j⟩ where |i⟩ and |j⟩ are eigenstates
corresponding to eigenvalues of levels i and j. With this set of parameters, the couplings matrix ⟨i | τ̂z γ̂0 σ̂0 | j⟩ at zero
detuning ϵδ = 0 is:

2.4×10−3 8.52×10−2 9.2×10−3 9.9632×10−1 1.9×10−3 6×10−5

 0 0

2.4×10−3 0 2.5×10−3 8.52×10−2 1.4×10−3 9.9636×10−1 0 6×10−5
 8.52×10−2 2.5×10−3 0 2.4×10−3
3×10 −5
0 9.9632×10 −1
1.9×10−3

 
 9.2×10−3 8.52×10−2 2.4×10−3 3×10−5 1.4×10−3 9.9632×10−1
0 0
(S6)

 9.9632×10−1 1.4×10−3 3×10−5 0 0 2.4×10−3 8.52×10−2 9.2×10−3

 
 1.9×10−3 9.9636×10−1 0 3×10−5 2.4×10−3 0 2.5×10−3 8.52×10−2 
6×10−5 0 9.9636×10−1 1.4×10−3 8.52×10−2 2.5×10−3 0 2.4×10−3
0 6×10−5 1.9×10−3 9.9632×10−1 9.2×10−3 8.52×10−2 2.4×10−3 0
 9

With this, we have the Rabi frequencies of the three transitions of the quasi-harmonic ladder given by Ω01 =
Ω23 = 2.65 MHz and Ω12 = 2.8 MHz, obtained by multiplying the corresponding coupling term by drive amplitude
term of value 40.7 common to all three transitions. The transition frequencies defined by fij = (Ej − Ei )/h are
f01 = 9098.0 MHz, f12 = 9096.1 MHz and f23 = 9096.0 MHz.

Fixed Parameters Adjusted Parameters Adjusted Parameters

for the dispersion with ϵδ for the time domain measurements
δθ π/6 ∆L,KK′ /h 4541.2 MHz ∆L,KK′ /h 4543.9 MHz
1
g µ
2 L B
28000 MHz T−1 ∆R,KK′ /h 4529.8 MHz ∆R,KK′ /h 4534.7 MHz
1
g µ
2 R B
28000 MHz T−1 Z
EL,K /h 9656.7 MHz Z
EL,K /h 9690.1 MHz
Z
EL,K′ /h 9189.9 MHz Z
EL,K′ /h 9222.1 MHz
Z
ER,K /h 9825,6 MHz Z
ER,K /h 9814.6 MHz
Z
ER,K ′ /h 9043,7 MHz Z
ER,K′ /h 9069.7 MHz
ωc /(2π) 6975 MHz g0 /(2π) 27.0 MHz g0 /(2π) 27.0 MHz
κ 1.44 MHz tc /h 41020 MHz tc /h 31190.0 MHz

Supplementary Table II. Parameters of the Hamiltonian in MHz used for numerical calculations.

B. Dephasing and relaxation

The dephasing is taken into account following the discussion in the supplementary material of ref. [34]. Dephasing
for each degree of freedom (charge left/right, spin up/down and valley K/K’) is calculated with perturbation theory.
For spin and valley, we restrict ourselves to first order while for charge we go to second order to deal with sweet spots
where the first order derivative cancels. We have
q
2
Γsφ,ij /(2π) = A2s |⟨j | τ̂0 γ̂0 σ̂z | j⟩ − ⟨i | τ̂0 γ̂z σ̂0 | i⟩| (S7)
q
2
Γvφ,ij /(2π) = A2v |⟨j | τ̂0 γ̂0 σ̂z | j⟩ − ⟨i | τ̂0 γ̂z σ̂0 | i⟩| (S8)
r hP i2
2 |⟨k | τ̂0 γ̂0 σ̂z | j⟩|2 |⟨k′ | τ̂0 γ̂0 σ̂z | i⟩|2
Γcφ,ij /(2π) (S9)
P
= A2c |⟨j | τ̂0 γ̂0 σ̂z | j⟩ − ⟨i | τ̂0 γ̂0 σ̂z | i⟩| + k Ek −Ej − k′ Ek′ −Ej A4c

For combining the dephasing rates associated with each derivative for charge noise we choose the ansatz of computing
the square root of the sum of the squares of each contributions. The total dephasing rate for each transition is then
the sum of the three contributions Γφ,ij = Γsφ,ij + Γvφ,ij + Γcφ,ij . The parameters Aα for α = {s, v, c} correspond
to the amplitude of fluctuations of the parameterspcontrolling the dephasing. For charge noise it corresponds to the
amplitude of fluctuations of the detuning Ac = ⟨ϵδ ⟩2 . For spin dephasing, As comprises contributions from the
nuclear spins environment and from the fluctuations of the exchange fields. For reproducing the experimental data,
we take Ac = 250 MHz, As < 10 MHz pand Av = 0.2 MHz. We note that the value of Ac corresponds to fluctuations of
the detuning of standard deviation ⟨ϵδ ⟩2 ≈ 1 µeV which is an indication of a very low charge noise, p as expected for
an ultra-clean carbon nanotube. In the main text however, we keep a more conservative value of ⟨ϵδ ⟩2 ≈ 10 µeV to
estimate the amplitude of charge noise because we lack independent experimental confirmation of the low detuning
fluctuations. The dephasing rates matrix restricted to the relevant subspace {|0⟩ , |1⟩ , |2⟩ , |3⟩} for the dynamics of
our system reads in MHz

0.029 0.158 0.129

 
0.029 0.137 0.157
Γφ /(2π) =  (S10)
0.158 0.137 0.030
0.129 0.157 0.030

For relaxation in the simulations, we consider phonons. At the transition frequency ∼ 9 GHz, the relevant modes
are stretching modes which have a fundamental mode frequency ν0 given by the relation ν0 L = 12.3 GHz · µm with
L the length of the CNT in µm [30]. Due to the nanoassembly technique, we do not know exactly the location of the
CNT along the gates. Depending on its location on the zig-zag contacts, the length of the CNT suspended over the
gate electrodes varies typically between 1.5 and 2 µm giving ν0 between 8.2 and 6.15 GHz. Relaxation is accounted for
by a radiative Purcell process. As we work at a temperature comparable to the transitions frequencies, the thermal
 10

ij is non vanishing, hence inverse Purcell effect which excite lower states to higher states are non zero
occupancy nth
and must be taken into account. We have
ν
√ !2
↓ gij / n
Γ1,ij /(2π) = κν /(2π) (nth
ij + 1) (S11)
νn − fij
ν
√ !2
↑ gij / n
Γ1,ij /(2π) = κν /(2π) nth
ij (S12)
νn − fij

with νn = nν0 the frequency of the n-th harmonic and nth ij = ehfij /kB T −1 . Here we take the scaling that the
1
√
charge-phonon coupling gij ν
scales as 1/ n [30]. The charge-phonon coupling matrix elements are the same as the
charge-photon matrix elements given in (S6). The bare charge-phonon coupling g0ν is however different than the bare
charge-photon coupling. g0ν is expected to be large and is a free parameter in the model. The phonon mode damping
rate κν is the third free parameter and as its scaling with the harmonic number is not clearly predicted, we kept it
constant. For the simulation we considered up to n = 5 harmonics so that all transitions of the quasi-harmonic ladder
have a contribution. Varying the length L of the CNT, we find that to reproduce the data, we have very similar
relaxation matrix with Γ↓1,02 and Γ↓1,13 dominating the relaxation process. Restricting again to the {|0⟩ , |1⟩ , |2⟩ , |3⟩}
subspace, the relaxation matrix for L = 1.5 µm, g0ν = 500 MHz and κν = 2.5 GHz is in MHz
6 × 10−3 8.4 × 10−1 3 × 10−3
 
−3 −3 −1
 1 × 10 7 × 10 8.4 × 10 
Γ1 /(2π) =  (S13)
4.6 × 10−2 2 × 10−3 6 × 10−3 
−5 −2 −3
4 × 10 4.6 × 10 1 × 10
The parameters used to compute the dephasing and relaxation rates for the simulation of the Rabi chevron and the
Ramsey fringes presented in the main text are summarized in table III

Dephasing Relaxation
Ac 250 MHz g0ν 500 MHz
As 10 MHz κν 2500 MHz
Av 0.2 MHz L 1.5 µm

Supplementary Table III. Parameters used to calculate the dephasing and relaxation rates.

C. Cavity readout

The transmission S21 of microwave field through the dqd-cavity system is given by [35]:
√
κ1 κ2
S21 = , (S14)
∆cd − i κ2 − Ξ
with ∆cd = ωc − ωd the detuning between the cavity and the drive angular frequencies, κ/(2π) the cavity linewidth.
The total susceptiblity Ξ is given by
2
X gij (nj − ni )
Ξ= Γij
, (S15)
ij ωij − ωd − i 2

with ni,j the population of level i, j, ωij = (Ej −Ei )/ℏ are transitions angular frequencies and Γij is the decoherence
Γ
rate associated with the transition between states |i⟩ and |j⟩ given by Γij = 1,ij 2 + Γφ,ij . The thermal equilibrium
populations are given by the Boltzmann weight
e−Ei /kB T
ni = P −E /k T , (S16)
ke
k B

with Ei the energy of the i-th level. In this work we have T = 300 mK. For simulations of the Rabi and Ramsey
experiments, the initial state is the thermal equilibrium state { 0.77, 0.18, 0.04, 0.01, 0, 0, 0, 0 }. The calculated popu-
lations at each evolution time are then used in equation (S14) to calculate the phase of the cavity field. To reproduce
 11

the phase contrast observed experimentally, we have to manually change a single parameter from the value computed
from the Hamiltonian. This parameter is g03 which must be increased by a factor approximately 40. We checked
that using this value to calculate the relaxation rates due to phonons reproduces the same simulation given that we
slightly reduce κν to 1.5 GHz. This highlights that parameters used to compute the relaxation rates with phonons
are not discriminant.

D. Simulated dispersions

9.0985
Frequency (GHz)

9.0975

9.0965
100 50 0 50 100
B (mT)
Supplementary Figure 14. Simulated dispersion of the 01 transition frequency with respect to the external mag-
netic field. It is computed with the parameters used to reproduce the time domain measurement.

9.0982
Frequency (GHz)

9.0979

9.0976
20 40 60 80 100
tc (GHz)
Supplementary Figure 15. Simulated dispersion of the 01 transition frequency with respect to the interdot tunnel
coupling. It is computed with the parameters used to reproduce the time domain measurement.