Professional Documents
Culture Documents
Such operational points are found from a detailed theoretical treatment and parameter study of the coupled
system. The quartonic readout circuit is also experimentally feasible and preserves the coherence properties of
the qubit. Our work reveals a new path for order-of-magnitude improvements of superconducting qubit readout
by engineering nonlinear light-matter couplings in parameter regimes unreachable by existing designs.
A. U (φ) B.
|0i 2χ |1i
JJs+Quarton Kb S7 S7
|0iqubit |1iqubit |0iqubit |1iqubit
JJs
Quarton ω
−2 −1 0 1 2 ωq ωr − χ ωd ωr + χ
φ/π
Frequency (GHz)
1→2 4→5 6→7
2→3 0.8 Q|0i Q|1i
16.4 Kb /2π
0.6 0.4
|0i
16.2 S7 0.4
0.30
0.20 0.6
240 0.3 0.13
0.15
0.4
210 0.10 0.15
0.2
0.05 0.2
180 0.18
0.00
40 50 60 70 80 90 40 50 60 70 80 90 40 50 60 70 80 90
EQ /2π (GHz) EQ /2π (GHz) EQ /2π (GHz)
FIG. 2. Parameter sweeps for optimal parameters for readout. (A) Linearizing the resonator by cancelling the JJ and QRM ϕ4 terms. (B)
Schematic eigenenergy spectrum and labeled quantities to be optimized. (C) Resonator transition frequencies for qubit states |0⟩, |1⟩ showing
good linearization at EQ /2π = 70 GHz. (D) Predicted QND fidelity from analytic readout behavior as a function of EQ . (E) Cross-Kerr
between the resonator and qubit (2χ) and qubit self-Kerr (Kb ) both increasing with EQ . (C-E) are respective horizontal line cuts (black
dashed)pin (F-H). (F-H) 2D sweeps of (F) frequency spread, (G) analytic QND fidelity, and (H) cross-Kerr, using constant resonator frequency
ωa = 8ECa EJaef f , constant qubit parameters, and varying EQ and EJaef f /ECa . The starred point has highly optimal parameters for
readout.
in Fig. 1A). See also Appendix B for more details on the ef- Note that we have made the common approximation [28]
fects of the normal metal resistor in the circuit, and Appendix to treat n series JJ arrays as a single element with poten-
I for its effects on the losses in the circuit. This would allow tial nEJ cos (ϕ/n), see Appendix C for justification via a
the entire system to be flux biased by a single current source. more detailed treatment without this approximation. We have
also included a small capacitive coupling term 8ECab n̂a n̂b to
model the Josephson junction capacitance in the quarton cou-
pler and other stray capacitance between the resonator and
B. Optimal parameters for readout qubit (assumed here to be ∼ 5 fF). In order to maintain net
zero linear coupling, we cancel the effect of the capacitive
Following Eq. (3), and adding the capacitive (kinetic) en- coupling term by engineering an opposite-signed linear in-
ergy terms associated with Cooper pair number n̂ operators ductive coupling in the quarton; this is done by perturbing the
but neglecting the normal metal resistor (see Appendix I for its value of α ≈ 1/2 or “tilting” [28] the quarton (see Methods).
perturbative treatment), we can derive the total system Hamil- We numerically solve (with ℏ = 1 units) for the eigenenergies
tonian: ωna nb of Eq. (4) in the Fock basis, without any rotating wave
approximations (RWA), and label the eigenstates |na , nb ⟩ by
Ĥ = 4ECa n̂2a + 4ECb n̂2b + 8ECab n̂a n̂b resonator (na ), qubit (nb ) excitation number (see Methods for
! details).
ϕ̂a
− 2EJa cos − EJb cos ϕ̂b
2 (4) For fast, high-fidelity and QND qubit readout, the sys-
! tem should simultaneously exhibit low resonator nonlinear-
ϕ̂a − ϕ̂b ity, strong qubit-resonator cross-Kerr, strong qubit self-Kerr,
− 2EJ cos + αEJ cos(ϕ̂a − ϕ̂b ).
2 and high predicted QNDness from analytics. We quantify res-
4
p
Res. frequency ω10 /2π 16.1 GHz approximately constant by fixing ωa ≈ 8ECa EJaef f . At
Res. effective EJ EJaef f /2π 269 GHz every point, we also solve for the optimal quarton tilt parame-
Res. mode EC ECa /2π 119 MHz ter 2α (which is heuristically optimized as described in Meth-
Res. capacitance Ca 163 fF ods). The sweep reveals a trade-off between resonator lin-
Qubit frequency ω01 /2π 7.78 GHz |q⟩
earization (minimizing S7 ) and maximizing 2χ. Addition-
Qubit EJ EJb /2π 20.3 GHz
Qubit mode EC ECb /2π 325 MHz ally, we see that increasing EJaef f of the resonator increases
|q⟩
Qubit capacitance Cb 59.6 fF the EQ that minimizes S7 as consistent with Fig. 2A. See
Quartic potential EQ /2π 70.0 GHz also Appendix D and E where we provide a more thorough
Quarton tilt 2α 1.02 analysis and summary of the Kerr effects in the system, in-
Quarton lone JJ αEJ /2π 95.2 GHz cluding the general case of more than 2 series JJs and exact
Quarton series JJ EJ /2π 186.7 GHz treatment of coupling terms such as (b† 2 + b2 )a† a (photon-
Quarton JJ capacitance CJ 5.2 fF |q⟩
|0⟩ enhanced squeezing). The discontinuities in S7 are the result
Res. spread for |0⟩q S7 /2π 21.9 MHz
|1⟩ of avoided crossings between the eigenstates (such as between
Res. spread for |1⟩q S7 /2π 15.7 MHz
|7, 1⟩, |5, 5⟩ seen at the right side of Fig. 2F). It is important
Qubit self-Kerr Kb /2π 475 MHz
to operate away from such crossings, as they indicate strong
Cross-Kerr 2χ/2π 252 MHz
hybridization of the resonator mode with the qubit mode, lead-
ing to qubit state leakage during readout and low QND fidelity
TABLE I. Summary of parameters for starred point in Fig. 2.
(see Fig. 2G). Irregular peaks in Fig. 2F can be seen to corre-
spond to lower QND fidelity regions in Fig 2G, where fre-
quency collisions involving lower energy (and thus more pop-
onator nonlinearity by defining the resonator frequency spread ulated in readout with n̄ = 2) resonator eigenstates tend to
|q⟩
Sn∗ := maxi,j≤n∗ |(ωiq −ω(i−1)q )−(ωjq −ω(j−1)q )|, where decrease the QND fidelity more.
∗
n is the number of resonator transitions we consider, and In Fig. 2C-E, we take a horizontal line cut in the 2D sweep
(ωiq − ω(i−1)q ) is the single photon resonator transition from (black dashed line in Fig. 2F-H) to examine the effect of EQ
eigenstates |i, q⟩ → |i − 1, q⟩. We would ideally like our lin- alone. Fig. 2C shows the resonator transition frequencies for
earized transmon resonator to behave as a perfect linear res- both qubit states and for various energy levels i → i + 1, a
|q⟩
onator, or S∞ = 0, but as shown in Fig. 2A, the linearized clear signature of linearization at EQ /2π = 70 GHz appears
transmon resonator lives in an effective potential that is a sum (grey dashed line), where resonator frequency spread for both
|0⟩ |1⟩
of the transmon JJs cosine function and the quarton coupler’s qubit states S7 , S7 is less than 25 MHz. Furthermore, Fig
first-order quartic function. This results in a quadratic func- 2D illustrates the variations in the analytically predicted QND
tion potential near the bottom of the potential well, with the fidelity for both qubit states, showing that both Q̄|0,1⟩ > 99%
corresponding low energy levels being linearly spaced, but an at EQ /2π = 70 GHz. In Fig. 2E, we see the expected [28]
increasingly less ideal quadratic potential and less linear en- trend of both the cross-Kerr and qubit self-Kerr Kb increas-
ergies for higher states. As such, we choose to linearize only ing with increasing EQ . At EQ /2π = 70 GHz, we find over
the first n∗ = 7 transitions and will drive the resonator to a 250 MHz of cross-Kerr and over 400 MHz of qubit self-Kerr.
low n̄ = 2 coherent state in subsequent readout simulations to |q⟩
Since κ ≈ 2χ, this means S7 ≪ κ and we can drive the res-
avoid exciting the higher, nonlinear resonator states. We also onator to a coherent state. In summary, the point marked with
compute the resonator-qubit cross-Kerr 2χ = ω11 −ω01 −ω10 a star in Fig. 2F-H is an example of a parameter set (see Ta-
and qubit self-Kerr Kb = ω02 −2ω01 . The frequency spectrum ble 1) that satisfies all the criteria for optimal quartonic read-
|q⟩
of the qubit and resonator along with the metrics S7 , 2χ, Kb out. We will proceed to use this parameter set for the sub-
are schematically illustrated in Fig. 2B. For predicting QND- sequent readout dynamics simulations. We also emphasize
ness of readout for qubit in state |k⟩, we find that a concise that the proposed quartonic readout scheme is versatile and
analytic estimate Q̄|k⟩ matches well with numerical readout many other suitable parameter sets exists for other resonator
simulations. This Q̄|k⟩ makes the realistic assumptions that and qubit frequencies.
the resonator quickly evolves into a steady state coherent state
[22] and that qubit decay into different qubit number eigen-
states during readout is the main source of reduced QND fi- C. Readout performance
delity (see Appendix A for details).
We now perform a parameter sweep with experimentally Using the optimized system parameters in Table I, we simu-
realistic parameters (see Table 1) to demonstrate how optimal late performance with η = 1, κ/2π = 300 MHz and a readout
parameters for readout can be found. We choose a transmon drive resulting in steady-state n̄ = 2 photons. Key perfor-
qubit with uncoupled frequency 7.5 GHz and a readout trans- mance results are summarized in Table II. Numerical full mas-
mon resonator with uncoupled frequency ωa /2π = 16 GHz. ter equation simulations show near-ideal state evolution over
We also use the previously defined EQ ≡ 3EJ /8 from the a readout pulse (Fig. 3A) duration of 5 ns with high average
E
quarton potential UQ (ϕ) ≈ 4!Q ϕ4 +O(ϕ6 ) [28]. In Fig. 2F-H, QND fidelity of 99.55% in Fig. 3 and high average readout
we sweep both EQ and the effective resonator Josephson en- fidelity 99.60% in Fig. 4. We emphasize that we performed
ergy EJaef f = EJa /2 while keeping the resonator frequency these simulations in the lab frame with the full JJ cosine po-
5
ε(t)/2π (GHz)
A 0.5
Qubit initial state Readout Time (ns) Readout fidelity QND-fidelity
|0⟩ 5 99.52% 99.13%
|1⟩ 5 99.69% 99.98%
0.0
100
B Qubit in |0i 0,0 TABLE II. Summary of simulated transmon qubit readout perfor-
0,4
mances.
Population
1,0
−1
10 1,4
2,0
3,0
10−2 4,0
5,0 numerical simulation closely matches the analytic expectation
6,0
7,0 of high Q̄ in Fig. 2, justifying the construction of the analytic
10−30 8,0
10 metric.
C Qubit in |1i 0,1
1,1 Simulated readout statistics are shown in Fig. 4. Here,
Population
2,1
−1
10 3,1 we numerically compute measurement trajectories using the
4,1
5,1
stochastic master equation [34] with heterodyne detection.
10−2 6,1
We include all the same dissipations in the previous Lindblad
7,1
8,1 master equation simulation, except we choose to monitor the
10−3 dissipation from the relevant resonator decays for our mea-
0 2 4 6 8 10
surement operator (see Methods). The 12,800 measurement
t (ns)
trajectories are demodulated to extract the I(t), Q(t) quadra-
tures, and following state-of-the-art experimental readout pro-
FIG. 3. Quartonic readout simulation showing near-ideal evo- tocols [11], we simulate the integrated I, Q quadrature signals
lution with high QNDness. (A) Square drive pulse for simula- (see Fig. 4A) by time-integrating the trajectories I(t), Q(t)
tion. (B-C) Eigenstate population for readout of qubit initially with weighting functions WQ (t) ∝ Q|1⟩ (t) − Q|0⟩ (t) and
in (B) qubit eigenstate |0⟩ and (C) |1⟩ with eigenstate convention
|resonator, qubit⟩. The state evolutions are highly QND with no sig-
WI (t) ∝ I|1⟩ (t) − I|0⟩ (t) . Each (I, Q) point in Fig. 4A
nificant leakage into other qubit states. is obtained by time-integrating a trajectory over the max read-
out pulse length of 5 ns, which after an optimal axis projec-
tion gives the histograms of signal in Fig. 4B. From Fig. 4B,
a readout fidelity of 99.52%, 99.69% (for qubit initialized in
tential and without applying rotating wave and dispersive ap-
|0⟩ , |1⟩, respectively) can be extracted from histogram over-
proximations, see Methods for more details.
laps [7].
As shown in Fig. 3A, we apply a resonator drive ε(t) =
In Fig. 4C, we simulate the readout performance as a func-
u(t)[2ε0 cos (ωd t)]u(5 − t), for Heaviside step function u(t),
tion of pulse length and plot the resulting readout fidelity and
which is a simple square pulse of width 5 ns followed by post-
signal-to-noise ratio (SNR). This is done by integrating the
readout resonator ring-down time (5 ns). The resulting system
measurement trajectories over times ranging from 1-5 ns. Re-
dynamics are numerically computed with the Lindblad mas-
sults here confirm that our proposed quartonic readout scheme
ter equation solver in QuTiP [33] with all dissipations treated
with a high κ, χ readout resonator indeed results in much
rigorously following Ref. [25] (see Methods for details). In
faster SNR and readout fidelity growth with measurement
Fig. 3B-C, we plot the time evolution of the eigenstate (for
time. We note that measurement times beyond 5 ns could im-
the undriven Hamiltonian) population for qubit initialized in
prove readout fidelity further, albeit at the expense of lower
eigenstate |0⟩ (Fig. 3B) and |1⟩ (Fig. 3C), respectively. As
QND fidelity below 99% for qubit initialized in |0⟩ since leak-
expected, the large κ of quarton coupled readout resonator al-
age into qubit state |4⟩ would increase to more than 1% (see
lows for very fast resonator response (≈ 1 ns) to the drive ε(t).
Fig. 3B).
We also observe near-ideal behavior, with the resonator reach-
ing an equilibrium coherent state from the drive while the
qubit state remains mostly unchanged. Final states at the end
III. DISCUSSION
of the readout and ring-down periods shows very little leakage
outside the starting qubit computational states, translating to
a high QND fidelity of 99.13% (99.98%) for qubit initialized A. Purcell Decay and Shot Noise Dephasing
in |0⟩ (|1⟩). Here, we use the standard definition of QND fi-
delity as being the probability of the qubit remaining in state Ideally, any readout scheme should not significantly worsen
|i⟩ after measuring |i⟩ [15]. The excellent QND fidelity of the qubit’s decoherence (T1 , T2 ). Here, we examine how com-
our proposed readout results from the non-perturbative cross- mon readout-induced decoherence channels such as Purcell
Kerr between the readout resonator and qubit, with the main decay [26] and thermal shot noise dephasing [35, 36] can be
parasitic coupling term being two-photon hopping of the form suppressed in the proposed quarton-coupler readout scheme.
â†2 b2 + h.c., which are highly suppressed from the massive Without the Purcell filter, the readout resonator is coupled
frequency detuning between the 7.73 GHz qubit and the 16.08 via its (normalized) charge operator n̂0 := i(↠− â) to the
GHz resonator. These two-photon hopping terms are still visi- readout environment (or bath) with frequency dependent rate
ble in Fig. 3B-C as they cause qubit leakage from starting state κ(ω), and the qubit is coupled via the tilted quarton to the
|i⟩ → |i + 2n⟩ for integer n. The high QND fidelity found in readout resonator, the eigenstates transitions of the system
6
3 1.0
B 10
Readout Fidelity
2.5
A C
|0iqb
Q (a.u.)
counts
SNR
0.0 0.5 2
101 |0iqb |1iqb
−2.5 |1iqb
0.0 0
0 5 −2.5 0.0 2.5 0 1 2 3 4 5
I (a.u.) signal (a.u.) t (ns)
FIG. 4. Quartonic readout statistics showing high readout fidelity for 5 ns square readout pulse.(A) Stochastic master equation simulated
time-integrated measurement results in the IQ plane; 12,800 points each represents a measurement trajectory, red (green) points are for qubit
|0⟩ (|1⟩). (B) Histogram of measurement signal from A. (C) Readout fidelity and SNR over time (duration of square pulse).
|ej ⟩ → |ei ⟩ are effectively coupled to the bath with rate Mechanism Decoherence rate Γ1 T1 = 1/Γ1
Γji = κ(ωji )| ⟨ej | n̂0 |ei ⟩ |2 [25]. So for qubit eigenstates
Purcell decay κ(ωq )| ⟨0, 0| n̂0 |0, 1⟩ |2 -
{|ej ⟩ , |ei ⟩} = {|1⟩ , |0⟩}, the relaxation rate is enhanced by
the Purcell decay rate ΓP ≡ Γ10 . By virtue of quarton cou- normal metal Eq. (6) 113 ms
2 q
plers being purely nonlinear couplers, it is possible to use quasiparticle ⟨0| sin ϕ̂
|1⟩ 8EJ
xqp ℏω 2∆
0.42 ms
2 πℏ
a gradiometric quarton [37] as the coupler and apply in-situ 2
ℏωq
h q i
2 ℏωq
flux-tuning to exactly cancel all linear coupling between the dielectric loss 4EC Qdiel
|⟨0|ϕ̂|1⟩| coth 2kB T + 1 72 µs
qubit and readout resonator [28, 38], thereby making ΓP = 0. flux noise | ⟨0| ∂∂Φ
Ĥ
|1⟩ |2 SΦ (ωq ) 3.9 ms
It is worth emphasizing that quarton-coupler’s zero Purcell
Dephasing rate Γ2 T2 = 1/Γ2
decay with large cross-Kerr can be achieved without Purcell
filters, in contrast to state-of-the-art dispersive readout where n̄t (n̄t +1)(2χ)2
thermal photon κ
0.51 ms
Purcell decay cannot be avoided without Purcell filters due to flux noise simulated 1.8 ms
the underlying linear coupling required for cross-Kerr.
In dispersive readout, large ΓP is typically suppressed by TABLE III. Summary of quarton coupled qubit decoherence proper-
adding Purcell filters [15, 32, 39–41] that minimize qubit ties
κ(ωq ) → 0 while keeping resonator κ(ωr ) unchanged. While
a Purcell filter is not required for quarton-coupler readout to
eliminate Purcell decay at κ(ωq ), it is nevertheless benefi-
cial for suppressing unwanted decay at other eigenstate tran-
sitions (see Appendix G for details). Purcell filtering is highly onator into a very low κ frequency band of the readout trans-
compatible with quarton-coupler readout since transition fre- mission line (e.g. protected by a filter) to reduce γm before
quencies to filter are usually many gigahertz away from the readout, and flux-tune the resonator back to the desired κ fre-
resonator frequency ωr we wish to preserve. State-of-the-art quency during readout. Such active flux-tuning schemes may
bandpass Purcell filters with quality factors Qf ≈ 10 can be also be replaced by a more hardware-efficient, passive scheme
used, which is sufficiently low [42] compared to the quality of shot noise protection in the form of a well-thermalized
factor of the readout resonator Qr = ωr /κ(ωr ) ≈ 50 for the [36, 46] high frequency readout resonator [12]. Unlike dis-
Purcell filter to operate ideally [43]. persive readout, quartonic readout performance does not ex-
Qubit dephasing from shot noise of thermal photons in the plicitly depend on qubit-resonator frequency detuning, so the
resonator is another important source of decoherence that in- quarton coupled readout resonator can be made very high fre-
creases with κ (for fixed χ/κ) [7, 44]. For an average thermal quency (e.g.≈ 16 GHz in Table 1) relative to current transmon
photon number n̄t , the shot noise dephasing rate is given by qubits and dispersive readout resonators (typically 3-8 GHz
[35] [8]). Since γm scales directly with the average thermal pho-
ton population following Bose-Einstein statistics: n̄t (T ) =
n̄t (n̄t + 1)(2χ)2 [exp (ℏωr /(kB T )) − 1]−1 , a marginally higher resonator fre-
γm = . (5) quency ωr = 2π × 12.5 GHz with state-of-the-art thermal-
κ(ωr )
ization (effective temperature T = 45 mK [46]) can have
We can substantially reduce γm in quartonic readout by 102 times lower n̄t and thus 102 times lower γm compared
leveraging both the frequency-tunability and the higher res- to a state-of-the-art dispersive readout resonator ωr = 2π × 8
onant frequency of quarton coupled readout resonators. The GHz, thereby nullifying the impact of 102 times larger κ. In
frequency-tunability of quarton coupled readout resonators fact, with even larger ωr /2π ≫ 12.5 GHz easily achievable
stems from the use of a lumped-element transmon-like mode for quartonic readout, the γm increase from larger κ may be
for readout, with inductance provided by JJs which can be re- more than offset by many orders of magnitude lower n̄t , mak-
placed with SQUIDs that can be in-situ flux-tuned [45]. This ing the quartonic readout’s γm net lower than state-of-the-art
allows for potential schemes that flux-tune the readout res- dispersive readout’s.
7
B. Relaxation from Normal Metal picted in Fig. 1. Although dissipation from resistors would
normally lead to large decoherence rates, the nonlinear nature
Because the quarton requires precise flux biasing with Φ̃ = of the quarton coupler suppresses the current that would pass
Φ0 /2 without flux biasing the larger loop through ground, through the normal metal. We derive (see Appendix I) the
we have incorporated a normal metal segment, modeled as transition rate ΓR,b for the qubit’s T1 relaxation to be
a resistor, to eliminate DC flux bias in the larger loop as de-
!
8e2 Rωb CJ CJ CJ2 2
ΓR,b = ⟨01| n̂a + n̂b − (n̂a − n̂b ) |00⟩
ℏ Ca + CCbJ+C
Cb
J
Cb + CCaJ+C
Ca
J
CΣ
!! (6)
8e2 R (EJ /ℏ)2 αEJ ϕ̂a − ϕ̂b 2
+ ⟨01| sin ϕ̂a − ϕ̂b − sin |00⟩
ℏ ωb EJ nS
where CJ is the quarton’s intrinsic capacitance and R is the D. Decoherence from Flux Noise
resistance of a normal metal segment. With normal metal
junctions of 10 µΩ resistance, we estimate that the qubit’s T1 Flux noise is a well-known decoherence channel with
lifetime due to resistive dissipation is 1/ΓR,b ≈ 0.11 s at the “quasi-universal” noise power spectrum [8]: SΦ (ω) =
operating point depicted in Fig. 2. We can calculate a sim- Hz γΦ
A2Φ 2π×1 γΦ ≈ 0.8 − 1.0 and A2Φ ≈
ω with
ilar dissipative loss for the resonator mode too, resulting in 2
1/ΓR,a ≈ 0.49 s. (1 µΦ0 ) /Hz. Qubit relaxation from flux noise follows [8]:
2
∂H
Γ1,Φ = ⟨0| |1⟩ SΦ (ωq ) (8)
∂Φ
loss does not depend explicitly on the unusually large param- should be achievable [58]. Depositing 300 nm of Pt to form a
eters such as EQ or κ in quartonic readout, so as expected, resistive segment 100 µm wide and 2 µm would then provide
T1,diel is essentially unaffected by the quarton. a normal metal segment with a resistance around 6.4 µΩ.
Microwave measurement hardware compatible with higher
frequency measurements have also been developed [59].
F. Future directions Since higher frequencies are typically less utilized, the ability
to operate readout resonators at much higher frequency due
The high fidelities simulated using only a simple square to the quarton coupler’s non-perturbative cross-Kerr not being
pulse demonstrate the robustness of our proposed scheme and explicitly frequency dependent may be an important practi-
leave ample room for further improvements. For instance, cal advantage compared to state-of-the-art dispersive readout
existing dispersive readout optimization techniques such as which suffers from frequency crowding [60].
pulse shaping [53, 54] or qubit shelving [55] should be readily For important use cases such as large-scale error correction
applicable to our quartonic readout scheme for some constant setups [1], it is necessary to have frequency-multiplexed read-
factor improvements in measurement time. Another avenue of out for hardware efficiency. This seems difficult at first glance,
improvement could be to leverage the inherent nonlinearity in as resonators typically must be spaced several line-widths
our transmon-based readout resonator for bifurcation [12] to apart to avoid cross-talk [42, 60] which for large κ/2π = 300
enhance readout performance. More comprehensive parame- MHz can strain the few gigahertz bandwidth of analog elec-
ter sweep and optimization could also show operation points tronics [61]. However, it has been shown in [42] that the nec-
|q⟩
with small Sn∗ for larger n∗ , which allows for more photons essary frequency spacing between each resonator can be sig-
n̄ to be used in readout for even higher SNR. However, higher nificantly reduced with individual Purcell filters, so this may
n̄ is also known to cause deleterious effects such as reduced provide a path to engineering multiplexed ultrafast quartonic
qubit T1 [56] which are not included in our model; it is worth readout of many qubits in the future. Alternatively, by sacri-
investigating if this and other large n̄ effects (known from dis- ficing some readout speed, κ/2π = 100 − 200 MHz designs
persive readout) like chaos-induced quantum demolition [21] can easily accommodate 10 multiplexed qubit readout chan-
are relevant for large n̄ quartonic readout. nels with existing hardware [61], while still operating in a
We also emphasize that while our results here use the parameter regime that is difficult or impossible to reach for
much higher available SNR of quartonic readout to reduce dispersive readouts with 2χ ≈ κ ⪅ 2π × 10 MHz. Frequency
measurement time, it is conceivable that important use cases multiplexing would also be much easier if one opted to use the
such as large-scale error correction setups [1] could prior- extra SNR from quartonic readout towards lowering η rather
itize hardware-efficiency over fast feedback time. It may than lowering measurement time, as this makes measurement
therefore be advantageous to instead use the higher avail- pulses much longer in time and thus narrower in frequency,
able SNR on tolerating lower measurement quantum effi- thereby reducing cross-talk.
ciency η in the hardware setup, thereby removing the need
for quantum-limited amplifiers [17–19] and their accompa-
nying impedance-matching circulators or isolators [18, 19].
Since these quantum-limited amplifiers typically improve η H. Takeaways
by about 10 times [8], and quartonic readout improves readout
SNR per unit time by more than 10 times, we envision state- In summary, we present an experimentally-feasible, quar-
of-the-art 50 ns or less measurement time without quantum- ton coupler-based scheme for ultrafast superconducting qubit
limited amplifiers [55] should be feasible, representing a dras- readout. Simulation results show that only 5 ns is needed
tic reduction in measurement hardware complexity. However, to reach readout and QND fidelity above 99%, representing
this would require further optimization for an operating point significant potential improvements to state-of-the-art experi-
with much lower qubit leakage rate such that quartonic read- ments that require 40 ns [12] and could thus lead to much
out with a much longer (order 50 ns) duration does not result faster feedback schemes such as quantum error correction
in significantly lower QND fidelity. protocols [1, 2]. In addition to the immediate benefits in
readout speed, the ability of the quartonic readout scheme
to operate with low readout photon number (n̄ = 2) and
G. Feasiblity and scalability huge qubit-resonator detuning could alleviate some practi-
cal readout-related issues such as chaos [21], thermal pho-
All proposed superconducting and normal metal circuit ton dephasing [44], and frequency crowding [60]. Unlike
necessary for quartonic readout should be compatible with many existing directions of superconducting qubit readout
standard microfabrication. In particular, the normal metal al- improvements that focus on optimizing fundamentally con-
lows for a simple flux bias scheme wherein all quarton cou- strained parameters like the quantum efficiency η ≤ 1 [17–
plers on a chip are designed with identical geometric loops 19] and dispersive cross-Kerr and environmental coupling
that can be simultaneously flux biased by a single coil pro- 2χ ≈ κ ⪅ 10 MHz [11, 12, 15], our work suggests that better
viding an uniform flux [57]. Our assumed 10 µΩ resistance non-perturbative nonlinear couplers can overcome traditional
should be achievable by using thin strips of platinum. At cryo- design constraints to reach new regimes of much larger χ, κ
genic temperatures, a platinum resistivity of 9.6 × 10−9 Ω cm and thus significantly improved readout speed.
9
A. Fock Basis Treatments To minimize Purcell decay and unwanted mixing of the
qubit and resonator modes, a main function of the quarton
With a non-perturbative coupler between qubit and res- coupler is to have strong nonlinear (cross-Kerr) coupling with-
onator modes, the circuit’s eigenstates may differ significantly out linear coupling. However, even a purely quartic (ϕ̂a − ϕ̂b )4
from a naive set of basis states. We represent the circuit eigen- coupling term, when expressed in the Fock basis with normal
states with the Fock basis, i.e. the Hilbert space is spanned by ordering, introduces some weak Jaynes-Cummings a† b + ab†
tensor product of Fock states in the resonator and qubit mode coupling terms, exactly analogous to a linear coupling term
subspace. Using the Fock basis is convenient because its ba- ϕ̂a ϕ̂b . This, along with unavoidable linear capacitive coupling
sis states can be chosen to be very close to the eigenstates. terms n̂a n̂b , motivates us to modify the quarton potential by
This is essential for utilizing heuristics such as minimizing lin- adjusting the tilt t := 2α to minimize the linear coupling [28].
ear coupling terms (a† b), but maintaining the cross-Kerr terms With a Fock basis close to the bare (Hj in Eq. (10)) eigen-
a† ab† b. states, we can tilt the quarton to minimize the quantity
The first step is writing the Hamiltonian of Eq. (4) in
some abstract Fock basis with the canonical transformations |(⟨0a | ⊗ ⟨1b |)H(|1a ⟩ ⊗ |0b ⟩|2
ϕ̂j → ϕzpfj (a†j + aj ), n̂j → inzpfj (a†j − aj ), then normal
ordering and separating out all the coupling terms to obtain a where H is the full circuit Hamiltonian and |nj ⟩ is the n-
Hamiltonian of the form photon Fock state in mode j. This can be done by sweeping
values of t, where we reoptimize the basis as described previ-
H = Ha + Hb + Hcoup (10) ously and calculate the linear coupling for each t.
where each Hj ∈ {Ha , Hb } has all terms that only contain the
operators aj , a†j , and each term in Hcoup contains operators C. Master equation simulation
from both modes. Normal ordering is important as this allows
the process to be analytic and the coupling term Hcoup to be
By choosing the Lindblad master equation formalism, we
as simplified as possible.
implicitly make the standard approximations [62] in its deriva-
To select an accurate basis to represent each mode j, we
1 tion (e.g. Markovian and Born). We discuss the validity of
need to optimize our numerical choices of ϕzpfj = 2nzpf .
j this in Appendix H. We follow [25] in constructing the Lind-
We do this by having the Fock basis states of mode j, blad dissipators in the master equation by assuming a zero-
{|0j ⟩, |1j ⟩, . . . |Nj ⟩}, to be as close as possible to the eigen- temperature bath and allowing only high to low energy eigen-
(j) (j)
states of Hj , {|e0 ⟩, . . . |eN ⟩}. Specifically, we aim to max- state transitions; it is also necessary to group transitions close
imize the quantity (≤ κ) in frequency to the same dissipator as these transitions
have correlated coupling to the same bath mode. In stochastic
Nj
X (j) master equation simulation, the same Lindblad dissipators are
|⟨kj |ek ⟩|2 used and the monitored stochastic operator is set to include
k=0 only the resonator transitions for the first few qubit states. For
where we choose Nj depending on the number of energy tran- computational efficiency, when simulating the quartonic read-
sitions we are interested in, which is usually up to 10. Since out system, we truncate the otherwise huge Fock-basis Hilbert
each bare Hamiltonian Hj includes terms with ϕzpfj from space and keep only the low photon number subspaces with
the other mode, the optimization of ϕzpfj for each basis isn’t a threshold determined iteratively via convergence of results.
completely independent. However, using the procedures and For stochastic master equation simulations, 12,800 trajecto-
heuristics explained in Appendix F, we are able to consistently ries are used with 200 simulation substeps for a measurement
achieve overlap probabilities of over 98% between the Fock generating time step of 1/(5ωd ). See Appendix G for a more
basis states and the eigenstates of Hj for the first 10 energy detailed presentation.
levels in each mode. This is usually sufficient for labeling the
relevant eigenenergies and our relevant heuristics.
We then label the eigenstates of H by iterating through each V. ACKNOWLEDGMENT
of the Fock basis states |na , nb ⟩ = |na ⟩ ⊗ |nb ⟩ and identifying
the eigenstate |λ⟩ (of H) with greatest overlap |⟨na , nb |λ⟩|2 . The authors thank Terry Orlando, Max Hays, Daniel Sank,
Then |λ⟩ is labeled as the eigenstate version of |na , nb ⟩. When David Toyli, Jens Koch, Aashish Clerk, Kyle Serniak, William
the terms in Hcoup are not causing near-degeneracies of Hj Oliver for fruitful discussions and insightful comments; the
eigenstates, the labeling choices are clear due to high (over authors would also like to thank David Rower for sample flux
98%) overlap of Hj eigenstates with the basis states. As de- noise simulation code.
scribed in the main text, degeneracies between resonator and This research was supported in part by the Army Research
qubit excitations can lead to higher leakage rates and lower Office under Award No. W911NF-23-1-0045 and the AWS
QND fidelity, so we aim to ensure high overlaps between Hj Center for Quantum Computing. Y.Y. acknowledges support
(bare) eigenstates and H (dressed) eigenstates. from the IBM PhD Fellowship and the NSERC Postgraduate
10
conditions to get:
q
ϕQ + ϕ2 − ϕ1 = 0 (B4)
This transforms the capacitive energy of the circuit via: qubit modes (that they are nonlinearly coupled with). There-
fore, we can safely ignore these internal modes, i.e. assume
4e2 ⊤ −1 cos (ϕa,b ) cos(ϕr,q ) ≈ cos (ϕa,b ), which leads us to the sim-
T = ⃗n C ⃗n
2 plified form of the potential energy used in main text.
4e2 ′⊤
= ⃗n W C −1 W ⊤⃗n′
2 (C8)
e2
Appendix D: Derivation of photon-enhanced squeezing
= 4⃗n′⊤ W C −1 W ⊤ ⃗n′
2
↔ One non-ideal quarton coupling effect is the addition of
= 4⃗n′⊤ EC ⃗n′ negative self-Kerrs and cross-Kerr to the qubit and resonator
due to photon number dependent (correlated) squeezing terms
↔
where the capacitive energy matrix EC is approximately diag- of the form (b† 2 + b2 )a† a and (a† 2 + a2 )b† b. These terms
onal but has small off-diagonal terms arising from finite junc- originate from the same Hamiltonian coupling term ϕ2a ϕ2b that
tion capacitance of JJs. For instance, with experimentally- gives the ideal cross-Kerr coupling, so they are unavoidable.
realistic estimates of capacitances: {CJ,q , CJ,r , Ca , Cb } = We can model the effects of these terms by looking at a toy
{3, 7.5, 80, 70} fF: model of two harmonic oscillators coupled only one of these
terms:
223.6 9.2 0 0
↔ 9.2 265.7 0 0 Htoy = ωa a† a + ωb b† b + ζ(b† 2 + b2 )a† a (D1)
EC = MHz (C9)
0 0 1291.3 0
0 0 0 3228.4 where we can assume ωζb ≪ 1 (generally true for quarton
coupling). We can perform a Schrieffer-Wolff transformation
with energy in units of h = 1. This makes intuitive sense, as on this Hamiltonian with the unitary
the junction capacitances in the quarton creates a direct (for
α JJ) and indirect (for series JJ) path for capacitive coupling 1 † ∗ 2
S = exp a a(z b − zb† 2 ) (D2)
(non-zero EC,12 ) between the two transmons; but the junc- 2
tion capacitances in the resonator does not contribute to any
coupling. and choose z = reiθ to be real with θ = 0 for simplicity. By
Putting everything together, the total circuit Hamiltonian setting the coefficients of the off-diagonal b2 terms in H̃ =
with transformed variables is: SHtoy S † to 0, we obtain the condition
H =U +T 2ζ †
tanh(2ra† a) = a a (D3)
ϕa ωb
= −2EJa cos cos ϕ̃r − EJb cos ϕb
2
ϕa − ϕb and with ζ/ωb ≪ 1, this is satisfied with r = ζ/ωb . Then by
− EJ [−α cos (ϕa − ϕb ) + 2 cos ( ) cos ϕ̃q ] expanding our transformed Hamiltonian H̃ to second order in
2
ζ, we have
+ 4EC,11 ñ2a + 4EC,22 ñ2b + 8EC,12 ña ñb
+ 4EC,33 ñ2r + 4EC,44 ñ2q 2ζ 2
H̃ = ωa − cosh(2ra† a) a† a + ωb b† b cosh(2ra† a)
(C10) ωb
It is clear from Eq. C10 that the quarton and res- 4ζ 2 † †
onator internal modes are transmon-like, with cosine po- + ωb sinh2 (ra† a) − a ab b cosh(2ra† a)
ωb
tential −2EJ(a) cos ϕq(r) , capacitive energy 4EC,44(33) ñ2q(r) ,
2ζ 2 † 2 2 4ζ 2 † 2 2 †
and no capacitive coupling. Their only coupling to the − a a cosh(2ra† a) − a a b b cosh(2ra† a)
resonator, qubit modes a, b is through nonlinear terms ωb ωb
cos (ϕa,b ) cos(ϕr,q ) which to lowest-order provides cross- (D4)
Kerr like coupling ϕ2a,b ϕ2r,q . However, these internal modes and expanding to second order in r and taking only the 4 wave
are extremely high frequency (high EC from low junction ca- mixing terms or lower, we have the transformed Hamiltonian
pacitance and high EJ from large individual JJs used in the
ζ2 2ζ 2 † † ζ2
chain), e.g. > 35 GHz, so they can be safely taken to be H̃ ≈ ωa − a† a + ωb b† b − a ab b − a† a† aa
frozen in the ground state where their cross-Kerr interaction ωb ωb ωb
with the resonator and qubit modes a, b can be ignored. Fur- (D5)
2
2ζ 2
thermore, we can choose to fabricate the junctions in the array which adds negative cross-Kerr 2ζ ωb and negative self-Kerr ωb
to have reasonable capacitances (e.g. about 5 fF such as in to mode a. This result holds for coupling via the other corre-
Eq. (C9)) such that these internal modes have EJ /EC > 50 lated squeezing term (a† 2 + a2 )b† b also, which analogously
2
2ζ 2
so they can really be viewed as extremely high frequency adds negative cross-Kerr 2ζ ωa and negative self-Kerr ωa to
transmons. This means the internal modes have steady fre- mode b. Note that the factor of 2 in the self-Kerr originates
quencies and do not impart charge noise to the resonator and from the self-Kerr K being K † †
2 a a aa.
13
the qubit and cross-Kerr between the qubit and the resonator.
TABLE V. Summary of cross-Kerr sources and analytic scalings.
Choices of the number of series junctions {nS , nJa , nJb } gen-
Mode a - b (resonator-qubit) erally have a significant influence on nonlinearity and there-
(EC,a EC,b )1/2 fore decisions should be made prudently to help achieve the
cross-Kerr from quarton coupler +χab = 2EQ
(EJ,a EJ,b )1/2 goals in optimization. Generally speaking, larger nS simply
from a’s squeezing
cross-Kerr χ2
− 2ωabb increases EQ relative to αEJ [28], whereas larger nJa , nJb
a2 b† b + h.c. drastically decreases the intrinsic negative self-Kerr in mode
from b’s squeezing
cross-Kerr χ2
− 2ωaba a, b. This is most relevant for the resonator mode a, which we
b2 a† a + h.c.
must linearize (achieve net zero self-Kerr). To that end, nJa
is an exceptionally valuable tuning knob, and we have opted
for nJa = 2 in the main text for the particular combination of
In typical quarton coupling circuits, the correlated squeez- resonator and qubit frequency we were working with. Other
ing term has magnitude ζ directly proportional to the ideal parameter ranges could certainly benefit from different nJa .
cross-Kerr χ:
since they both originate from the ϕ2a ϕ2b coupling Hamilto- A good basis is essential for effectively simulating cou-
nian. pled quantum systems. As described in Methods, we split our
Hamiltonian into three terms to isolate the coupling terms:
Appendix E: Summary of Kerr effects in quartonic readout H = Ha + Hb + Hcoup
circuit
and in order to accurately and efficiently represent the circuit
Kerr nonlinearity relative to linear inductance is generally eigenstates and their associated eigenenergies, we want to find
weakened by chaining together more JJs in series [28]. If we a Fock basis that well represents the eigenstates of Hj for each
generalize the Hamiltonian in the main text to allow for any mode j = a, b. We aim to maximize
number of chained JJs in each mode, we can get a Hamilto-
Nj
nian: X
|⟨kj |ek,j ⟩|2
Ĥ = 4ECa n̂2a + 4ECb n̂2b + 8ECab n̂a n̂b k=0
! !
ϕ̂a ϕ̂b where |ek,j ⟩ is the k th eigenstate of Hj . This translates to
− nJa EJa cos − nJb EJb cos
nJa nJb the general problem with finding a Fock basis for a poten-
! tial U (ϕ) that isn’t necessarily quadratic. One approach is
ϕ̂a − ϕ̂b to sweep possible values of ϕzpf = 1/(2nzpf ) and search
+ αEJ cos(ϕ̂a − ϕ̂b + ϕ̃) − nS EJ cos
nS for the maximum overlap as seen in Fig. 7. There, we also
(E1) demonstrate the performance of two analytical heuristics, first
Tables IV-V below summarize the various sources of self- by minimizing the magnitude of the a† a† term coefficient, and
and cross-Kerr effects in this generalized quartonic readout by minimizing the coefficient of a† a terms. We may choose to
setup, along with their expected analytic scalings. Two main implement analytical heuristics for computational efficiency.
causes of self- and cross-Kerr (other than inherent JJ self- As seen in Fig 7, minimizing the normal ordered a† a coef-
Kerr) are the bare mode quarton Kerr effects derived in [28] ficients tend to give high average overlaps between our basis
and the correlated squeezing Kerr effects as derived above. A and Hj eigenstates. Since the Fock basis is a complete basis,
main goal in parameter optimization is to have the self-Kerr the final eigenenergies of H should be independent of the ex-
(and higher level nonlinearity) in the resonator be net zero act choices of the ϕzpf values, so the exact optimality of our
while maintaining high (hundreds of megahertz) self-Kerr in ϕzpf values is unimportant.
14
10 15 20 25 10 15 20 25
Drive
Frequency
1 R U P D O L ] H G
ergy eigenstates |ej ⟩ → |ei ⟩. Furthermore, we define indepen- states (blue lines in Fig. 8). By construction, the stochas-
dent baths indexed by k, each coupled to the k’th set of eigen- tic master equation will produce the same average dynam-
state transitions {ωk } that have overlapping line width [25]. ics as the deterministic master equation, while providing re-
Eigenstate j → i transitions with frequencies ωji = ωj − ωi alistic measurement trajectories. Since the starting state of
are considered to have overlapping line widths if they satisfy our readout simulation is always a pure state (eigenstate of
|ωji − ωj ′ i′ | ≤ c ∗ κ (for some order unity constant c). In qubit 0,1), we use QuTiP’s stochastic Schrodinger equation
summary: solver ssesolve [33]. Iteratively choosing QuTiP stochastic
solver parameters [33] for convergence, we find that about
|ei ⟩ ⟨ej | ∈ dˆk iff |ωji − ωj ′ i′ | ≤ c κ for ωj ′ i′ ∈ {ωk } (G3) ntraj ≈12,800 trajectories, nsubsteps ≈200 substeps for a
given time step of 1/(5ωd ) worked well. The resulting nu-
This is physically important as eigenstate transitions within merical heterodyne measurement trajectories are demodulated
about κ in frequency are correlated in their coupling to the by multiplying a phase exp (iωd t) and integrated to generate
same bath mode [25], and independent of bath modes cou- Fig. 4 of main text, from which SNR and readout fidelity can
pling transitions ≫ κ away. In practice, the labelling in be obtained.
Eq. (G3) above is done via a density-based clustering algo-
rithm (e.g. DBSCAN) on an array of all the allowed eigen-
state transitions {ωji }, to identify each of the k baths. The
results are shown in Fig. 8, where a threshold rate of 10 kHz Appendix H: Approximations in the master equation
was set to discard dissipations too slow to affect our O(10)
ns time simulations. Fig. 8’s column 2 panels repeat plots in Master equations are derived with [62] many approxima-
column 1 but with opacity of lines set in proportion to the tions such as the Markovian or Born approximations. Our
transition’s κeff,ij . This shows that the rates are dominated by quartonic readout designs use very large κ/2π = 300 MHz,
the eigenstate transitions that most resembles resonator single so it is worth checking if these approximations are still valid.
photon loss, which are the transitions we monitor in readout In the literature, Purcell filters routinely have quality fac-
measurement. We index this bath by k ∗ and will use it as the tors of about 30 and many hundreds of MHz of decay rate
monitored operator in the subsequent stochastic master equa- κf (e.g. [12] κf /2π = 310 MHz and [32] κf /2π = 224
tion simulation. MHz), and the standard Lindblad master equation was suc-
Note that in Eq. (G2), we use a realistic effective decay cessful in reproducing experimental results [43]. Therefore,
rate κeff,ij that is weighted by the coupling frequency de- the standard approximations used to derive the master equa-
2
pendence (κ ∝ ωji )[7] and the Purcell filtering response tion such as the Markovian and Born approximations should
(κ ∝ [1 + (2(ωji − ωr )/κf )2 ]−1 ) [43]. The effect of Pur- still be valid in for our design with similar κ/2π = 300 MHz
cell filter’s suppression of unwanted bath coupling far from and Q = ωa /κ ≈ 50.
the drive frequency is clearly visible in column 2 of Fig. 8, in However, our Purcell filter has much larger κf = 4κ =
practice the use of the Purcell filter improves QND and read- 2π × 1.2 GHz decay rate, corresponding to a low quality fac-
out fidelity by a few percent, making it necessary for reaching tor of Q ≈ 13. Although this is the same order of magnitude
> 99% fidelity. We also chose a nominal value of c = 1 as experimentally demonstrations [12, 32], it is an open ques-
in Eq. (G3), but the results of the Linblad-form master equa- tion whether the coupling of this Purcell filter to the bath is
tion simulation seem robust to a range of c values we tested strong enough to violate the master equation’s underlying as-
(around 0.5 - 2). sumptions. Therefore, we have opted to not include the Pur-
Having constructed our Lindblad dissipators, we can use cell filter as part of the quantum system in the master equa-
the Lindblad-form master equation of Eq. (G1) to find the av- tion simulations, but have instead treated it more classically
erage dynamics of the resonator-qubit during readout, produc- as a filtering function on the resonator’s decay to the bath
ing plots like Fig. 3 in the main text. This is sufficient for (κ ∝ [1 + (2(ωji − ωr )/κf )2 ]−1 ). Note also that we chose
determining properties such as QND fidelity. However, to rig- a small ratio of κf /κ = 4 but this has been experimentally
orously obtain readout fidelity, we instead simulate measure- demonstrated [42].
ment trajectories using the stochastic master equation [33]:
FIG. 9. Rearrangement of the readout coupling circuit and illustration of the Caldeira-Legget model of the resistor to follow dissipative
calculations in [65].
with the Caldeira-Legget model as where L0 is the original Lagrangian. We also assume that
∆ϕR is small enough that we don’t need to include it in our
X 1 1 2
Legendre transformation. This gives us the modified Hamil-
2
HR = Cβ ϕ̇β + ϕ (I1)
2 2Lβ β tonian
β
P
with a total phase difference ∆ϕR = β ϕβ . Here, we as-
sume that the resistance is perturbative and much smaller than
P
the reactance of the rest of the system, so ϕR = β ϕβ ≪ H = H0 + HR + αEJ sin(ϕ̂a − ϕ̂b )∆ϕR
1 and that the quarton junction capacitances CJ are much
!
ϕ̂a − ϕ̂b
smaller than the resonator and qubit capacitances. Our new − EJ sin ∆ϕR
nS
Lagrangian (allowing for general {nS , nJa } as in Eq. (E1))
will be
q̂a q̂b
(I4)
− CJ +
1 1 1 CΣ /(Cb + CJ ) CΣ /CJ
L= Ca ϕ̇2a + Cb ϕ̇2b + CJ (ϕ̇a − ϕ̇b − ∆ϕ̇R )2
q̂a q̂b
2 2 2 − + ∆ϕ̇R
+ nJa EJa cos(ϕa /nJa ) + EJb cos(ϕb ) CΣ /CJ CΣ /(Ca + CJ )
ϕa − ϕb − ∆ϕR
+ αEJ cos(ϕa − ϕb − ∆ϕR ) − nS EJ cos
nS
+ LR where CΣ = Ca Cb +CJ Ca +CJ Cb and H0 is the Hamiltonian
(I2) without the resistor (see Eq. (E1)) .
which we can expand to first order in ∆ϕR , ∆ϕ̇R so that Treating the resistive terms with nonzero ϕR , ϕ̇R as per-
turbations, we can apply Fermi’s golden rule to obtain the
L = L0 + LR − CJ (ϕ̇a − ϕ̇b )∆ϕ̇R transition probability from state |i⟩ → |f ⟩ with ∆ϕR , ∆ϕ̇R
ϕa − ϕb ∆ϕR as uncorrelated noise terms. Using the transition formula
+ αEJ sin(ϕa − ϕb )∆ϕR − nS EJ sin ⟩|2
nS nS Γ = |⟨i|Â|f
ℏ2 SF F (ω) for operator  and noise source F from
(I3) Ref. [66], we obtain our transition rate of:
n̂a n̂b n̂a n̂b 2
Γ = ⟨i|2eCJ + − + |f ⟩ SV V (ωif )/ℏ2
CΣ /(Cb + CJ ) CΣ /CJ CΣ /CJ CΣ /(Ca + CJ )
! (I5)
ϕ̂a − ϕ̂b 2
+ ⟨i|αEJ sin(ϕ̂a − ϕ̂b ) − EJ sin |f ⟩ Sϕϕ (ωif /ℏ2 .
nS
Following the derivations in [65], we can find the spectral densities for ϕ̇R = VR , ϕR as
2
SVR VR (ω) = ℏωR ,
1 − e−ℏω/kB T
2 (I6)
ℏR 2 2π
Sϕϕ (ω) =
ω 1 − e−βℏω/kB T Φ0
17
2
and we take the limit of T → 0 in our cold superconducting environments so SV V (ω) = 2Rℏω, Sϕϕ (ω) = 2Rℏ ω
2π
Φ0 . This
gives our qubit dissipation rates as
8e2 Rωb C2
CJ CJ 2
Γ= ⟨01| n̂a + n̂b − J (n̂a − n̂b ) |00⟩
ℏ Ca + CJ Cb /(Cb + CJ ) Cb + CJ Ca /(Ca + CJ ) CΣ
!!
2 2
8e R (EJ /ℏ) αEJ ϕ̂a − ϕ̂b 2 (I7)
+ ⟨01| sin ϕ̂a − ϕ̂b − sin |00⟩
ℏ ωb EJ nS
= ΓC + ΓQ .
1 C −1
FIG. 10. Branch variables used in flux noise analysis (ϕa − P a −1 ϕ̃)
Ĥirr = T − 2EJa cos
2 i Ci
−1
C
−EJb cos ϕb + P b −1 ϕ̃
i Ci
The second term ΓQ in Eq. (I7) represents the resistive dis- −1 (J1)
sipation from energy flowing through the quarton. This term CQ
+EJα cos ϕb − ϕa − P −1 ϕ̃
demonstrates why tilting the quarton helps reduce the resistive i Ci
loss, where the tilt is defined as t = 2α. Heuristically, if we CQ−1
1
assume a small phase difference ϕa − ϕb with some variance −2EJs cos (ϕb − ϕa − P −1 ϕ̃ ,
2 i Ci
∆ϕ2 = ⟨(ϕ̂a − ϕ̂b )2 ⟩, expanding the operator in the second
term yields
where T is the usual capacitive energy term and CQ = Cs +
Cα . Similarly, if the time-dependent noise ϕ̃ threads the quar-
! ton loop, we obtain
αEJ ϕ̂a − ϕ̂b
sin ϕ̂a − ϕ̂b − sin
EJ nS 1
Ĥirr = T − 2EJa cos (ϕa + Aϕ̃) − EJb cos ϕb + B ϕ̃
2
∆ϕ2 ∆ϕ2
t 1
≈ 1− (ϕ̂a − ϕ̂b ) − 1− (ϕ̂a − ϕ̂b )
+EJα cos ϕb − ϕa + (B − A)ϕ̃
nS 6 nS 6n2S
1
∆ϕ2 n2S − 1 ϕ̂a − ϕ̂b −2EJs cos (ϕb − ϕa + (B − A + 1)ϕ̃ ,
= t− 1+ 2
2
6 nS nS (J2)
with
Cb Cs
which shows that the resistive is minimized for some t > 1. A= (J3)
2Ca Cb + (Ca + Cb )(Cα + Cs )
In our design, we don’t need to tilt the quarton to minimize
the resistive loss as long as the resistive dissipation isn’t the
limiting factor in our T1 values. Instead, we can aim to min- Ca Cs
B= (J4)
imize the linear coupling between the resonator and qubit to 2Ca Cb + (Ca + Cb )(Cα + Cs )
reduce Purcell decay. Using the parameters in Table I, Eq. (I7)
evaluates to 1/Γ ≈ 0.113 s. Here, 1/ΓC = 0.871 s, 1/ΓQ = We can then compute the T1 decoherence contribution of each
0.130 s. loop as
18
[1] S. Krinner, N. Lacroix, A. Remm, A. Di Paolo, E. Genois, ducting qubit using a nonlinear purcell filter, arXiv preprint
C. Leroux, C. Hellings, S. Lazar, F. Swiadek, J. Herrmann, arXiv:2309.04315 (2023).
et al., Realizing repeated quantum error correction in a [13] M. Reed, L. DiCarlo, B. Johnson, L. Sun, D. Schuster, L. Frun-
distance-three surface code, Nature 605, 669 (2022). zio, and R. Schoelkopf, High-fidelity readout in circuit quan-
[2] V. Sivak, A. Eickbusch, B. Royer, S. Singh, I. Tsioutsios, tum electrodynamics using the jaynes-cummings nonlinearity,
S. Ganjam, A. Miano, B. Brock, A. Ding, L. Frunzio, et al., Physical review letters 105, 173601 (2010).
Real-time quantum error correction beyond break-even, Nature [14] N. Didier, J. Bourassa, and A. Blais, Fast quantum nonde-
616, 50 (2023). molition readout by parametric modulation of longitudinal
[3] C. H. Bennett, G. Brassard, C. Crépeau, R. Jozsa, A. Peres, and qubit-oscillator interaction, Physical review letters 115, 203601
W. K. Wootters, Teleporting an unknown quantum state via dual (2015).
classical and einstein-podolsky-rosen channels, Phys. Rev. Lett. [15] Y. Sunada, S. Kono, J. Ilves, S. Tamate, T. Sugiyama,
70, 1895 (1993). Y. Tabuchi, and Y. Nakamura, Fast readout and reset of a super-
[4] L. Steffen, Y. Salathe, M. Oppliger, P. Kurpiers, M. Baur, conducting qubit coupled to a resonator with an intrinsic purcell
C. Lang, C. Eichler, G. Puebla-Hellmann, A. Fedorov, and filter, Phys. Rev. Appl. 17, 044016 (2022).
A. Wallraff, Deterministic quantum teleportation with feed- [16] C. C. Bultink, B. Tarasinski, N. Haandbæk, S. Poletto,
forward in a solid state system, Nature 500, 319 (2013). N. Haider, D. Michalak, A. Bruno, and L. DiCarlo, General
[5] J. Johnson, C. Macklin, D. Slichter, R. Vijay, E. Weingarten, method for extracting the quantum efficiency of dispersive qubit
J. Clarke, and I. Siddiqi, Heralded state preparation in a super- readout in circuit qed, Applied Physics Letters 112 (2018).
conducting qubit, Physical review letters 109, 050506 (2012). [17] N. Bergeal, F. Schackert, M. Metcalfe, R. Vijay,
[6] D. Ristè, J. G. van Leeuwen, H.-S. Ku, K. W. Lehnert, and V. Manucharyan, L. Frunzio, D. Prober, R. Schoelkopf,
L. DiCarlo, Initialization by measurement of a superconduct- S. Girvin, and M. Devoret, Phase-preserving amplification near
ing quantum bit circuit, Phys. Rev. Lett. 109, 050507 (2012). the quantum limit with a josephson ring modulator, Nature
[7] A. Blais, A. L. Grimsmo, S. M. Girvin, and A. Wallraff, Cir- 465, 64 (2010).
cuit quantum electrodynamics, Reviews of Modern Physics 93, [18] C. Macklin, K. O’brien, D. Hover, M. Schwartz, V. Bolkhovsky,
025005 (2021). X. Zhang, W. Oliver, and I. Siddiqi, A near–quantum-limited
[8] P. Krantz, M. Kjaergaard, F. Yan, T. P. Orlando, S. Gustavsson, josephson traveling-wave parametric amplifier, Science 350,
and W. D. Oliver, A quantum engineer’s guide to superconduct- 307 (2015).
ing qubits, Applied physics reviews 6 (2019). [19] K. Peng, M. Naghiloo, J. Wang, G. D. Cunningham, Y. Ye, and
[9] F. Arute, K. Arya, R. Babbush, D. Bacon, J. C. Bardin, K. P. O’Brien, Floquet-mode traveling-wave parametric ampli-
R. Barends, R. Biswas, S. Boixo, F. G. Brandao, D. A. Buell, fiers, PRX Quantum 3, 020306 (2022).
et al., Quantum supremacy using a programmable supercon- [20] D. Sank, Z. Chen, M. Khezri, J. Kelly, R. Barends, B. Camp-
ducting processor, Nature 574, 505 (2019). bell, Y. Chen, B. Chiaro, A. Dunsworth, A. Fowler, et al.,
[10] A. Wallraff, D. I. Schuster, A. Blais, L. Frunzio, J. Majer, M. H. Measurement-induced state transitions in a superconducting
Devoret, S. M. Girvin, and R. J. Schoelkopf, Approaching unit qubit: Beyond the rotating wave approximation, Physical re-
visibility for control of a superconducting qubit with dispersive view letters 117, 190503 (2016).
readout, Physical review letters 95, 060501 (2005). [21] J. Cohen, A. Petrescu, R. Shillito, and A. Blais, Reminis-
[11] T. Walter, P. Kurpiers, S. Gasparinetti, P. Magnard, A. Potočnik, cence of classical chaos in driven transmons, PRX Quantum
Y. Salathé, M. Pechal, M. Mondal, M. Oppliger, C. Eich- 4, 020312 (2023).
ler, et al., Rapid high-fidelity single-shot dispersive readout of [22] M. Khezri, E. Mlinar, J. Dressel, and A. N. Korotkov, Measur-
superconducting qubits, Physical Review Applied 7, 054020 ing a transmon qubit in circuit qed: Dressed squeezed states,
(2017). Physical Review A 94, 012347 (2016).
[12] Y. Sunada, K. Yuki, Z. Wang, T. Miyamura, J. Ilves, K. Mat- [23] J. Koch, M. Y. Terri, J. Gambetta, A. A. Houck, D. I. Schus-
suura, P. A. Spring, S. Tamate, S. Kono, and Y. Naka- ter, J. Majer, A. Blais, M. H. Devoret, S. M. Girvin, and R. J.
mura, Photon-noise-tolerant dispersive readout of a supercon- Schoelkopf, Charge-insensitive qubit design derived from the
19
cooper pair box, Physical Review A 76, 042319 (2007). abatement of spontaneous emission in circuit quantum electro-
[24] A. Blais, R.-S. Huang, A. Wallraff, S. M. Girvin, and R. J. dynamics, Applied Physics Letters 107 (2015).
Schoelkopf, Cavity quantum electrodynamics for supercon- [41] N. T. Bronn, E. Magesan, N. A. Masluk, J. M. Chow, J. M.
ducting electrical circuits: An architecture for quantum com- Gambetta, and M. Steffen, Reducing spontaneous emission in
putation, Phys. Rev. A 69, 062320 (2004). circuit quantum electrodynamics by a combined readout/filter
[25] F. Beaudoin, J. M. Gambetta, and A. Blais, Dissipation and ul- technique, IEEE Transactions on Applied Superconductivity
trastrong coupling in circuit qed, Physical Review A 84, 043832 25, 1 (2015).
(2011). [42] J. Heinsoo, C. K. Andersen, A. Remm, S. Krinner, T. Walter,
[26] A. Houck, J. Schreier, B. Johnson, J. Chow, J. Koch, J. Gam- Y. Salathé, S. Gasparinetti, J.-C. Besse, A. Potočnik, A. Wall-
betta, D. Schuster, L. Frunzio, M. Devoret, S. Girvin, et al., raff, et al., Rapid high-fidelity multiplexed readout of supercon-
Controlling the spontaneous emission of a superconducting ducting qubits, Physical Review Applied 10, 034040 (2018).
transmon qubit, Physical review letters 101, 080502 (2008). [43] E. A. Sete, J. M. Martinis, and A. N. Korotkov, Quantum theory
[27] R. Dassonneville, T. Ramos, V. Milchakov, L. Planat, E. Du- of a bandpass purcell filter for qubit readout, Physical Review
mur, F. Foroughi, J. Puertas, S. Leger, K. Bharadwaj, J. De- A 92, 012325 (2015).
laforce, C. Naud, W. Hasch-Guichard, J. J. Garcı́a-Ripoll, [44] J. Gambetta, A. Blais, M. Boissonneault, A. A. Houck, D. I.
N. Roch, and O. Buisson, Fast high-fidelity quantum nondemo- Schuster, and S. M. Girvin, Quantum trajectory approach to cir-
lition qubit readout via a nonperturbative cross-kerr coupling, cuit qed: Quantum jumps and the zeno effect, Phys. Rev. A 77,
Phys. Rev. X 10, 011045 (2020). 012112 (2008).
[28] Y. Ye, K. Peng, M. Naghiloo, G. Cunningham, and K. P. [45] M. A. Castellanos-Beltran, K. Irwin, G. Hilton, L. Vale, and
O’Brien, Engineering purely nonlinear coupling between super- K. Lehnert, Amplification and squeezing of quantum noise
conducting qubits using a quarton, Phys. Rev. Lett. 127, 050502 with a tunable josephson metamaterial, Nature Physics 4, 929
(2021). (2008).
[29] L. Neumeier, M. Leib, and M. J. Hartmann, Single-photon tran- [46] Z. Wang, S. Shankar, Z. Minev, P. Campagne-Ibarcq, A. Narla,
sistor in circuit quantum electrodynamics, Phys. Rev. Lett. 111, and M. H. Devoret, Cavity attenuators for superconducting
063601 (2013). qubits, Physical Review Applied 11, 014031 (2019).
[30] M. Leib, P. Zoller, and W. Lechner, A transmon quantum an- [47] G. Catelani, R. J. Schoelkopf, M. H. Devoret, and L. I. Glaz-
nealer: decomposing many-body ising constraints into pair in- man, Relaxation and frequency shifts induced by quasiparti-
teractions, Quantum Science and Technology 1, 015008 (2016). cles in superconducting qubits, Physical Review B 84, 064517
[31] S. E. Nigg, H. Paik, B. Vlastakis, G. Kirchmair, S. Shankar, (2011).
L. Frunzio, M. Devoret, R. Schoelkopf, and S. Girvin, Black- [48] A. Somoroff, Q. Ficheux, R. A. Mencia, H. Xiong, R. Kuzmin,
box superconducting circuit quantization, Physical Review Let- and V. E. Manucharyan, Millisecond coherence in a supercon-
ters 108, 240502 (2012). ducting qubit, Physical Review Letters 130, 267001 (2023).
[32] E. Jeffrey, D. Sank, J. Mutus, T. White, J. Kelly, R. Barends, [49] F. Henriques, F. Valenti, T. Charpentier, M. Lagoin, C. Gou-
Y. Chen, Z. Chen, B. Chiaro, A. Dunsworth, et al., Fast accu- riou, M. Martı́nez, L. Cardani, M. Vignati, L. Grünhaupt,
rate state measurement with superconducting qubits, Physical D. Gusenkova, et al., Phonon traps reduce the quasiparticle
review letters 112, 190504 (2014). density in superconducting circuits, Applied physics letters 115
[33] J. R. Johansson, P. D. Nation, and F. Nori, Qutip: An open- (2019).
source python framework for the dynamics of open quantum [50] R. Gordon, C. Murray, C. Kurter, M. Sandberg, S. Hall, K. Bal-
systems, Computer Physics Communications 183, 1760 (2012). akrishnan, R. Shelby, B. Wacaser, A. Stabile, J. Sleight, et al.,
[34] K. Jacobs and D. A. Steck, A straightforward introduction to Environmental radiation impact on lifetimes and quasiparticle
continuous quantum measurement, Contemporary Physics 47, tunneling rates of fixed-frequency transmon qubits, Applied
279 (2006). Physics Letters 120 (2022).
[35] P. Bertet, I. Chiorescu, G. Burkard, K. Semba, C. J. P. M. Har- [51] L. B. Nguyen, G. Koolstra, Y. Kim, A. Morvan, T. Chistolini,
mans, D. P. DiVincenzo, and J. E. Mooij, Dephasing of a super- S. Singh, K. N. Nesterov, C. Jünger, L. Chen, Z. Pedramrazi,
conducting qubit induced by photon noise, Phys. Rev. Lett. 95, et al., Blueprint for a high-performance fluxonium quantum
257002 (2005). processor, PRX Quantum 3, 037001 (2022).
[36] F. Yan, D. Campbell, P. Krantz, M. Kjaergaard, D. Kim, J. L. [52] C. Wang, X. Li, H. Xu, Z. Li, J. Wang, Z. Yang, Z. Mi, X. Liang,
Yoder, D. Hover, A. Sears, A. J. Kerman, T. P. Orlando, et al., T. Su, C. Yang, et al., Towards practical quantum computers:
Distinguishing coherent and thermal photon noise in a cir- Transmon qubit with a lifetime approaching 0.5 milliseconds,
cuit quantum electrodynamical system, Physical Review Letters npj Quantum Information 8, 3 (2022).
120, 260504 (2018). [53] D. T. McClure, H. Paik, L. S. Bishop, M. Steffen, J. M. Chow,
[37] K. O’Brien, Gradiometric quarton for nonlinear coupling of su- and J. M. Gambetta, Rapid driven reset of a qubit readout res-
perconducting qubits and resonators, Bulletin of the American onator, Physical Review Applied 5, 011001 (2016).
Physical Society (2023). [54] B. Lienhard, Machine Learning Assisted Superconducting
[38] M. Kounalakis, C. Dickel, A. Bruno, N. Langford, and Qubit Readout, Ph.D. thesis, Massachusetts Institute of Tech-
G. Steele, Tuneable hopping and nonlinear cross-kerr interac- nology (2021).
tions in a high-coherence superconducting circuit, npj Quantum [55] L. Chen, H.-X. Li, Y. Lu, C. W. Warren, C. J. Križan, S. Kosen,
Information 4, 38 (2018). M. Rommel, S. Ahmed, A. Osman, J. Biznárová, et al., Trans-
[39] M. D. Reed, B. R. Johnson, A. A. Houck, L. DiCarlo, J. M. mon qubit readout fidelity at the threshold for quantum error
Chow, D. I. Schuster, L. Frunzio, and R. J. Schoelkopf, Fast re- correction without a quantum-limited amplifier, npj Quantum
set and suppressing spontaneous emission of a superconducting Information 9, 26 (2023).
qubit, Applied Physics Letters 96 (2010). [56] R. Hanai, A. McDonald, and A. Clerk, Intrinsic mechanisms
[40] N. T. Bronn, Y. Liu, J. B. Hertzberg, A. D. Córcoles, A. A. for drive-dependent purcell decay in superconducting quantum
Houck, J. M. Gambetta, and J. M. Chow, Broadband filters for circuits, Physical Review Research 3, 043228 (2021).
20