Arxiv:2003.03307V2 [Quant-Ph] 10 Feb 2021

Home , Classical capacity, Quantum cryptography, Quantum simulator, Qubit, Qutrit, Transmon

Quantum Information Scrambling on a Superconducting Qutrit Processor

M.S. Blok,1, 2, ∗ V.V. Ramasesh,1, 2, ∗ T. Schuster,2 K. O’Brien,2 J.M. Kreikebaum,1, 2 D. Dahlen,1, 2 A. Morvan,1, 2 Beni Yoshida,3 N.Y. Yao,2 and I. Siddiqi1, 2 1Lawrence Berkeley National Laboratory, Berkeley, CA 94720, United States of America 2Department of Physics, University of California, Berkeley, California 94720 USA 3Perimeter Institute for Theoretical Physics, Waterloo, Ontario N2L 2Y5, Canada (Dated: February 11, 2021) The dynamics of quantum information in strongly-interacting systems, known as quantum information scrambling, has recently become a common thread in our understanding of black holes, transport in exotic non-Fermi liquids, and many-body analogs of quantum chaos. To date, verified experimental implementations of scrambling have focused on systems composed of two-level qubits. Higher-dimensional quantum systems, however, may exhibit different scrambling modalities and are predicted to saturate conjectured speed limits on the rate of quantum information scrambling. We take the first steps toward accessing such phenomena, by realizing a quantum processor based on superconducting qutrits (three-level quantum systems). We demonstrate the implementation of universal two-qutrit scrambling operations and embed them in a five-qutrit quantum teleportation protocol. Measured teleportation fidelities, Favg = 0.568 ± 0.001, confirm the presence of scrambling even in the presence of experimental imperfections and decoherence. Our teleportation protocol, which connects to recent proposals for studying traversable wormholes in the laboratory, demonstrates how quantum technology that encodes information in higher-dimensional systems can exploit a larger and more connected state space to achieve the resource efficient encoding of complex quantum circuits.

I. INTRODUCTION dynamics enable thermalization in closed classical systems, scrambling dynamics enable thermalization of iso- While the majority of current generation quantum pro- lated many-body quantum systems by creating highly- cessors are based on qubits, qutrit-based (and more gen- entangled states. In general, it is difficult to deter- erally qudit-based [1]) systems have long been known to mine whether arbitrary many-body Hamiltonians lead to exhibit significant advantages in the context of quan- scrambling dynamics; this is one area where future quantum technology: they have been touted for their small tum processors could offer experimental guidance. Our code-sizes in the context of quantum error correction [2], realization of verifiable scrambling behavior in a multi- high-fidelity magic state distillation [3], and more robust qutrit system is a step towards this goal. quantum cryptography protocols [4, 5]. To date, advan- In particular, a quantum processor can, in principle, di- tages of individual qutrits have been explored experimen- rectly measure the scrambling-induced spread of initially tally in fundamental tests of quantum mechanics [6] and localized information via the decay of so-called out-of- in certain quantum information protocols [7–10], while time-ordered correlation functions (OTOCs) [13, 16–23]. entanglement between qutrits has been demonstrated in This capability was recently demonstrated in a qubit- probabilistic photonic systems [11, 12]. A qutrit plat- based system, using a seven-qubit teleportation algo- form capable of implementing deterministic high-fidelity rithm analogous to the five-qutrit protocol we implement gates would be a powerful tool for both quantum sim- in this work [24–26]. A key property of the teleportation- ulation and information processing. In this work, we based protocol is that it enables verification of scrambling develop a prototypical multi-qutrit processor based on behavior even in the face of decoherence and experimen- superconducting transmon circuits (Fig. 1), and—as a tal imperfection. proof-of-principle demonstration—use it to (i) perform arXiv:2003.03307v2 [quant-ph] 10 Feb 2021 The superconducting qutrit processor we develop here a maximally-scrambling two-qutrit unitary and (ii) ver- features long coherence times; multiplexed readout of ify its scrambling behavior using a five-qutrit quantum individual qutrits; fast, high-fidelity single-qutrit opera- teleportation algorithm. tions; and two types of two-qutrit gates for generating en- Quantum scrambling, the subject of much recent inter- tanglement. Using this gateset on our processor, we con- est, is the quantum analogue of chaotic dynamics in clas- struct a maximally scrambling qutrit unitary and char- sical systems: scrambling describes many-body dynam- acterize it using quantum process tomography. Finally, ics which, though ultimately unitary, scatter initially- to demonstrate the ability of our platform to perform localized quantum information across all of the system’s genuine multi-qutrit algorithms, we perform a five-qutrit available degrees of freedom [13–15]. Just as chaotic teleportation protocol which serves as an additional verification of genuine two-qutrit scrambling behavior. This protocol is inspired by the Hayden-Preskill variant of the black-hole information paradox [14, 24]. While we have ∗ These authors contributed equally to this work. chosen quantum scrambling as a demonstration of our 2

a b c 2 0 ω01 t 2 1 ω 0 12 t P(0) I P(1)

P(2) 1

Pulse duration t (μs) Q

Figure 1. Superconducting qutrit processor — a, Optical micrograph of the five-transmon processor used in this experiment. Transmon circuits (light blue) couple to an integrated Purcell-filter and readout bus (red) via individual linear resonators (gold), enabling multiplexed state measurement. Exchange coupling between nearest-neighbor transmons is mediated by resonators (purple), while microwave drive lines (green) enable coherent driving of individual qubits. b, Coherent Rabi dynamics of a single qutrit induced by simultaneous microwave driving at frequencies ω01 and ω12. Achievable Rabi frequencies are in the range of tens of MHz, three orders of magnitude faster than decoherence timescales. c, Example single-shot readout records of an individual qutrit, generally achievable with fidelities above 0.95. This is largely limited by decay during readout. processor, our work opens the door more broadly to the ring geometry (Fig. 1a) of which we use five (denoted experimental study of quantum information processing Q1,...,Q5) to realize quantum circuits. Each transmon utilizing qutrit-based logic. encodes a single qutrit and is coupled both to a dedicated microwave control line (for performing gates, Fig. 1b) and its own readout resonator (for state measurement, Fig. 1c). Transmons are quantum nonlinear oscillators II. QUTRIT PROCESSOR that can be operated as qubits, using only their two lowest-lying energy states |0i and |1i. While their higher Nearly all current quantum processors are based on col- energy states (|2i, |3i, etc.) in principle enable transmons lections of the simplest possible quantum system: qubits, to be operated as higher-dimensional qutrits or qudits consisting of two states per site. In contrast, qudits – fea- [38] , the experimental implementation of multi-qutrit turing d > 2 states – can store an exponentially greater algorithms is challenging due to a lack of two-qutrit en- amount of information compared to qubits and can there- tangling gates and increased noise associated with the fore, in certain cases, implement quantum algorithms us- higher transmon states. ing smaller systems and fewer multi-site entangling gates [27, 28]. In cavity QED experiments, continuous vari- able quantum information has been encoded [29, 30], pro- tected [31] and entangled [32] in linear cavity modes, by A. High-fidelity single qutrit operations coupling them to an artificial non-linear atom. Encod- ing discrete quantum information in an intrinsically non- In order to implement high-fidelity single-qutrit oper- linear system has the advantage that energy transitions ations in transmons, one must overcome multiple sources have distinct frequencies that can be driven individually of noise and coherent errors that are naturally introduced and with faster gates. upon including the |2i state of transmon in the compu- For qutrit systems specifically, many quantum infor- tational subspace. The primary such sources include: mation protocols have been proposed that would yield a (i) Relaxation due to spontaneous emission—Energy significant advantage over qubits [2–5, 33]. Qutrits have relaxation or T1 processes in transmon qubits cause un- been successfully realized in various physical degrees of wanted, incoherent transitions from the |1i state to the freedom including the polarization state of multiple pho- |0i state, and arise from lossy interfaces or electromag- tons [34], spin S = 1 states of solid-state defect centers netic channels. For transmon qutrits, the addition of [35], hyperfine states in trapped atoms and ions [36, 37] the |2i state introduces another decay channel, from and the lowest energy states of an anharmonic oscillator |2i → |1i [41]. Due to bosonic enhancement, in which in superconducting circuits [38]. However, no platform the spontaneous emission rate scales linearly with pho- to date has demonstrated a deterministic, universal two- ton number, the time constant associated with |2i → |1i qutrit gate. decay is roughly half that of the |1i → |0i transition. Our superconducting qutrit processor features eight To address this, we use a high-quality fabrication pro- transmons [39, 40] connected in a nearest-neighbor cess and careful microwave engineering to minimize the 3 effects of energy relaxation. The fabrication recipe, de- 2. Microwave fields leaking onto neighboring qutrit(s) tailed in Appendix A, is optimized to remove lossy ox- will result in an unwanted cross-resonance interac- ide layers at the substrate-metal and metal-metal inter- tion, making the desired Rabi frequency dependent faces through three separate buffered-oxide etches and on the state of the neighboring qutrit(s) [46]. an ion-mill step. In addition, finite-element simulations are used to identify and mitigate each loss channel as- For a given drive frequency, our cross-talk can be char- sociated with microwave radiation. Specific mitigation acterized in terms of a five-by-five complex-valued matrix strategies include: an integrated, broadband Purcell fil- C(ω), which relates the field amplitudes ~o(ω) seen by ter suppressing leakage into the readout bus (Fig. 1a); each of the five qutrits to the input field amplitudes ~i(ω) readout and bus resonator geometries designed to shift on each drive line: ~o(ω) = C(ω)~i(ω). While this cross- the frequencies of lossy higher modes away from those talk matrix strongly depends on frequency, it does not of qutrits and resonators of band; and extensive wire- exhibit any nonlinearity at the powers used in the exper- bonding connecting the ground planes on either sides of iment. Therefore, we compensate for the cross-talk by the readout bus, coupling resonators, readout resonators, inverting the matrix C(ω) at each frequency to yield lin- and control lines. As a results of these techniques, the ear combinations of drive lines which result in non-zero average T1 times on the chip are 56.0 ± 10 µs for the fields only at a single qutrit. |1i → |0i transition, and 34.8 ± 4 µs for the |2i → |1i To measure the cross-talk matrix C(ω) of our system, transition. we drive two control lines simultaneously, and, for each (ii) Dephasing due to the charge-sensitivity of higher pair of driven control lines, determine the relative am- transmon levels—The transmon was originally developed plitudes and phases of driving that cancel the resulting to reduce the dependence of energy levels on offset charge field at each qutrit on the chip. Depending on the relative noise. While typical values of transmon parameters (i.e. frequencies between the drive field and the qutrit transi- the Josephson energy EJ and the charging energy EC ) re- tion, we use either an AC Stark shift or a Rabi oscillation sult in low charge-noise sensitivity in the qubit subspace, as the diagnostic of an unwanted microwave field. This the charge dispersion increases rapidly with increasing measurement is repeated for each of ten drive frequencies energy levels [39, 42]. For qutrits specifically, the charge of interest (i.e. the |0i ↔ |1i and |1i ↔ |2i transition fre- dispersion of the |2i state is at least an order of mag- quencies of all five qutrits), each pair of lines, and each nitude greater than that of the |1i state, resulting in a qutrit. charge-limited dephasing time ten times lower than that While this method proved effective for our system, it of the qubit subspace. can become prohibitively measurement-intensive for fu- To mitigate charge noise in the |2i state, we tune the ture processors with hundreds of qudits. Looking for- transmon parameters even further than typical into the ward, in order to scale to such systems it will be crucial “transmon regime”: specifically, we choose a ratio of to pinpoint and mitigate the source of this cross-talk at Josephson and charging energies EJ /EC ≈ 73 (typically, the hardware level. this ratio is near 50 for transmon qubits [43–45]). This With the above three issues—relaxation, dephasing, reduces the charge dispersion of the |2i and |1i states and cross-talk—taken care of, we now turn to our re- to 12 kHz and 250 Hz, respectively. As a result, we realization of high-fidelity single-qutrit gates. In qutrits, alize dephasing times, averaged over the five qutrits, of these require the combination of a microwave pulse (to ∗ T2 = (39±21) µs [(14±5) µs] for the |0i → |1i [|1i → |2i] perform the rotation in the relevant subspace) and a vir- transitions, which can be further extended with a Hahn- tual phase-gate. This phase gate is necessary because a echo to T2,echo = (61.2 ± 11) µs [(28 ± 5) µs]. microwave field that addresses, e.g. the |0i → |1i transi- (iii) Increased cross-talk due to frequency crowding— tion, also generates a relative phase ei(φG+φD ) |2i on the If pulses, intended to apply unitary operations on a sin- idling state, where φG is the geometric (Berry) phase as- gle qutrit, are not well localized in space to the desired sociated with the rotation in the {|0i , |1i} subspace and qutrit, they can induce unintended unitary operations φD is the dynamical phase due to the AC Stark shift of on nearby qutrits. This ‘cross-talk’ is increasingly trou- the idling |2i state. These phases represent logical Z er- blesome when including the |2i state, as the frequency rors in the computational subspace, which we measure spectrum of state transitions becomes more crowded due and correct. to the inclusion of the |1i → |2i transition frequencies. Single-qutrit gates in our system are performed within On our processor, we find significant cross-talk between 30 ns (see Apendix C for the universal set of gates used the microwave drive lines for most transmons. When in our experiment). We characterize them by performing driving Rabi oscillations on a given qutrit, this cross-talk qubit-based randomized benchmarking in two different has two unwanted effects: qubit subspaces [47]. These yield fidelities on par with state-of-the-art qubit processors: f01 = 0.9997 ± 0.0001 1. All other qutrits are off-resonantly driven. Depend- and f12 = 0.9994 ± 0.0001 for gates within the {|0i , |1i} ing on the relative frequencies between the qutrits, and {|1i , |2i} subspace, respectively. While this bench- this can manifest either as an unwanted shift in marking method demonstrates that single-qutrit coher- state populations or as an AC Stark shift. ence is no longer a major bottleneck for high-fidelity op- 4 erations, it should be noted that this method is not sen- a Control Target sitive to certain sources of errors, including phase errors 2 in the idle state and multi-qutrit errors. In the future, a 1 full characterization of qutrit operations will require the 0 t =125 ns development of genuine qutrit randomized benchmarking b g Control Target protocols. Control in 0 2 Multiplexed, individual qutrit readout is performed Control in 1 1 dispersively via a linear readout resonator coupled to Control in 2 0 each transmon. As shown in Fig. 1c, with a suit- able choice of the readout frequency and amplitude all c three states of the transmon can be resolved in a single π 0 θ π shot. Averaged over all qutrits, our readout fidelity is Control Favg = 0.96 ± .02. In ensemble measurements (e.g. when Target 0 π π π performing tomography), we can correct for this readout imperfection by storing a ‘confusion matrix’ Mij of the conditional probabilities to measure a state |ii given an actual state |ji. Applying its inverse to the measurement results allows one to infer the actual state populations.

B. Two-qutrit entanglement

Having outlined our implementation of high-fidelity single-qutrit operations, we next demonstrate two methods for generating controllable two-qutrit entangling gates in our system: the first based on the cross- resonance interaction [46, 48], and the second based on Figure 2. Two-qutrit EPR pair generation via the the longitudinal, or cross-Kerr, interaction. Both of these cross-resonance interaction — a Nearest-neighbor qutrits operate on neighboring qutrits, giving our full eight- coupled by an exchange interaction can be entangled via qutrit processor an effective ring geometry; restricting to the cross-resonance effect, where one qutrit (the control) is five qutrits, this reduces to a linear geometry. The two microwave-driven at the |0i ↔ |1i transition frequency of the types of interaction give rise to two distinct forms of en- other (the target). Resulting Rabi oscillations of the target tangling gates: the cross-resonance interaction leads (up qutrit exhibit a Rabi frequency dependent on the state of the to a phase) to a conditional subspace rotation, while the control qutrit. Here, we drive the control with a field whose cross-Kerr interaction leads to a controlled-SUM gate— amplitude is chosen to make the Rabi frequencies correspond- the qudit analogue of the controlled-NOT gate. We now ing to control states |0i and |2i identical, resulting in a unitary operation which, after 125 ns, interchanges states |0i and |1i describe both of these gates in detail. of the target qutrit when the control qutrit is the |1i state. b, Cross-resonance gate—The cross-resonance interac- When the target qutrit is in the |2i state, the cross-resonance tion has been used extensively in superconducting interaction is off-resonant and does not affect the population. qubit systems to generate high-fidelity controlled-NOT c (top), Sequence used to prepare an EPR pair with two 01 gates [49–51]. The origin of the interaction is a weak applications of the cross-resonance gate. πx represents be a transverse coupling between two qubits that are highly rotation of π about the x axis in the {|0i , |1i} subspace of off-resonant such that the coupling is almost negligible. the qutrit. The number in the circle for the two-qutrit gate When one of the two systems (the control) is driven at represents the condition (state of the control qutrit) for which the frequency of the other (the target), the combination a gate is applied on the target. (bottom) The density matrix of the coupling and the drive induces Rabi oscillations on (reconstructed via state tomography) of the resulting EPR pair, with a state fidelity of F = 0.98 ± 0.002. the target qubit, with a Rabi frequency that depends on EPR the state of the control qubit. This results in two Rabi frequencies, ω0 and ω1, for the two control qubit states, equivalent to a σ ⊗ σ interaction, and periodically en- |0i and |1i, both of which also depend on the drive power. z x tangles and disentangles qubits at a rate ∆ω = |ω0 − ω1|. This interaction is naturally described by the Hamilto- When the two oscillations are completely out of phase nian (t∆ω = π), the resulting unitary qubit Hcr /~ = ω0 |0ih0| ⊗ σx + ω1 |1ih1| ⊗ σx. (1) UCNOT = |0ih0| ⊗ I + |1ih1| ⊗ σx (2) Higher-order deviations from this Hamiltonian can is locally equivalent to a controlled-NOT gate. largely be mitigated with dynamical decoupling, local On our chip, the cross-resonance interaction arises rotations, or pulse shaping. This Hamiltonian is locally from the static capacitive coupling between neighboring 5 transmons. To extend the cross-resonance gate to qutrit Q1/Q2 Q2/Q3 Q3/Q4 Q4/Q5 systems, we must consider how this interaction behaves α11 -279 -138 -276 -262 in the full two-qutrit Hilbert space. We will show that we α12 160 158 -631 -495 can realize a similar gate to that of qubit systems, where α21 -528 -335 243 -528 we swap the |0i and |1i populations of the target qutrit α22 -743 -342 -748 -708 conditional on the control qutrit being in the |1i state. The dynamics of the two-qutrit cross-resonance inter- Table I. Measured cross-Kerr interaction strengths between nearest-neighbor pairs of transmons, in units of kHz. action are shown in Fig. 2(a-b). Compared to qubits, we have two additional considerations: the effect of the cross-resonance drive on the |2i state of the target qutrit, In order to benchmark the performance of the and the Rabi oscillations induced on the target qutrit by conditional-π gate, we use two applications of the gate the |2i state of the control qutrit. The first effect results to create a maximally entangled two-qutrit EPR pair: in an overall phase being applied to the |2i state of the target via the AC Stark shift (Fig. 2b), since the effective |00i + |11i + |22i |EPRi = √ . (5) Rabi drive is off-resonant with the |1i → |2i transitions of 3 the target qutrit (in contrast to the |0i → |1i transition, with which it is resonant). The second effect extends the We measure the success of EPR preparation using qutrit qubit cross-resonance Hamiltonian to qutrits according state tomography [38], enabled by our high-fidelity local to qutrit pulses and qutrit readout. From this we recon- struct the density matrix of our final state (Fig. 2c) and Hqutrit/ = ω |0ih0| ⊗ π01 + ω |1ih1| ⊗ π01 cr ~ 0 x 1 x compute an EPR fidelity of FEPR = 0.98 ± 0.002, which 01 + ω2 |2ih2| ⊗ πx , (3) we observe to be mostly limited by decoherence. Finally, while the conditional-π gate is fast and ca- where we have added the (drive-dependent) frequency ω2, pable of generating high-fidelity entangled states, it acts which induces Rabi oscillations in the target qutrit when only in a two-qubit subspace of the full two-qutrit Hilbert the control qutrit is in the |2i state. space. Consequently, despite the conditional-π gate be- 01 In the above Hamiltonian, πx is the 3 × 3 Gell-Mann ing universal for qutrit computation when combined with matrix that couples the |0i and |1i states, and does noth- arbitrary single qutrit gates [54], implementing general ing to the |2i state. We emphasize that this model for the two-qutrit unitaries will require a high number of such interactions, while physically motivated and consistent gates. with our experimental observations, is only approximate. This can be undesirable given the finite error associ- Much like for the two-qubit cross-resonance gate [52, 53], ated with each gate. To address this, in what follows we further study is needed to develop a precise model of the introduce a second qutrit entangling gate, the controlled- dynamics. SUM gate, which we will implement via the cross-Kerr The qutrit cross-resonance interaction Eq. (3) involves interaction instead of the cross-resonance interaction. Ul- three distinct Rabi frequencies, corresponding to the timately, we will use both gates to perform the full five- three states of the control qutrit (Fig. 2a). At low drive qutrit teleportation protocol, applying the conditional- power, these frequencies scale linearly with the drive π gate to EPR preparation/measurement and the more amplitude, while at higher powers the dependence be- powerful controlled-SUM gate to a complex two-qutrit comes nonlinear. We utilize this nonlinearity to choose scrambling operation. a drive power such that two of the Rabi frequencies Cross-Kerr gate—As neighboring qutrits in our pro- are equal, ω0 = ω2. Next, similar to the qubit im- cessor are capacitively coupled to one another, they are plementation, we choose the gate time tg according to subject to an “always-on” cross-Kerr entangling interac- tg|ω0 −ω1| = π. Up to state-dependent phases, this gives tion, which we harness to perform controlled-phase and rise to a conditional-π gate (also known as a generalized controlled-SUM gates (the two are locally equivalent). controlled-X gate [54]): The always-on cross-Kerr interaction, analogous to the dispersive interaction between a qubit and cavity, is mod- U = (|0ih0| + |2ih2|) ⊗ I Cπ eled by the dispersive cross-Kerr Hamiltonian + |1ih1| ⊗ (|0ih1| + |1ih0| + |2ih2|). (4) HcK/~ = α11 |11ih11| + α12 |12ih12| + (6) Our experiment features a frequency difference ω0 −ω1 = α21 |21ih21| + α22 |22ih22| , 4 MHz and a corresponding gate time tg = 125 ns. Dur- ing this gate, simultaneously with the cross-resonance where we work in a rotating frame that cancels the drive applied to the control qutrit, we apply a concurrent Hamiltonian’s action on the |0ji , |i0i states. The cross- drive to the target qutrit which ensures that it ends up Kerr coefficients αij represent the rate at which phases in a computational basis state at tg. This target drive, accumulate on the |iji state during idling. Their val- with Rabi frequencies in the tens of MHz, has the ad- ues are sensitive to both the magnitude of the capaci- ditional benefit of suppressing dephasing in the target tive coupling between transmons as well as the transmon qutrit’s {|0i , |1i} subspace during the gate [55]. spectra. While in principle one could design a processor 6 with pre-specified values of αij, this is impractical given and |2i states. The simplest construction uses four seg- current technology because the transmon spectra depend ments of cross-Kerr evolution. 12 on the critical current of their Josephson junction, which Denoting a swap pulse on qutrit q as πq , the total cannot be precisely set in fabrication [56]. To maintain pulse sequence is given by the decomposition: robustness against such variability, we therefore design −iTAHcK 12 −iTB HcK UCφ = e · π · e the cross-Kerr gate to work regardless of the particular 0 (11) 12 −iTC HcK 12 −iTD HcK 12 values of the coefficients. For reference, these values for · π1 · e · π0 · e · π1 , our particular processor are shown in Table I. where T , T , T , T denote the time durations of We will now demonstrate how to combine a generic A B C D each segment of cross-Kerr evolution. To determine these cross-Kerr Hamiltonian with single-qutrit rotations to times from the interaction coefficients, note that each form a controlled-SUM gate. The foundation of our con- evolution segment serves to add a phase φ = α T to the struction is the local equivalence between the controlled- ij ij |iji state (here, i, j ∈ {1, 2}), while the swap pulses inter- SUM gate: change the indices 1 ↔ 2 on the acted upon qutrit. Our d choice of swap gates guarantees that each state spends X n exactly one segment under each interaction coefficient, UCSUM = |nihn| ⊗ X (7) n=1 and returns to itself at the end of the gate. The unitary thus amounts to a phase applied to each state equal to a and the controlled-phase gate UCφ: linear combination of the four evolution times, with the transformation matrix determined by the interaction co- d efficients. For generic interaction coefficients this linear X n UCφ = |nihn| ⊗ Z , (8) transformation from interaction times to applied phases n=1 is full rank, which guarantees that the method can in principle generate any combination of two-qutrit phases. which we will soon show is achievable via the cross-Kerr With our particular interaction coefficients, this interaction. Here controlled-phase gate implementation takes a total time d d ∼ 1.5 µs for each of the qutrit pairs (Q1,Q2) and X X X = |j + 1ihj| ,Z = ei2πj/d |jihj| (9) (Q3,Q4). We characterize the performance of the full j=1 j=1 controlled-SUM gate via quantum process tomography on the full two-qutrit subspace (Fig. 3b) and achieve a are the generators of the set of qudit Pauli operators. The fidelity of 0.889, primarily limited by decoherence occur- two gates are related via conjugation by a single-qutrit ring throughout the cross-Kerr time evolution. Hadamard H gate: Combined with the single-qutrit control demonstrated in the previous section, each of the conditional-π gate † (I ⊗ H )UCφ(I ⊗ H) = UCSUM, (10) and the conditional-SUM gate enable universal quantum computation on our qutrit processor [54]. Having both which, as for qubits, interchanges the Pauli X and Z op- gates on hand provides additional flexibility in gate com- erators. This realization also easily generalizes to slight pilation as the different entangling gates may be more modifications of the controlled-SUM gate: one can re- adept in different scenarios. alize instead a controlled-MINUS gate by inverting the Hadamard, and one can interchange the control and the target qudits by switching which qudit is Hadamard con- III. QUTRIT SCRAMBLING jugated. Specifying to qutrits, the controlled-phase gate UCφ We now turn to proof-of-principle experiments demon- imparts a phase 2π/3 to the |11i, |22i states and the strating the capabilities of our qutrit processor, specif- opposite phase −2π/3 to the |12i, |21i states. ically by demonstrating and verifying quantum scram- We have developed two methods for implementing this bling. We will demonstrate the implementation of a gate via the cross-Kerr interaction. The first, described two-qutrit scrambling unitary, and verify that it scram- in what follows, is conceptually simpler and uses only four bles using a five-qutrit quantum teleportation algorithm. single-qutrit pulses, but is prone to local dephasing and Quantum scrambling, loosely defined as the delocaliza- to errors arising from a nonzero cross-Kerr interaction tion of initially local operators induced by many-body with qutrits outside the target-control pair. The second time dynamics, is an area of active current research; it method addresses these errors via dynamical decoupling is likely that quantum simulations performed on future in the specific context of the five-qutrit teleportation pro- processors will deepen our understanding of this phe- tocol, and is described in conjunction with the protocol nomenon and many-body quantum dynamics more gen- in Section I. erally. To date, experimental realizations of scrambling The essential concept behind these constructions is to have been performed with qubits; however, many in- intersperse time-evolution under the native cross-Kerr teresting scrambling-related phenomena occur in higher- Hamiltonian with single-qutrit pulses exchanging the |1i dimensional systems; see, e.g., [57–59]. Using quantum 7

Theory Experiment a Local gate Single Qutrit Two Qutrit Single Qutrit Two Qutrit

⊗ π

⊗ = π ⊗ π ⊗ = b Entangling gate ⊗

= φ ⊗ ⊗⊗⊗ ⊗⊗⊗ c Scrambler ⊗

⊗

⊗ ⊗ ⊗ ⊗ ⊗ ⊗ ⊗ ⊗ ⊗ ⊗ ⊗ ⊗ ⊗ ⊗ ⊗ ⊗ ⊗ ⊗ ⊗ ⊗

Figure 3. Quantum process tomography of the two-qutrit scrambling unitary — Results of process tomography both experimental and ideal. Plotted is part of the Pauli transfer matrix (where the vertical axis only includes the single qutrit operators, and PG represents the single-qutrit Pauli group excluding I) which shows the unitary’s action on single-qutrit Pauli operators when the unitary is a local gate (a), non-scrambling entangling gate (b) and the scrambling-unitary (c), . This directly verifies the key characteristic of scrambling by Us, that it maps all non-identity single-qutrit Pauli operators to two-qutrit operators. scrambling to demonstrate the capabilities of our proces- that completely delocalizes all single-qubit operators. A sor thus also represents a step towards such simulations system of two qutrits is the smallest bipartite system in of these high-dimensional composite systems. which maximal scrambling is possible. Specifically, we implement a simple Clifford scrambling In the next two sections, we verify the maximally- unitary Us, which effects the following permutation of the scrambling nature of our implementation in two ways: nine two-qutrit computational basis states [25]: (i) explicitly through quantum process tomography, and (ii) through a five-qutrit teleportation protocol inspired Us |m, ni = |2m + n, m + ni . (12) by quantum gravitational physics. As described in Ref. [25], the scrambling behavior of this unitary can be seen in the effect it has on the single-qutrit Pauli operators X and Z: A. Verifying scrambling through quantum process tomography U (Z ⊗ I) U † = Z ⊗ Z2 (13) † 2 2 U (I ⊗ Z) U = Z ⊗ Z (14) On our processor, the scrambling unitary Us is con- U (X ⊗ I) U † = X2 ⊗ X (15) structed by applying two controlled-SUM gates in se- U (X ⊗ I) U † = X ⊗ X. (16) quence, switching the control and target qubit between the two (Fig. 3c). In order to fully characterize our imple- Each single-qutrit Pauli operator is transformed into a mentation of the scrambling unitary Us we perform full two-qutrit operator, as required by scrambling. Addi- quantum process tomography. For a two-qutrit unitary, tionally, the transformations above show that Us is also this is a highly measurement-intensive procedure, which a Clifford scrambler, i.e. a scrambling unitary which is entails reconstructing, via two-qutrit state tomography, a member of the Clifford group: this is evidenced by the a 9 × 9 output density matrix for a complete set of 81 fact that the action of Us on Pauli operators is closed. input states. This requires millions of measurements – For a bit of context, we note that scrambling is not pos- each involving state preparation, application of Us, and sible in a two-qubit system: there is no unitary operation state measurement (in a different basis) – with the precise 8 number of repetitions determined by the desired statisti- Qutrits Scrambling cal uncertainty. For experimental platforms with a duty Black Hole cycle in the Hz range (e.g. trapped ions) this procedure would be prohibitively long; in contrast, superconducting Hawking circuits can reach repetition rates up to ∼ 100 kHz, such photon that the full process matrix can be measured in approxi- Wormhole mately 1 hour. This gives superconducting platform the unique ability to not only quantify the process fidelity of the implemented scrambling unitary but also to measure precisely how it scrambles. Early Radiation We implement the scrambling unitary using two se- quential controlled-SUM gates based on the cross-Kerr gate described in the previous section. In Fig. 3c, we depict the results of quantum process tomography on our implementation. Through this tomography, we find that the fidelity of the scrambling operation on our hard- Black Hole Hawking ware is 0.875, with two dominant error mechanisms: (i) photon dephasing and (ii) amplitude-damping during the cross- Kerr evolution. Quantum process tomography also allows us to directly visualize the entire action of the unitary via its effect on local operators; i.e., we can directly visualize (Fig. 3) Early Radiation the relations (13)-(16). In particular, in Fig. 3c we can Figure 4. Quantum teleportation circuit — Five-qutrit verify explicitly that Us transforms all single-qutrit operators into fully two-qutrit operators – the definition of a teleportation protocol used to test for scrambling by the two- qutrit unitary U , and its interpretation in the context of maximally scrambling unitary. We can also verify that it s black hole physics. a, We begin with the first qutrit in a does so as a Clifford unitary, i.e. all Pauli operators are quantum state |ψi, and prepare the remaining qutrits into transformed into different Pauli operators. For compar- two EPR pairs. The first qutrit Q1 is ‘scrambled’ with half of ison, we also illustrate the quantum process maps, both the first EPR pair by the unitary U, while the conjugate U ∗ theoretical and experimental, of a single-qutrit unitary is applied to Q3 and Q4. An EPR measurement on a pair of ∗ that does not delocalize any information (Fig. 3a), as qutrits (Q2 and Q3) exiting each of U, U serves to teleport well as a controlled-SUM gate, which is entangling but the initial quantum state to Q5 if and only if U is scrambling. not fully scrambling (Fig. 3b). In the former, we ver- b, In the context of the Hayden-Preskill thought experiment, ify that no single-qutrit operators are delocalized, while Q1 corresponds to Alice’s diary, which subsequently falls into in the latter we observe that all single-qutrit operators a black hole (Q2) that is maximally entangled with its past are delocalized except for Z(†) ⊗ I and I ⊗ X(†), which Hawking radiation (Q3), held by an outside observer Bob. In the gravitational picture, this shared entanglement functions commute with the controlled-SUM gate. similarly to a geometric wormhole between the black hole and its early radiation. The black hole dynamics are modelled as a scrambling unitary U, followed by emission of a Hawking ∗ B. Verifying scrambling through quantum photon (Q2). Finally, Bob applies U and measures in the teleportation EPR basis to recover Alice’s diary.

While process tomography provides an elegant and ex- haustive way to “image” scrambling performed by our unitary over local operators; in this context, maximal two-qutrit unitary, such an approach is infeasible for ver- scrambling by Us is captured by the fact that the av- ifying scrambling in larger systems, as the number of re- erage OTOC decays to its minimum allowed value (1/9 quired measurements scales exponentially with system for a two-qutrit system) [25]. Crucially, the protocol is size. As an example of a protocol which can circumvent constructed in such a way that faithful teleportation of a this difficulty, and thus could be used on larger quan- quantum state |ψi requires quantum information scram- tum processors, we turn to a teleportation protocol that bling to occur, as well as the absence of experimental quantifies the scrambling behavior of a unitary by us- errors. The teleportation fidelity can then in turn be ing scrambling to enable the teleportation of quantum used to upper-bound the OTOC, even in the face of such states [24–26]. Compared to tomography, a downside of errors. the teleportation protocol is that it is limited as an essen- As shown in Fig. 4, the verification protocol requires tially ‘one-parameter’ diagnostic of scrambling: it quan- both the scrambling unitary, Us, and its time-reversal, ∗ tifies the average amount that operators on a given qutrit Us , to be performed in parallel on separate pairs of are transformed to have support on an additional qutrit. qutrits. The qutrit pairs undergoing these two time- It does so via an average of OTOCs associated with the evolutions are initially highly correlated: two qutrits out 9 of the four (one from each pair) begin as a maximally en- of operators [24, 64]. This second form of teleportation ∗ tangled EPR pair. After applying Us,Us , the probability encapsulates the physics observed here. of measuring this same pair in the EPR state decreases Experimentally, the teleportation protocol requires to 1/9 – the same value as in a maximally mixed state – three essential ingredients: initial preparation of EPR due to scrambling with the additional qutrits. However, ∗ pairs, implementation of Us and Us , and measurement simply observing this decrease is not enough to robustly in the EPR basis. We have discussed our implemen- demonstrate scrambling, since the same decrease could tation of the first two of these ingredients in previous also have resulted from deterioration of the EPR pair sections. As our architecture only natively allows for under experimental decoherence. computational-basis measurements, we realize the EPR Here, teleportation comes to our aid: we place one of measurement on Q2 and Q3 by performing the EPR cre- the remaining two qutrits (Q1, the ‘input’) in an arbi- ation sequence in reverse, which transforms |EPRi to the trary pure single-qutrit state |ψi, and the other (Q4) in computational basis state |00i. We then measure all five an EPR pair with a fifth qutrit (Q5, the ‘output’). In qutrits in the computational basis and use the measure- the absence of experimental error, maximally scrambling ments outcomes showing Q2 and Q3 in |00i as a herald dynamics guarantee that whenever Q2 and Q3 happen for successful teleportation. We verify success by tomo- to be measured in their initial EPR state, the state |ψi graphically reconstructing the quantum state of Q5 after of Q1 is teleported to Q5. Unlike measuring a low EPR EPR measurement (by inserting state tomography pulses probability, high-fidelity teleportation can only arise from on Q5 before the final measurement), represented by the quantum information scrambling, not as a result of deco- density matrix ρout. herence, making the teleportation fidelity a robust di- All three of the steps described above work particu- agnostic of information scrambling [24, 25]. Further- larly well on pairs of qutrits in “isolation”, i.e. when more, while experimental error will lead to a decay of the pair’s neighboring qutrits are left in their ground the teleportation fidelity from unity, any measured fi- states. However, this is necessarily not the setting of any delity above the classical limit (0.5 for qutrits) places a five-qutrit quantum circuit, including the teleportation non-trivial upper-bound on the averaged OTOCs, and protocol. Here, we have to contend with the “always- thus the scrambling behavior, of Us [25]. on” nature of the cross-Kerr interaction, which leads to This association between scrambling and teleporta- unwanted entangling dynamics between all neighboring tion originated in black hole physics, and the teleporta- qutrit pairs, even as we attempt to perform our quan- tion protocol explored here is in fact inherited directly tum gates. To combat these unwanted interactions, we from this context. Its most straightforward interpre- develop a novel set of dynamical decoupling sequences tation is based on the Hayden-Preskill variant of the optimized for qutrits (see Appendix D for details). black hole information paradox [24, 60], as outlined in We perform teleportation of twelve different single- Fig. 4. Here, one pictures an observer Alice who has qutrit quantum states |ψi, corresponding to all possi- dropped a ‘diary’ consisting of the quantum state |ψi ble eigenstates of single-qutrit Pauli operators, i.e. three into a black hole. Meanwhile, an outside observer Bob eigenstates each for X,Z,XZ,XZ2 (the remaining Pauli wonders whether this information is recoverable, as op- operators are either inverses of these or the identity). We posed to being irreversibly destroyed by the black hole. calculate the teleportation fidelity from the reconstructed In seminal works [24, 60], it was shown that if (i) the density matrix of the output qubit as Fψ = hψ| ρout |ψi. black hole’s dynamics are approximated as a fully scram- Our chosen set of states is known as a ‘state 2-design’, bling quantum unitary Us and (ii) Bob possesses a large which guarantees that the average teleportation fidelity P resource of entanglement with the black hole (e.g. a col- Favg = (1/12) ψ Fψ measured for these twelve sam- lection of its early Hawking radiation), then the state |ψi pled states is in fact equal to the average teleportation can in fact be recovered via quantum operations on any fidelity over all possible states [25, 65]. Furthermore, few degrees of freedom emitted from the black hole (e.g. this average teleportation fidelity in fact allows us to up- one additional Hawking photon). per bound the average OTOCs associated with the im- In the gravitational picture, the shared entanglement plemented unitary by (4F − 1)−2, without making any between the black hole and its early radiation func- assumptions about the nature of the noise affecting our tions similarly to a geometric wormhole between the quantum processor [25]. two systems. Indeed, in a close variant of this pro- The results of teleportation for two different choices of tocol, teleportation has a precise interpretation as Al- unitary dynamics are shown in Fig. 5. First, as a con- ice’s state traversing a two-sided wormhole, in the spe- trol, we perform the protocol with the identity operation cific case where the unitary corresponds to time-evolution in place of Us. To ensure that the two have similar mag- under many-body dynamics with a holographic grav- nitudes and modes of experimental error, we implement ity dual [61–63]. Quite surprisingly, recent work has the identity with precisely the same number and type revealed that the same protocol also features a more of gates as the maximally scrambling unitary; the two generic, non-gravitational mechanism for teleportation, differ only in the magnitude of certain software-defined which displays remarkably similar features to gravita- phase gates. Since the identity operator does not scram- tional teleportation but is based only on the spreading ble quantum information, we observe trivial teleportation 10 a Scrambling Qutrit 1 ψ EPR pair EPR measurement π π θ π π θ Qutrit 2 0

Qutrit 3 0 π π π π π π

π θ π Qutrit 4 0 Scrambling Qutrit 5 0 π π π EPR pair b c

Figure 5. Results of the five-qutrit teleportation protocol — a, An expanded view of the five-qutrit teleportation protocol in Fig. 4, showing the native operations used to realize each portion of the protocol. b, Measured teleportation fidelities for twelve teleported states, which combine to give an unbiased estimate of the teleportation fidelity averaged over all single-qutrit pure states. An average fidelity above 1/2 – the classical limit for teleportation of qutrits – verifies non-zero quantum information scrambling by the maximally scrambling unitary, despite the presence of experimental error. When the scrambling unitary is replaced with an identity operation with same complexity as the scrambler, the average teleportation state fidelity reduces to 1/3, the same as a random guess when one does not have access to the input state. c, Representation of each of the twelve reconstructed density matrices after teleportation, expressed in the basis of Gell-Mann matrices (λ0 − λ8) with dotted lines showing the ideal result.

fidelities Fψ ≈ 1/3 for all input states (Fig. 5b). Indeed, and two-qutrit gates, as well as characterization methods using quantum state tomography we verify that the final such as state- and process-tomography that provide use- state of qubit Q5 is near maximally mixed regardless of ful information for benchmarking and debugging large- input state. scale computations. Using the verification of qutrit infor- Finally, we perform the teleportation protocol with mation scrambling as a proof-of-principle task, we have the maximally scrambling unitary Us, which in theory demonstrated two distinct entangling gates as well as should allow for perfect teleportation of any input quan- a simple dynamical-decoupling protocol that allows the tum state. In contrast to the identity operation, we gates to be run simultaneously on adjacent qutrits. In- observe that all but one of the input states are tele- terestingly, this experiment can be also interpreted as ported with fidelity Fψ > 1/2, with an average fidelity an implementation of quantum error-correction. In this of Favg = 0.568 ± 0.001. This allows us to experimen- language, the teleportation protocol is equivalent to a tally upper bound the averaged OTOC by 0.618 ± 0.004. three-qutrit quantum error-correcting code, which pro- This marks a significant decay from its non-scrambling tects information from the erasure of any one of the three value of unity and verifies that our platform is capable of qutrits [66]. Exploring other forms of error-correction in coherently simulating multi-qutrit scrambling dynamics. qudit-based processors is a natural extension of our work. A number of intriguing future directions are suggested by this work. First, our platform opens the door to IV. CONCLUSION exploring the potential advantages of ternary quantum logic, including a more efficient decomposition of the In summary, we have demonstrated a five-qutrit quan- Toffoli gate [27] as well as magic state distillation tum processor built from superconducting transmon cir- protocols that outperform conventional qubit-based cuits. Our qutrit toolbox features high fidelity single- strategies in terms of both error-threshold and yield [3]. 11

Second, the dynamical decoupling techniques introduced temperature for 8 minutes. We then etch the niobium here naturally apply to other spin-1 systems includ- with chlorine chemistry in an inductively coupled reac- ing solid-state defect centers and multi-level atomic, tive ion etcher with about 50 nm overetch into the sil- molecular and trapped ion systems [67, 68]. Third, icon. After etching, resist is removed with Microposit one can imagine further enlarging the qudit dimension 1165 at 80◦C for 60 min. The Josephson junction fab- by further leveraging the intrinsic anharmonicity of rication process begins by stripping native Nb and Si transmons [30, 32], enabling the study of many-body oxides with 30 seconds in BOE. Resist is then spun: we phases and entanglement dynamics in higher-spin use 500 nm of MMA EL-13 and 150 nm of AllResist AR- quantum systems [69]. Finally, building upon recent P 6200.9, both baked at 150◦C for 60 and 90 seconds, excitement surrounding quantum supremacy protocols respectively. We write ”Manhattan style” junction pat- using pseudorandom quantum circuit sampling [70], it terns [71, 72] (proximity-effect-corrected with Beamer by would be natural to investigate analogous qudit-based Genisys software) at 100 keV in a Raith EBPG 5150 us- protocols, where supremacy might be achieved using a ing a 200 pA beam current and 200 µm aperture. After substantially smaller number of elements. writing, the exposed AR-P is first developed in n-amyl acetate chilled to 0◦C; after this the AR-P development is halted with 10s immersion in IPA; finally MMA is developed in 3:1 IPA:MIBK for 10 min. We then dry the ACKNOWLEDGMENTS resulting structure with N2 and descum it with an 80W, 200 mbar oxygen plasma etch. This etching step is split Acknowledgements The authors gratefully acknowl- into 4 separate substeps, with 90 degree substrate rota- edge the conversations and insights of James Colless, tions between each substep for improved junction unifor- Emmanuel Flurin, Daniel Harlow, William Livingston, mity. Newly-formed oxides at the bottom of the devel- Leigh S. Martin, Brad Mitchell and Brian Swingle. oped structure are then removed with a 15s dip in BOE. This work was supported by multiple grants awarded The wafer is then loaded into a Plassys MEB550s evap- by the Department of Energy. From the Office of orator and pumped overnight before the junction evapo- Advanced Scientific Computing Research: the Advanced ration steps: first, an Al base electrode is evaporated and Quantum Testbed and the Quantum Algorithm Teams the tunnel barrier then formed by thermal oxidiation, in- Programs. From the Office of High Energy Physics: the troducing a 95%/5% Ar/O mix into the chamber at 10 HEP QUANTISED and the COMPHEP pilot “Probing mbar for 10 min. A second aluminum electrode is then information scrambling via quantum teleportation” (DE- evaporated to complete the junction and a third evapo- AC02-05CH11231). T.S. acknowledges support from ration is necessary to climb the second 250 nm capacitor the National Science Foundation Graduate Research step edge. The junction pattern includes a 6 x 8 µm Fellowship Program under Grant No. DGE 1752814. Al wire on top of the Nb for electrical contact between the junction and capacitor. After liftoff for 2 hours in acetone at 67◦C, the same resist stack is spun, and 10 x 15 µm rectangles are opened over the Al/Nb overlap Appendix A: Processor and fabrication details region. The exposed metals are then ion milled and Al is subsequently e-beam evaporated to ensure a low loss galvanic connection between Nb and Al [73]. More de- The processor we use features five fixed-frequency (sin- tails on junction fabrication, including the steps leading gle junction) transmon qutrits on a chip with an eight- to higher uniformity, can be found in [56]. After fabri- transmon ring geometry. The readout and coupling res- cation, the wafer is diced into 1x1 cm dies; cleaned in onators, Purcell filter, transmon capacitors, microwave Microposit 1165 for 12 hours at 80C; sonicated in DI drive lines and ground plane are composed of niobium, water, acetone, and IPA; descummed in 100 W oxygen while the transmon junctions are aluminum with an alu- plasma for 1 min and then wirebonded into a gold plated minum oxide barrier (Fig 1a main text). copper cryopackage on a 300 µm air gap. The processor is fabricated on intrinsic >8000 ohm-cm silicon <100> wafers. Initial cleaning of the silicon wafer Each transmon is coupled to (i) a linear readout res- occurs in piranha solution—a mixture of sulfuric acid and onator to enable multiplexed dispersive measurement, hydrogen peroxide—at 120◦C for 10 minutes, followed (ii) two coupling resonators to enable entangling interac- by 5:1 buffered oxide etch (BOE) for 30 seconds to re- tions with nearest neighbors, and (iii) a microwave drive move surface contaminants and native oxides. A 200-nm line. Readout resonators are loaded so that their effective thick film of niobium is then sputtered onto the wafer, linewidth κext ≈ 1 MHz. All readout resonators on the with deposition pressures optimized to yield a slightly chip are coupled to a common λ/2 resonator, a Purcell compressive film. Following this, junctions and all other filter with an external Q ≈ 10 [74]. The Purcell filter’s structures are patterned using 3 rounds of electron-beam passband overlaps with all readout resonator frequencies, lithography. We use MicroChem MMA EL-13 copolymer allowing fast readout; all qutrit frequencies lie outside as a resist, developing it in a 3:1 mixture of IPA:MIBK the passband, suppressing qutrit relaxation through this (isopropyl alcohol and methyl isobutyl ketone) at room channel. Slotline modes of all structures are suppressed 12 using wirebonds; a wirebond also enables the readout bus the Hamiltonian to overlap a coupling resonator. † † Hint/~ = g a b + b a (B2)

Appendix B: Experimental setup We measure the value of g on a chip with tunable transmons, but otherwise identical to the one used in the experiment. Spectroscopy of the avoided crossing reveals The processor is installed in the 10 mK stage of a Blue- an interaction amplitude g of roughly 3 MHz. Fors XLD dilution refrigerator. Room-temperature and cryogenic electronics for performing control and measurement of the qutrit chip are shown in Fig. 6. A Holzworth 2. Coherence of third transmon level multi-channel synthesizer generates three local oscillator tones: a qubit control LO at 4.72 GHz, a readout LO at 6.483 GHz, and a pump at 7.618 GHz for the The dominant noise processes affecting transmons tend traveling-wave parametric amplifier (TWPAs). Qutrit to worsen for states higher up the transmon ladder. For control pulses are formed by IQ modulating the ampli- our qutrit-based processor there are two salient manifes- fied qubit LO (split six ways) with IF signals from a tations of this: Tektronix AWG (sample rate 2.5 GS/s) with frequencies • Due to bosonic enhancement, amplitude damping between 0.5 and 1.1 GHz. We use both single-channel, (spontaneous emission) decay from state |2i to |1i hybrid-enabled SSB modulation and high-pass filtering proceeds roughly twice as fast as the decay from |1i to eliminate the lower sideband of the pulse, with addi- to |0i. tional band-pass filtering at room-temperature to eliminate noise from the AWG itself. Readout signals are • Dephasing due to charge noise, which randomizes generated using two-channel SSB modulation with IF sig- the relative phases between the |0i, |1i, and |2i nals from the same Tektronix AWG. All input signals are states, occurs roughly an order of magnitude faster further attenuated in the cryostat. for each state up the transmon ladder: in particu- Readout signals are amplified by the TWPA at 10 mK, lar, for the |2i state relative to the |1i state. high-electron mobility transistor amplifiers (HEMT) at 4K, and further amplification at room-temperature be- As stated in the main text, careful fabrication, microwave fore being digitized at 1.25 GSa/s and demodulated in engineering, and parameter selection were required to ob- software. tain high coherence in the transmon qutrit. The fabrication and microwave engineering are detailed in Appendix A, and served to mitigate the T1 decay. Here we describe 1. Chip characterization the parameter selection – specifically, the choice of the transmon EJ /EC ratio – which was chosen to minimize Transmon parameters are given in Table II. The the effect of dephasing. frequencies are extracted using standard spectroscopy Transmons are characterized by two parameters: the methods. Lifetimes are extracted by fitting decay curves Josephson energy EJ , and the capacitive energy EC [39]. to a single model with five parameters: two energy- Increasing the EJ /EC ratio exponentially decreases the relaxation times (T 1→0 and T 2→1) and a dephasing time sensitivity of all transmon eigenstates to charge noise, at 1 1 the expense of also lowering the transmon’s anharmonic- for each basis state. We perform randomized benchmark- th ing to measure pulse errors of single qubit operations in ity. Specifically, the charge dispersion m of the m level the different subspaces, shown in the table. is given by We further measure the coefficients of the cross-Kerr r m + 3 24m+5 2 E 2 4 √ m J − 8EJ /EC (or ‘ZZ’) interaction by performing a Ramsey measure- m ≈ (−1) EC e , ment with neighboring qutrits in the |1i or |2i states. m! π 2EC The cross-Kerr Hamiltonian between neighboring qutrits (B3) is while the relative anharmonicity αr is given by −1/2 αr ≈ −(8EJ /EC ) . (B4) HKerr/~ = α11|11ih11| + α12|12ih12| +α |21ih21| + α |22ih22| (B1) Typical transmon qubit designs use ratios EJ /EC ≈ 50. 21 22 We initially used such a ratio, which resulted in charge Table I in the main text gives the value of these coef- dispersion of 102 kHz and <10 kHz of |2i and |1i states, ficients. Residual cross-Kerr interaction coefficients be- respectively. However, with these parameters, charge- tween non-nearest-neighbor transmons were found to be parity fluctuations [98] dephase the coherence between negligible. the |2i and |1i states within 5 µs, making high-fidelity The cross-Kerr interaction is the dispersive limit of gates impossible to implement. To mitigate this dephas- an exchange interaction between nearest-neighbor trans- ing, we switched to a design with EJ /EC ≈ 73, which mons mediated by the coupling resonators, governed by resulted in charge dispersions of 12 kHz and 261 Hz for 13

300 K Control (in) Readout (out/in) ctrl DAC RO I Q AWG

ctrl switch LO I Q ctrl ctrl 1 RO 1 ctrl to to ctrl 8 6 Demod LO I Q ctrl SA

SA RF

H Demod Alazar (ADC) 4K 20 20 20 20 20 20 20 20 10 20

Cu I Q

LO 10 20 10 10 10 10 Still 10 10 10 10 SA 6 6 6 6 CP 6 6 6 6 3 3 3 3 3 3 3 3 20 MXC 10 KL KL KL 20 20 20 20 20 20 20 20 T

20 RF Cu RO Signal KL KL KL KL KL KL KL KL Cu

KL Attenuator ... 3 IQ mixer Hybrid 20 (20, 10, 6, 3 dB)

Filter Circulator DC Source (HP, LP)

MW Source Isolator SA Spectrum Analyzer

Amplifier (HEMT, TWPA) Bias T Directional Coupler

Figure 6. Experimental setup described in detail in the text.

Q1 Q2 Q3 Q4 Q5

Qutrit |0i ↔ |1i frequency, ω01/2π (GHz) 5.447 5.634 5.776 5.619 5.431 Qutrit |1i ↔ |2i frequency, ω12/2π (GHz) 5.177 5.368 5.512 5.351 5.160 Readout frequency, ωRO/2π (GHz) 6.384 6.324 6.731 6.673 6.618 |1i→|0i Lifetime T1 (µs) 70 49 43 55 63 |2i→|1i Lifetime T1 (µs) 38 29 39 32 36 ∗ Ramsey decay time T2 , |1i/|0i (µs) 73 13 41 48 20 ∗ Ramsey decay time T2 , |2i/|1i (µs) 13 10 16 23 10 ∗ Ramsey decay time T2 , |2i/|0i (µs) 16 6 15 26 11 Echo time T2Echo, |1i/|0i (µs) 71 51 46 64 74 Echo time T2Echo, |2i/|1i (µs) 29 22 22 35 32 Echo time T2Echo, |2i/|0i (µs) 39 26 34 45 39 Readout fidelity, |0i 0.99 0.99 0.97 0.98 0.99 Readout fidelity, |1i 0.97 0.95 0.94 0.95 0.96 Readout fidelity, |2i 0.95 0.94 0.92 0.95 0.96 Per-Clifford error, |1i/|0i subspace 3.6e-4 3.9e-4 5.5e-4 2.7e-4 - Per-Clifford error, |2i/|1i subspace 3.6e-4 6.0e-4 5.0e-4 7.5e-4 -

Table II. Measured properties of the five qutrits 14 the |2i and |1i states, respectively. This also reduced drive line: ~o(ω) = C(ω)~i(ω). We did observe a strong the anharmonicity from roughly 300 MHz to roughly 250 frequency-dependence of the cross-talk matrix. MHz. The linearity of the cross-talk enabled us to compensate for it by inverting the matrix C(ω) at each drive frequency, yielding combinations of microwave drive lines which would route the drive field to only a single qutrit. 3. Crosstalk The main challenge in this scheme was the measurement of C(ω). Our strategy was to focus on two drive lines at a time, and find for each line the relative amplitudes and As discussed in the main text, each transmon features a phases which exactly cancelled the field at the location of dedicated microwave control line through which we drive all of the qutrits on our chip—depending on the relative single- and two-qutrit gates. However, we find significant frequencies, we used either a Stark shift or a Rabi oscil- (order unity compared to intended coupling) crosstalk lation as a symptom of unwanted microwave field. This between the microwave drive lines for each qutrit. This measurement was repeated for each of ten drive frequen- crosstalk is non-local, not confined to nearest or next- cies of interest (i.e. the |0i ↔ |1i and |1i ↔ |2i transition nearest neighbors. When driving Rabi oscillations on a frequencies of all five qutrits), each pair of lines, and each given qutrit, it produces two unwanted effects: qutrit on the chip. Our crosstalk cancellation method is extremely 1. All other qutrits will be off-resonantly Rabi driven. measurement-intensive and was feasible only because of Depending on the relative frequencies between the the relatively few qutrits in this work. On future quan- qutrits, this can either manifest as an unwanted tum processors with tens or hundreds of quantum sys- change in a qutrit’s state populations (if the fre- tems, the number of measurements required for our can- quencies are relatively close) or a Stark shift (if the cellation scheme would be prohibitively expensive. In ad- frequency difference is large compared to the Rabi dition, the strong frequency dependence of the cross-talk frequency). matrix limits the speed at which one can apply single- qudit pulses in this manner: for pulses approximately 10 ns in length, we observed cross-talk which we could not 2. Microwave field leaking onto one or more neigh- compensate for using our method, likely because of this boring qutrits will result in an unwanted cross- frequency dependence combined with Fourier broadening resonance interaction, making the desired Rabi os- of the pulses. Going forward, it is thus important to pin- cillation frequency vary with the state of the neigh- point the source of the microwave cross-talk, in order to boring qutrit(s). This effect was anticipated in develop scalable solutions at the hardware level. [46].

We observed no indications of nonlinearity in the cross- Appendix C: Qutrit Operations and gate-set talk at the drive powers we used. That is, for a given drive frequency, the cross-talk can be characterized in 1. Single Qutrit Operations terms of a five-by-five complex-valued matrix C(ω) re- lating the field amplitudes ~o(ω) seen by each of the A convenient set of generators to describe qutrit rota- five qutrits to the input field amplitudes ~i(ω) on each tions are the Gell-Mann matrices:

0 1 0 0 −i 0 1 0 0 01 01 01 λ1 ≡ sx = 1 0 0 λ2 ≡ sy =  i 0 0 λ3 ≡ sz = 0 −1 0 0 0 0 0 0 0 0 0 0

0 0 1 0 0 −i 0 0 0 02 02 12 λ4 ≡ sx = 0 0 0 λ5 ≡ sy = 0 0 0 λ6 ≡ sx = 0 0 1 1 0 0 i 0 0 0 1 0

0 0 0 1 0 0 12 1 λ7 ≡ sy = 0 0 −i λ8 = √ 0 1 0 0 i 0 3 0 0 −2

They are the generators of the Lie algebra of the special unitary group SU(3) and can be thought of as the natural 15 extension of Pauli matrices (generators of the Lie algebra Three convenient gates to describe qudit-logic, which of the SU(2) group). For each qutrit (with basis states can be constructed from our universal gate-set, are the |0i , |1i , |2i), we calibrate a set of microwave pulses that X and Z gates resonantly drive the |0i ↔ |1i-transition and a separate set of pulses to address the |1i ↔ |2i-transition, providing universal control over the qubit subspaces {|0i , |1i} and X |ii = |i + 1 mod di (C5) {|1i , |2i}. Our microwave control pulses directly perform rotations that correspond to exponentiating Gell-Mann i 01 01 01 12 12 01 Z |ii = ω |ii , (C6) matrices sx , sy , sz , sx , sy . The Z-rotation (sz ) is implemented as a virtual Z gate in software by adjusting where ω = exp(i2π/d). the phases of subsequent microwave pulses in that sub- and the Hadamard gate space [84]. We extend this technique to the 12 subspace to also obtain the following rotation that is not one of 1 X H = √ ωij|iihj| (C7) the Gell-Mann matrices but that is very useful for single d qutrit control since it is a virtual rotation with negligible i,j error: 0 0 0 2. Two-Qutrit operations: Controlled-SUM Gate 12 sz = 0 1 0 (C1) 0 0 −1 In general d-dimensional qudits, two-qudit controlled- SUM and controlled-phase gates can be deﬁned using the In principle one could also drive the |0i ↔ |2i- Pauli X and Z gates We have: transition to directly implement rotations corresponding 02 02 d to sx , sy . While it would be worthwhile to add these ro- 01 X n tations to the available gate-set to compile circuits with UCSUM = |nihn| ⊗ X lower depth, these two-photon transitions are more chal- n=1 d lenging to address as they require high-power. Luckily, X n all rotations generated by the remaining Gell-Mann ma- UCφ = |nihn| ⊗ Z (C8) trices can be constructed from our available operations, n=1 for example Here the superscript 01 indicates that the controlled-

θ 02 π 12 θ 01 π 12 SUM is applied with Q0 as the control qudit and Q1 as −i 2 sx/y −i 2 sx −i 2 sx/y i 2 sx e = e · e · e (C2) the target; such a label is not necessary for UCφ, which is symmetric between the two qudits. The two gates are where the right most operator is the first to act on the equivalent up to a single-qudit Hadamard gate H on the state, so time goes from right to left. Similarly, λ8 can second qudit: 01 12 be constructed from sz and sz k † 01 We write a rotation in one of these subspaces as θj = (I ⊗ H )UCφ(I ⊗ H) = UCSUM, (C9) −i θ sk e 2 j , where k = {01, 12} defines the subspace of the rotation, j = {x, y, z} the rotation axis and θ the rotation where the qudit Hadamard gate is defined to transform the Z gate into the X gate under conjugation. Reversing angle. As an example, the two available x-rotations and † their corresponding rotation matrices are the order of H and H yields a controlled-MINUS gate, and changing which qubit receives the conjugation interchanges the control and target. The entangling gates cos θ/2 −i sin θ/2 0 01 10 01 10 01 UCφ, UCSUM, UCSUM, UCMIN, and UCMIN are therefore θx = −i sin θ/2 cos θ/2 0 , 0 0 1 all equivalent up to local (single-qudit) operations. 1 0 0 In our system, we directly implement the two-qutrit 12 0 cos θ/2 −i sin θ/2 θx = . (C3) UCφ gate by interspersing periods of evolution under the 0 −i sin θ/2 cos θ/2 cross-Kerr Hamiltonian (Eq. B1) with single-qutrit gates. This notation, combined with some useful qutrit and Intuitively, evolution under the cross-Kerr Hamiltonian experiment specific operations, is also adopted in circuit imparts phases to the two-qutrit states |11i, |12i, |21i, diagrams displayed in the main text and in this docu- and |22i, with values determined by the coefficients αij. ment. By interspersing this phase accumulation with single- Our gate-set consists of all Z-rotations along an arbi- qutrit pulses exchanging the various states, we can ensure trary angle, combined with the Clifford operations oper- that each state accumulates exactly the phase required ations in the 01 and 12 subspace : for the controlled-phase gate. We present two methods for implementing the π −π θk with θ = {π, −π, , }, controlled-phase gate in the manner described above. j 2 2 The first uses fewer single-qutrit pulses and is concep- where k = {01, 12} and j = {x, y, z}. (C4) tually simpler, but is not dynamically decoupled from 16 the cross-Kerr interaction with neighboring qutrits. The For specificity, we have parameterized this pulse se- second is dynamically decoupled and is the one used in quence with a single delay time, T ; an appropriately the teleportation experiment. chosen T realizes the controlled-phase gate. The delay First method: As depicted in Fig. 7, here we use time T is determined by the values of the cross-Kerr co- four periods of cross-Kerr evolution, separated by pulses efficients αij for each pair Q1/Q2 and Q3/Q4, and thus swapping the |1i and |2i states of a single qutrit. differs between the pairs; however, in practice, we find 12 Denoting this swap pulse as πq , where q is the qutrit that a delay of 192 ns works well for both. number, and evolution under the cross-Kerr Hamiltonian This pulse sequence constitutes a dynamically- for a time T as ZZT , the total pulse sequence is decoupled implementation of the UCφ unitary, as its operation is successful regardless of the states of the neigh- ZZ · π12 · ZZ · π12 · ZZ · π12 · ZZ · π12. TA 0 TB 1 TC 0 TD 1 boring qutrits. The dynamical decoupling arises because (C10) the single-qutrit pulses shuffle each qutrit’s states |0i, where the times TA, TB, TC , TD depend on the cross- |1i, and |2i such that an equal amount of time is spent in Kerr interaction parameters αij. For any choice of times, each state, regardless of the initial state of either qutrit. this operation imparts zero phase to the states |00i, |01i, This shuffling ‘averages out’ the cross-Kerr interaction |02i, |20i, |10i, and non-zero relative phases φ11, φ12, with neighboring qutrits, such that no entanglement is φ21, and φ22 to the other basis states. These phases are generated. linear combinations of the delay times T , T , T , and A B C The particular teleportation algorithm we implement T . The transformation from delay times to induced D requires applying U on two pairs of qutrits Q /Q and phases is full rank (except for pathological values of the Cφ 1 2 Q /Q . We use this dynamically decoupled pulse se- cross-Kerr coefficients), meaning that, given enough total 3 4 quence for both pairs, and apply the gates simultaneously delay time, this method can in principle generate an ar- to reduce decoherence associated with a longer total gate bitrary two-qudit phase gate (the states that receive zero time. Naively, the dynamical decoupling effect is weak- phase above can be made to gain arbitrary phase using ened by this simultaneity, since the ‘neighboring qutrits’, only single-qutrit phase gates). On our particular chip, with respect to each individual pair, are no longer static. the coefficients α allow us to implement the controlled- ij Fortunately, we verify both theoretically and empirically phase in this manner in roughly 1.5 µs for qutrit pairs that we can nevertheless decouple the unwanted inter- Q /Q and Q /Q . 1 2 3 4 action by reversing the order of the Π12 and Π01 gates The drawbacks of this method become apparent when between the two pairs (Fig. 8 b). one tries to use it in a multi-qutrit algorithm. If the two qutrits undergoing the controlled-phase are coupled to other qutrits via the same cross-Kerr Hamiltonian (as is the case for our chip), the above method will not Appendix D: Dynamically-decoupled EPR work when the other qutrits are in superpositions of ba- preparation sis states, in which case entanglement between them and the desired qutrits will be generated. The second method We prepare the two intial EPR pairs of the teleporta- addresses this problem. tion algorithm using the controlled-π gate as discussed Second method: As depicted in Fig. 8 a, this method in the main text. The basic sequence is presented in Fig. uses six equal time periods of cross-Kerr evolution. These 8, and serves to prepare an EPR pair on either Q2/Q3 are interspersed with single-qutrit pulses swapping the or Q4/Q5 individually, while all other qutrits are in the 12 01 ground state |0i. Simultaneous EPR pair preparation, as |0i/|1i and |1i/|2i subspaces, denoted πq and πq , respectively. The total pulse sequence consists of three required by the algorithm, necessitates a more compli- repetitions of: cated sequence that incorporates dynamical decoupling. This necessity is demonstrated by Fig. 9(a-b), which com- 12 12 01 01 ZZT · π0 ⊗ π1 · ZZT · π0 ⊗ π1 (C11) pares the result of individual EPR preparation to joint EPR preparation without dynamical decoupling. Joint preparation fidelities are much lower than those of indi- T T T T a b c d vidual preparation. From the measured density matrices, this loss seems to be largely due to a decrease in the off- Qutrit 0 0 π π diagonal elements (i.e. the coherences). Qutrit 1 0 π To understand the source of this decrease in coherence, π we measured the density matrix of the Q2/Q3 EPR pair while projecting the neighboring qutrit, Q4, into each of Figure 7. Four-segment pulse sequence implement- its basis states. Each of the three conditional density ma- ing UCφ. Four local π-pulses that effectively permute the trices we obtained was much purer (i.e. had much higher eigenstates are interspersed with periods of free evolution coherence) than the unconditional density matrix; how- TA,TB ,TC ,TD. The times depend on the cross-Kerr interac- ever, the phases of each coherence differed depending on tion parameters and are chosen such that the pulse sequence the state of Q4. These measurements suggest that the implements the desired diagonal unitary operation. source of the decoherence was indeed unwanted entan- 17

a T T T T T T

Qutrit 1 π π π π π π Qutrit 2 π π π π π π

Qutrit 3 π π π π π π Qutrit 4 π π π π π π

Figure 8. Six-segment pulse sequence implementing UCφ, dynamically-decoupled from static neighbors. (a) By alternating six periods of free evolution with local permutation pulses, a diagonal phase gate can be implemented while also protecting the two qutrits from dephasing and static interactions with neighbors provided that they are static. (b) To perform two of these gates in parallel and maintain the decoupling, in this case between qutrit 2 and qutrit 3, the order of local permutations is swapped between pairs: qutrit 1 and qutrit 2 first are permuted in the |0i/|1i subspace (blue) while the first local operation on qutrit 3 and 4 is in the |1i/|2i subspace (orange). This effectively decouples qutrit 2 and 3 in addition to performing the desired gates and decoupling qutrit 1 and 4 from their other neighbors (0 and 5).

glement between Q3 and Q4 arising from the cross-Kerr can be accomplished by splitting the idling time into interaction. three equal time periods, and applying single-qutrit X Qutrit State Tomography [38] after each step of the gates to one of the qutrits between each of the periods. EPR preparation sequence allows us to pinpoint the por- This shuffling of the populations decouples the entangling tions of the sequence that contribute most strongly to interaction into a product of local Z interactions. Using the unwanted entanglement. The cross-Kerr interaction the same principle, we divide the controlled-π operations affects the |2i states most strongly, and we find corre- in the relevant portion of simultaneous EPR preparation spondingly that most of the entanglement occurs after into three equal periods of 125 ns, and apply qutrit X the |2i state of Q3 gets populated. We take advantage gates on Q4 in between the waiting periods. These local of this by only dynamical decoupling the cross-Kerr in- gates on Q4 do not commute with the controlled-π op- teraction after this point. As shown in Fig√ 9, the initial eration on Q4/Q5. We therefore tune up a faster cross- preparation of Bell states (|00i + |11i)/ 2, which does resonance pulse that realizes the full controlled-π gate not involve the state |2i, is performed without dynamical between Q4 and Q5 within the first of the three 125 ns decoupling to reduce the error associated with additional periods. The decoupling pulses on Q4 can then be ap- single-qutrit gates. plied without interfering with the preceding entangling The mechanism underlying our decoupling sequence operation. This sequence enables simultaneous EPR pair is most easily understood by first considering a simpler preparation with fidelities 0.88 ± 0.002 and 0.92 ± 0.002 problem, of decoupling an unwanted cross-Kerr interac- on Q2/Q3 and Q4/Q5, respectively. tion between two qutrits during an idling period. This

[1] M. Luo and X. Wang, Universal quantum computation tography with 3-state systems, Physical Review Letters with qudits, SCIENCE CHINA Physics, Mechanics & As- 85, 3313 (2000). tronomy 57, 1712 (2014). [6] R. Lapkiewicz, P. Li, C. Schaeff, N. K. Langford, [2] E. T. Campbell, Enhanced fault-tolerant quantum com- S. Ramelow, M. Wie´sniak,and A. Zeilinger, Experimen- puting in d-level systems, Physical Review Letters 113, tal non-classicality of an indivisible quantum system, Na- 10.1103/physrevlett.113.230501 (2014). ture 474 (2011). [3] E. T. Campbell, H. Anwar, and D. E. Browne, Magic- [7] A. Fedorov, L. Steffen, M. Baur, M. P. d. Silva, and state distillation in all prime dimensions using quantum A. Wallraff, Implementation of a Toffoli gate with su- reed-muller codes, Physical Review X 2, 041021 (2012). perconducting circuits, Nature 481 (2012). [4] D. Bruss and C. Macchiavello, Optimal eavesdropping [8] P. Kurpiers, P. Magnard, T. Walter, B. Royer, M. Pechal, in cryptography with three-dimensional quantum states, J. Heinsoo, Y. Salathé, A. Akin, S. Storz, J.-C. Besse, Physical review letters 88, 127901 (2002). S. Gasparinetti, A. Blais, and A. Wallraff, Deterministic [5] H. Bechmann-Pasquinucci and A. Peres, Quantum cryp- quantum state transfer and remote entanglement using 18

a: Individual preparation

π θ π Qutrit 2 0 Qutrit 2 0

Qutrit 3 Qutrit 3 0 π π π 0 π Qutrit 4 0 Qutrit 4 θ π 0

Qutrit 5 0 Qutrit 5 0 π π π

b: Simultaneous preparation

250 ns 250 ns

π Qutrit 2 0 θ π Slow Cross-Resonance Slow Cross-Resonance Qutrit 3 0 π π Qutrit 4 0 θ π Slow Cross-Resonance Slow Cross-Resonance Qutrit 5 0 π

c: Simultaneous preparation with dynamical decoupling

250 ns 125 ns 125 ns 125 ns

π Qutrit 2 0 θ π Slow Cross-Resonance Slow Cross-Resonance Slow Cross-Resonance Qutrit 3 0 π Dynamically Decoupled π Qutrit 4 0 θ π Slow Cross-Resonance Fast Cross-Resonance Qutrit 5 0 π

Figure 9. Dynamically-decoupling the EPR pair preparation — a, Density matrices of the Q2/Q3 (left, purple) and Q4/Q5 (right, orange) EPR pairs, prepared individually (i.e. with all other qutrits in the ground state). State fidelities for this dataset are 0.94±0.002 and 0.98±0.002 respectively. b, Density matrices of the same EPR pairs when prepared simultaneously without any dynamical decoupling. Fidelites are markedly lower in this case, 0.81 ± 0.002 and 0.82 ± 0.002 respectively for the Q2/Q3 and Q4/Q5 pairs. As discussed in the text, the loss of fidelity is due to unwanted entanglement arising from the cross-Kerr interaction between the two EPR pairs. c, EPR pairs prepared simultaneously using dynamical decoupling, with fidelities 0.88 ± 0.002 and 0.92 ± 0.002 , respectively. 19

microwave photons, Nature 558 (2018). B. Yoshida, N. Y. Yao, and C. Monroe, Verified quantum [9] S. Rosenblum, P. Reinhold, M. Mirrahimi, L. Jiang, information scrambling, Nature 567, 61 (2019). L. Frunzio, and R. J. Schoelkopf, Fault-tolerant detec- [27] P. Gokhale, J. M. Baker, C. Duckering, N. C. tion of a quantum error, Science 361 (2018). Brown, K. R. Brown, and F. T. Chong, Asymp- [10] K. Dai, P. Zhao, M. Li, X. Tan, H. Yu, and Y. Yu, totic improvements to quantum circuits via qutrits, in Demonstration of quantum permutation parity deter- Proceedings of the 46th International Symposium on Computer Architecture, mine algorithm in a superconducting qutrit, Chinese ISCA ’19 (Association for Computing Machinery, New Physics B 27, 060305 (2018), publisher: IOP Publish- York, NY, USA, 2019) p. 554–566. ing. [28] S. S. Bullock, D. P. O’Leary, and G. K. Brennen, Asymp- [11] Y.-H. Luo, H.-S. Zhong, M. Erhard, X.-L. Wang, L.-C. totically Optimal Quantum Circuits for $d$-Level Sys- Peng, M. Krenn, X. Jiang, L. Li, N.-L. Liu, C.-Y. Lu, and tems, Physical Review Letters 94, 230502 (2005), pub- et al., Quantum teleportation in high dimensions, Physi- lisher: American Physical Society. cal Review Letters 123, 10.1103/physrevlett.123.070505 [29] C. Sayrin, I. Dotsenko, X. Zhou, B. Peaudecerf, (2019). T. Rybarczyk, S. Gleyzes, P. Rouchon, M. Mirrahimi, [12] P. Imany, J. A. Jaramillo-Villegas, M. S. Alshaykh, J. M. H. Amini, M. Brune, J.-M. Raimond, and S. Haroche, Lukens, O. D. Odele, A. J. Moore, D. E. Leaird, M. Qi, Real-time quantum feedback prepares and stabilizes pho- and A. M. Weiner, High-dimensional optical quantum ton number states, Nature 477 (2011). logic in large operational spaces, npj Quantum Informa- [30] R. K. Naik, N. Leung, S. Chakram, P. Groszkowski, tion 5, 1 (2019). Y. Lu, N. Earnest, D. C. McKay, J. Koch, and D. I. [13] S. H. Shenker and D. Stanford, Black holes and the Schuster, Random access quantum information proces- butterfly effect, Journal of High Energy Physics 2014, sors using multimode circuit quantum electrodynamics, 10.1007/jhep03(2014)067 (2014). Nature Communications 8, 1 (2017). [14] P. Hayden and J. Preskill, Black holes as mirrors: quan- [31] N. Ofek, A. Petrenko, R. Heeres, P. Reinhold, Z. Leghtas, tum information in random subsystems, Journal of High B. Vlastakis, Y. Liu, L. Frunzio, S. M. Girvin, L. Jiang, Energy Physics 2007, 120 (2007). M. Mirrahimi, M. H. Devoret, and R. J. Schoelkopf, Ex- [15] Y. Sekino and L. Susskind, Fast scramblers, Journal of tending the lifetime of a quantum bit with error correc- High Energy Physics 2008, 065 (2008). tion in superconducting circuits, Nature 536, 441 (2016). [16] A. Larkin and Y. N. Ovchinnikov, Quasiclassical method [32] Y. Y. Gao, B. J. Lester, K. S. Chou, L. Frunzio, M. H. in the theory of superconductivity, Sov Phys JETP 28, Devoret, L. Jiang, S. M. Girvin, and R. J. Schoelkopf, 1200 (1969). Entanglement of bosonic modes through an engineered [17] D. A. Roberts, D. Stanford, and L. Susskind, Local- exchange interaction, Nature 566, 509 (2019). ized shocks, Journal of High Energy Physics 2015, [33] T. Bækkegaard, L. B. Kristensen, N. J. S. Loft, C. K. 10.1007/jhep03(2015)051 (2015). Andersen, D. Petrosyan, and N. T. Zinner, Realization [18] J. Maldacena, S. H. Shenker, and D. Stanford, A of efficient quantum gates with a superconducting qubit- bound on chaos, Journal of High Energy Physics 2016, qutrit circuit, Scientific Reports 9, 13389 (2019), number: 10.1007/jhep08(2016)106 (2016). 1 Publisher: Nature Publishing Group. [19] P. Hosur, X.-L. Qi, D. A. Roberts, and B. Yoshida, Chaos [34] B. P. Lanyon, T. J. Weinhold, N. K. Langford, J. L. in quantum channels, Journal of High Energy Physics O’Brien, K. J. Resch, A. Gilchrist, and A. G. White, Ma- 2016, 10.1007/jhep02(2016)004 (2016). nipulating Biphotonic Qutrits, Physical Review Letters [20] M. Gärttner,J. G. Bohnet, A. Safavi-Naini, M. L. Wall, 100 (2008). J. J. Bollinger, and A. M. Rey, Measuring out-of-time- [35] F. Dolde, V. Bergholm, Y. Wang, I. Jakobi, B. Nayde- order correlations and multiple quantum spectra in a nov, S. Pezzagna, J. Meijer, F. Jelezko, P. Neumann, trapped-ion quantum magnet, Nature Physics 13, 781 T. Schulte-Herbrüggen,J. Biamonte, and J. Wrachtrup, (2017). High-fidelity spin entanglement using optimal control, [21] J. Li, R. Fan, H. Wang, B. Ye, B. Zeng, H. Zhai, X. Peng, Nature Communications 5 (2014). and J. Du, Measuring out-of-time-order correlators on [36] L. Sadler, J. Higbie, S. Leslie, M. Vengalattore, and a nuclear magnetic resonance quantum simulator, Phys. D. Stamper-Kurn, Spontaneous symmetry breaking in a Rev. X 7, 031011 (2017). quenched ferromagnetic spinor bose–einstein condensate, [22] K. X. Wei, C. Ramanathan, and P. Cappellaro, Exploring Nature 443, 312 (2006). localization in nuclear spin chains, Phys. Rev. Lett. 120, [37] C. Senko, P. Richerme, J. Smith, A. Lee, I. Cohen, 070501 (2018). A. Retzker, and C. Monroe, Realization of a Quantum [23] S. Lepoutre, J. Schachenmayer, L. Gabardos, B. Zhu, Integer-Spin Chain with Controllable Interactions, Phys- B. Naylor, E. Maréchal, O. Gorceix, A. M. Rey, ical Review X 5 (2015). L. Vernac, and B. Laburthe-Tolra, Out-of-equilibrium [38] R. Bianchetti, S. Filipp, M. Baur, J. M. Fink, C. Lang, quantum magnetism and thermalization in a spin-3 L. Steffen, M. Boissonneault, A. Blais, and A. Wallraff, many-body dipolar lattice system, Nature Communica- Control and tomography of a three level superconducting tions 10, 10.1038/s41467-019-09699-5 (2019). artificial atom, Phys. Rev. Lett. 105, 223601 (2010). [24] B. Yoshida and A. Kitaev, Efficient decoding [39] J. Koch, T. M. Yu, J. Gambetta, A. A. Houck, D. I. for the hayden-preskill protocol, arXiv preprint Schuster, J. Majer, A. Blais, M. H. Devoret, S. M. Girvin, arXiv:1710.03363 (2017). and R. J. Schoelkopf, Charge-insensitive qubit design de- [25] B. Yoshida and N. Y. Yao, Disentangling scrambling and rived from the cooper pair box, Physical Review A 76, decoherence via quantum teleportation, Physical Review 10.1103/physreva.76.042319 (2007). X 9, 10.1103/physrevx.9.011006 (2019). [40] J. A. Schreier, A. A. Houck, J. Koch, D. I. Schuster, B. R. [26] K. A. Landsman, C. Figgatt, T. Schuster, N. M. Linke, Johnson, J. M. Chow, J. M. Gambetta, J. Majer, L. Frun- 20

zio, M. H. Devoret, and et al., Suppressing charge noise [53] E. Magesan and J. M. Gambetta, Effective hamiltonian decoherence in superconducting charge qubits, Physical models of the cross-resonance gate, Physical Review A Review B 77, 10.1103/physrevb.77.180502 (2008). 101, 10.1103/physreva.101.052308 (2020). [41] In principle, one would also have to consider the tran- [54] Y.-M. Di and H.-R. Wei, Synthesis of multivalued quan- sition from |2i → |0i, but this channel is largely sup- tum logic circuits by elementary gates, Phys. Rev. A 87, pressed, due to parity, in transmons. 012325 (2013). [42] M. J. Peterer, S. J. Bader, X. Jin, F. Yan, A. Kamal, T. J. [55] F. F. Fanchini, J. E. M. Hornos, and R. d. J. Napolitano, Gudmundsen, P. J. Leek, T. P. Orlando, W. D. Oliver, Continuously decoupling single-qubit operations from a and S. Gustavsson, Coherence and decay of higher energy perturbing thermal bath of scalar bosons, Physical Re- levels of a superconducting transmon qubit, Phys. Rev. view A 75, 022329 (2007), publisher: American Physical Lett. 114, 010501 (2015). Society. [43] R. Barends, J. Kelly, A. Megrant, A. Veitia, D. Sank, [56] J. M. Kreikebaum, K. P. O’Brien, A. Morvan, and E. Jeffrey, T. C. White, J. Mutus, A. G. Fowler, I. Siddiqi, Improving wafer-scale josephson junction re- B. Campbell, and et al., Superconducting quantum cir- sistance variation in superconducting quantum coher- cuits at the surface code threshold for fault tolerance, ent circuits, Superconductor Science and Technology 33, Nature 508, 500–503 (2014). 06LT02 (2020). [44] A. Kandala, K. Temme, A. D. Córcoles, A. Mezzacapo, [57] Y. Gu, X.-L. Qi, and D. Stanford, Local criticality, diffu- J. M. Chow, and J. M. Gambetta, Error mitigation ex- sion and chaos in generalized sachdev-ye-kitaev models, tends the computational reach of a noisy quantum pro- Journal of High Energy Physics 2017, 125 (2017). cessor, Nature 567, 491–495 (2019). [58] A. Nahum, J. Ruhman, S. Vijay, and J. Haah, Quantum [45] Y. Salathé, M. Mondal, M. Oppliger, J. Heinsoo, entanglement growth under random unitary dynamics, P. Kurpiers, A. Potoˇcnik,A. Mezzacapo, U. Las Heras, Physical Review X 7, 031016 (2017). L. Lamata, E. Solano, and et al., Digital quantum sim- [59] Q. Zhuang, T. Schuster, B. Yoshida, and N. Y. Yao, ulation of spin models with circuit quantum electrody- Scrambling and complexity in phase space, Physical Re- namics, Physical Review X 5, 10.1103/physrevx.5.021027 view A 99, 10.1103/physreva.99.062334 (2019). (2015). [60] P. Hayden and J. Preskill, Black holes as mirrors: quan- [46] C. Rigetti and M. Devoret, Fully microwave-tunable uni- tum information in random subsystems, Journal of High versal gates in superconducting qubits with linear cou- Energy Physics 2007, 120 (2007). plings and fixed transition frequencies, Phys. Rev. B 81, [61] P. Gao, D. L. Jafferis, and A. C. Wall, Traversable worm- 134507 (2010). holes via a double trace deformation, Journal of High [47] T. Proctor, K. Rudinger, K. Young, M. Sarovar, and Energy Physics 2017, 10.1007/jhep12(2017)151 (2017). R. Blume-Kohout, What randomized benchmarking ac- [62] J. Maldacena, D. Stanford, and Z. Yang, Diving tually measures, Physical review letters 119, 130502 into traversable wormholes, Fortschritte der Physik 65, (2017). 1700034 (2017). [48] J. M. Chow, A. D. Corcoles, J. M. Gambetta, C. Rigetti, [63] A. R. Brown, H. Gharibyan, S. Leichenauer, H. W. B. R. Johnson, J. A. Smolin, J. R. Rozen, G. A. Lin, S. Nezami, G. Salton, L. Susskind, B. Swingle, and Keefe, M. B. Rothwell, M. B. Ketchen, and et al., M. Walter, Quantum gravity in the lab: Teleportation by Simple all-microwave entangling gate for fixed-frequency size and traversable wormholes (2019), arXiv:1911.06314 superconducting qubits, Physical Review Letters 107, [quant-ph]. 10.1103/physrevlett.107.080502 (2011). [64] T. Schuster, B. Kobrin, P. Gao, I. Cong, E. Khabi- [49] S. Sheldon, E. Magesan, J. M. Chow, and J. M. Gam- boulline, N. Linke, M. Lukin, C. Monroe, B. Yoshida, betta, Procedure for systematically tuning up cross- and N. Y. Yao, Teleportation via operator spreading in talk in the cross-resonance gate, Physical Review A 93, the traversable wormhole protocol, Journal of High En- 10.1103/physreva.93.060302 (2016). ergy Physics 2017, 125 (2017). [50] A. Kandala, A. Mezzacapo, K. Temme, M. Takita, [65] D. A. Roberts and B. Yoshida, Chaos and complex- M. Brink, J. M. Chow, and J. M. Gambetta, Hardware- ity by design, Journal of High Energy Physics 2017, efficient variational quantum eigensolver for small 10.1007/jhep04(2017)121 (2017). molecules and quantum magnets, Nature 549, 242–246 [66] R. Cleve, D. Gottesman, and H.-K. Lo, How to share a (2017). quantum secret, Physical Review Letters 83, 648 (1999). [51] P. Jurcevic, A. Javadi-Abhari, L. S. Bishop, I. Lauer, [67] C. Senko, P. Richerme, J. Smith, A. Lee, I. Cohen, D. F. Bogorin, M. Brink, L. Capelluto, O. Günlük, A. Retzker, and C. Monroe, Realization of a quantum T. Itoko, N. Kanazawa, A. Kandala, G. A. Keefe, integer-spin chain with controllable interactions, Physi- K. Kruslich, W. Landers, E. P. Lewandowski, D. T. cal Review X 5, 021026 (2015). McClure, G. Nannicini, A. Narasgond, H. M. Nayfeh, [68] S. Choi, N. Y. Yao, and M. D. Lukin, Dynamical En- E. Pritchett, M. B. Rothwell, S. Srinivasan, N. Sundare- gineering of Interactions in Qudit Ensembles, Physical san, C. Wang, K. X. Wei, C. J. Wood, J.-B. Yau, E. J. Review Letters 119, 183603 (2017). Zhang, O. E. Dial, J. M. Chow, and J. M. Gambetta, [69] S. Pai, M. Pretko, and R. M. Nandkishore, Localization Demonstration of quantum volume 64 on a superconduct- in fractonic random circuits, Physical Review X 9, 021003 ing quantum computing system (2020), arXiv:2008.08571 (2019). [quant-ph]. [70] F. Arute, K. Arya, R. Babbush, D. Bacon, J. C. Bardin, [52] V. Tripathi, M. Khezri, and A. N. Korotkov, Opera- R. Barends, R. Biswas, S. Boixo, F. G. Brandao, D. A. tion and intrinsic error budget of a two-qubit cross- Buell, et al., Quantum supremacy using a programmable resonance gate, Physical Review A 100, 10.1103/phys- superconducting processor, Nature 574, 505 (2019). reva.100.012301 (2019). [71] M. V. Costache, G. Bridoux, I. Neumann, and S. O. 21

Valenzuela, Lateral metallic devices made by a multi- [85] C. Macklin, K. O’Brien, D. Hover, M. E. Schwartz, angle shadow evaporation technique, Journal of Vac- V. Bolkhovsky, X. Zhang, W. D. Oliver, and I. Siddiqi, uum Science & Technology B 30, 04E105 (2012), A near–quantum-limited josephson traveling-wave para- https://doi.org/10.1116/1.4722982. metric amplifier, Science 10.1126/science.aaa8525 (2015). [72] A. Potts, G. J. Parker, J. J. Baumberg, and P. A. J. de [86] E. Knill, D. Leibfried, R. Reichle, J. Britton, R. B. Groot, Cmos compatible fabrication methods for submi- Blakestad, J. D. Jost, C. Langer, R. Ozeri, S. Sei- cron josephson junction qubits, IEE Proceedings - Sci- delin, and D. J. Wineland, Randomized benchmarking ence, Measurement and Technology 148, 225 (2001). of quantum gates, Physical Review A 77, 10.1103/phys- [73] A. Dunsworth, A. Megrant, C. Quintana, Z. Chen, reva.77.012307 (2008). R. Barends, B. Burkett, B. Foxen, Y. Chen, B. Chiaro, [87] M. Blake, R. A. Davison, and S. Sachdev, Thermal diffu- A. Fowler, and et al., Characterization and reduction of sivity and chaos in metals without quasiparticles, Physi- capacitive loss induced by sub-micron josephson junction cal Review D 96, 10.1103/physrevd.96.106008 (2017). fabrication in superconducting qubits, Applied Physics [88] D. Ben-Zion and J. McGreevy, Strange metal from lo- Letters 111, 022601 (2017). cal quantum chaos, Physical Review B 97, 10.1103/phys- [74] E. A. Sete, J. M. Martinis, and A. N. Korotkov, Quan- revb.97.155117 (2018). tum theory of a bandpass purcell filter for qubit readout, [89] J. Maldacena, The large n limit of superconformal field Phys. Rev. A 92, 012325 (2015). theories and supergravity, International Journal of The- [75] F. Motzoi, J. M. Gambetta, P. Rebentrost, and F. K. oretical Physics 38, 1113 (1999). Wilhelm, Simple pulses for elimination of leakage in [90] M. Kardar, Statistical Physics of Particles (Cambridge weakly nonlinear qubits, Phys. Rev. Lett. 103, 110501 University Press, 2007). (2009). [91] A. Vaziri, G. Weihs, and A. Zeilinger, Experimental [76] T. Bækkegaard, L. B. Kristensen, N. J. S. Loft, C. K. two-photon, three-dimensional entanglement for quan- Andersen, D. Petrosyan, and N. T. Zinner, Realization tum communication, Physical Review Letters 89, 240401 of efficient quantum gates with a superconducting qubit- (2002). qutrit circuit, Scientific Reports 9, 10.1038/s41598-019- [92] F. Arute, K. Arya, R. Babbush, D. Bacon, J. C. Bardin, 49657-1 (2019). R. Barends, R. Biswas, S. Boixo, F. G. S. L. Brandao, [77] M. Fujiwara, M. Takeoka, J. Mizuno, and M. Sasaki, Ex- D. A. Buell, and et al., Quantum supremacy using a pro- ceeding the classical capacity limit in a quantum optical grammable superconducting processor, Nature 574, 505 channel, Phys. Rev. Lett. 90, 167906 (2003). (2019). [78] A. Kulikov, M. Jerger, A. Potoˇcnik, A. Wallraff, and [93] R. Blume-Kohout, J. K. Gamble, E. Nielsen, A. Fedorov, Realization of a quantum random gener- K. Rudinger, J. Mizrahi, K. Fortier, and P. Maunz, ator certified with the kochen-specker theorem, Physi- Demonstration of qubit operations below a rigorous fault cal Review Letters 119, 10.1103/physrevlett.119.240501 tolerance threshold with gate set tomography, Nature (2017). Communications 8, 14485 (2017). [79] K. Dai, P. Zhao, M. Li, X. Tan, H. Yu, and Y. Yu, [94] E. A. Sete, J. M. Martinis, and A. N. Korotkov, Quan- Demonstration of quantum permutation parity deter- tum theory of a bandpass purcell filter for qubit read- mine algorithm in a superconducting qutrit, Chinese out, Physical Review A 92, 10.1103/physreva.92.012325 Physics B 27, 060305 (2018). (2015). [80] H. Blöte,J. L. Cardy, and M. Nightingale, Conformal [95] X.-M. Hu, C. Zhang, B.-H. Liu, Y. Cai, X.-J. Ye, Y. Guo, invariance, the central charge, and universal finite-size W.-B. Xing, C.-X. Huang, Y.-F. Huang, C.-F. Li, and G.- amplitudes at criticality, Physical review letters 56, 742 C. Guo, Experimental high-dimensional quantum tele- (1986). portation, Phys. Rev. Lett. 125, 230501 (2020). [81] I. Affleck, T. Kennedy, E. H. Lieb, and [96] S. Kirchhoff, T. Kebler, P. J. Liebermann, E. Asse- H. Tasaki, Rigorous results on valence- mat, S. Machnes, F. Motzoi, and F. K. Wilhelm, Op- bond ground states in antiferromagnets, in timized cross-resonance gate for coupled transmon sys- Condensed Matter Physics and Exactly Soluble Models tems, Physical Review A 97, 10.1103/physreva.97.042348 (Springer, 2004) pp. 249–252. (2018). [82] F. D. M. Haldane, Nonlinear field theory of large-spin [97] J. L. Allen, R. Kosut, and E. Ginossar, Minimal heisenberg antiferromagnets: semiclassically quantized time robust two qubit gates in circuit qed (2019), solitons of the one-dimensional easy-axis néelstate, Phys- arXiv:1902.08056 [quant-ph]. ical Review Letters 50, 1153 (1983). [98] D. Riste, C. C. Bultink, M. J. Tiggelman, R. N. Schouten, [83] P. Gao and H. Liu, Regenesis and quantum traversable K. W. Lehnert, and L. DiCarlo, Millisecond charge- wormholes, Journal of High Energy Physics 2019, parity fluctuations and induced decoherence in a super- 10.1007/jhep10(2019)048 (2019). conducting transmon qubit, Nature Communications 4, [84] D. C. McKay, C. J. Wood, S. Sheldon, J. M. Chow, and 10.1038/ncomms2936 (2013). J. M. Gambetta, Efficient z gates for quantum comput- [99] J. Preskill, Quantum Computing in the NISQ era and ing, Physical Review A 96, 10.1103/physreva.96.022330 beyond, Quantum 2, 79 (2018). (2017).