ThreeQubit Randomized Benchmarking - arXiv · ThreeQubit Randomized Benchmarking ... the gates we...
6
arXiv:1712.06550v1 [quant-ph] 18 Dec 2017 Three Qubit Randomized Benchmarking David C. McKay, * Sarah Sheldon, John A. Smolin, Jerry M. Chow, and Jay M. Gambetta IBM T.J. Watson Research Center, Yorktown Heights, NY 10598, USA (Dated: December 19, 2017) As quantum circuits increase in size, it is critical to establish scalable multiqubit fidelity metrics. Here we investigate three-qubit randomized benchmarking (RB) with fixed-frequency transmon qubits coupled to a common bus with pairwise microwave-activated interactions (cross-resonance). We measure, for the first time, a three-qubit error per Clifford of 0.106 for all-to-all gate connectivity and 0.207 for linear gate connectivity. Furthermore, by introducing mixed dimensionality simulta- neous RB — simultaneous one- and two-qubit RB — we show that the three-qubit errors can be predicted from the one- and two-qubit errors. However, by introducing certain coherent errors to the gates we can increase the three-qubit error to 0.302, an increase that is not predicted by a proportionate increase in the one- and two-qubit errors from simultaneous RB. This demonstrates three-qubit RB as a unique multiqubit metric. Quantum circuits are being built with an increasingly larger number of qubits and, accordingly, the problem of characterization is becoming more acute. The funda- mental reason for quantum speedups — the exponential growth of the state space with the number of qubits — means that tomographic methods for reconstructing the system will require exponential resources. Indeed, the number of required measurements for quantum process tomography scales as 16 n [1] where n is the number of qubits. For a 3 qubit system this is already 4096 ex- periments. Therefore, to avoid scaling issues, methods have focused on characterizing the primitive set of gates used to construct the universal gateset. At minimum, for n qubits, this set contains several one-qubit gates for all n qubits and n − 1 two-qubit gates [2]. Unless there is a native interaction for greater than two qubits, mul- tiqubit gates are built from sequences of one- and two- qubit gates. This is the case in many quantum computing technologies, such as superconducting qubits, where the interactions are local [3]. But how good is the assumption that multiqubit algorithmic fidelities will be predicted by the fidelities of the gate primitives measured in isolation? There are strong indications that this assumption fails due to crosstalk and addressability errors. For example, one of the main objectives for multiqubit circuits is to run surface code algorithms to construct logical fault-tolerant qubits. This entails constructing local five-qubit gates via sequential application of two-qubit CNOT gates in par- allel across a circuit with many qubits. Surface codes are predicted to have a high threshold for correcting errors, but they are typically simulated with correlated noise only between qubits for which there is a direct gate [4]. This does not take into account errors to spectator (i.e., non-participating neighboring) qubits and between pairs of qubits during parallel gate operations. In a recent five-qubit test of a logical qubit, the logical qubit fidelity was greatly improved by compensating for ZZ terms to spectator qubits during the two-qubit gate [5]. In addi- tion, several studies have observed that algorithmic and primitive gate fidelity do not always agree. For example when four algorithms were run on two different five-qubit processors there was no definitive agreement from prim- itive to algorithmic fidelity [6]. In a five-qubit device with measured two-qubit gate fidelities of 0.99, the state fidelity of a five-qubit GHZ state was 0.82 after apply- ing four two-qubit gates [7]. Therefore, to predict the true algorithmic fidelity we need to measure multiqubit fidelity metrics. Fortunately, the issue of scaling can be circumvented if the goal is to characterize a process based on a few measures, e.g., average gate fidelity. Based on this idea, there have been several proposed techniques for efficiently characterizing large circuits, such as Monte Carlo sam- pling [8, 9], compressed sensing [10], matrix product state tomography [11], and twirling protocols [12]. These tech- niques have been applied to measure the fidelity of a 7- qubit gate in an NMR system [13], to reconstruct a code state in a 7-qubit ion trap system [14] and to character- ize a 14-qubit ion trap simulator [15]. However, a com- mon drawback to these techniques is that the result is sensitive to preparation and measurement errors, some- times exponentially so. In addition, Refs [13, 14] char- acterized the state of a multiqubit system, but not the underlying gates. These problems are addressed by ran- domized benchmarking [16, 17] (RB), in which sequences of random Clifford gates are composed such that the uni- tary operator obtained by applying the entire sequence of gates is equal to the identity. The probability of a qubit returning to the state it started in versus sequence length decays exponentially. The decay constant is a sim- ple measure of the average fidelity of the Clifford set in- dependent of preparation and measurement errors. RB is a method widely used to characterize gates in supercon- ducting circuits [7, 18–20], ion-traps [16, 21–23], neutral- atom-traps [24], NMR systems [25] and for solid-state spin qubits [26]. Extensions to RB have been proposed and implemented to measure specific gate error via in- terleaving [27], purity [28, 29] and leakage [30, 31]. RB is, by construction, designed to address fidelities in multiqubit systems in two ways. For one, RB over
Transcript of ThreeQubit Randomized Benchmarking - arXiv · ThreeQubit Randomized Benchmarking ... the gates we...
arX
iv1
712
0655
0v1
[qu
ant-
ph]
18
Dec
201
7
Three Qubit Randomized Benchmarking
David C McKaylowast Sarah Sheldon John A Smolin Jerry M Chow and Jay M GambettaIBM TJ Watson Research Center Yorktown Heights NY 10598 USA
(Dated December 19 2017)
As quantum circuits increase in size it is critical to establish scalable multiqubit fidelity metricsHere we investigate three-qubit randomized benchmarking (RB) with fixed-frequency transmonqubits coupled to a common bus with pairwise microwave-activated interactions (cross-resonance)We measure for the first time a three-qubit error per Clifford of 0106 for all-to-all gate connectivityand 0207 for linear gate connectivity Furthermore by introducing mixed dimensionality simulta-neous RB mdash simultaneous one- and two-qubit RB mdash we show that the three-qubit errors can bepredicted from the one- and two-qubit errors However by introducing certain coherent errors tothe gates we can increase the three-qubit error to 0302 an increase that is not predicted by aproportionate increase in the one- and two-qubit errors from simultaneous RB This demonstratesthree-qubit RB as a unique multiqubit metric
Quantum circuits are being built with an increasinglylarger number of qubits and accordingly the problemof characterization is becoming more acute The funda-mental reason for quantum speedups mdash the exponentialgrowth of the state space with the number of qubits mdashmeans that tomographic methods for reconstructing thesystem will require exponential resources Indeed thenumber of required measurements for quantum processtomography scales as 16n [1] where n is the number ofqubits For a 3 qubit system this is already 4096 ex-periments Therefore to avoid scaling issues methodshave focused on characterizing the primitive set of gatesused to construct the universal gateset At minimumfor n qubits this set contains several one-qubit gates forall n qubits and n minus 1 two-qubit gates [2] Unless thereis a native interaction for greater than two qubits mul-tiqubit gates are built from sequences of one- and two-qubit gates This is the case in many quantum computingtechnologies such as superconducting qubits where theinteractions are local [3] But how good is the assumptionthat multiqubit algorithmic fidelities will be predicted bythe fidelities of the gate primitives measured in isolationThere are strong indications that this assumption failsdue to crosstalk and addressability errors For exampleone of the main objectives for multiqubit circuits is to runsurface code algorithms to construct logical fault-tolerantqubits This entails constructing local five-qubit gates viasequential application of two-qubit CNOT gates in par-allel across a circuit with many qubits Surface codes arepredicted to have a high threshold for correcting errorsbut they are typically simulated with correlated noiseonly between qubits for which there is a direct gate [4]This does not take into account errors to spectator (ienon-participating neighboring) qubits and between pairsof qubits during parallel gate operations In a recentfive-qubit test of a logical qubit the logical qubit fidelitywas greatly improved by compensating for ZZ terms tospectator qubits during the two-qubit gate [5] In addi-tion several studies have observed that algorithmic andprimitive gate fidelity do not always agree For example
when four algorithms were run on two different five-qubitprocessors there was no definitive agreement from prim-itive to algorithmic fidelity [6] In a five-qubit devicewith measured two-qubit gate fidelities of 099 the statefidelity of a five-qubit GHZ state was 082 after apply-ing four two-qubit gates [7] Therefore to predict thetrue algorithmic fidelity we need to measure multiqubitfidelity metrics
Fortunately the issue of scaling can be circumventedif the goal is to characterize a process based on a fewmeasures eg average gate fidelity Based on this ideathere have been several proposed techniques for efficientlycharacterizing large circuits such as Monte Carlo sam-pling [8 9] compressed sensing [10] matrix product statetomography [11] and twirling protocols [12] These tech-niques have been applied to measure the fidelity of a 7-qubit gate in an NMR system [13] to reconstruct a codestate in a 7-qubit ion trap system [14] and to character-ize a 14-qubit ion trap simulator [15] However a com-mon drawback to these techniques is that the result issensitive to preparation and measurement errors some-times exponentially so In addition Refs [13 14] char-acterized the state of a multiqubit system but not theunderlying gates These problems are addressed by ran-domized benchmarking [16 17] (RB) in which sequencesof random Clifford gates are composed such that the uni-tary operator obtained by applying the entire sequenceof gates is equal to the identity The probability of aqubit returning to the state it started in versus sequencelength decays exponentially The decay constant is a sim-ple measure of the average fidelity of the Clifford set in-dependent of preparation and measurement errors RB isa method widely used to characterize gates in supercon-ducting circuits [7 18ndash20] ion-traps [16 21ndash23] neutral-atom-traps [24] NMR systems [25] and for solid-statespin qubits [26] Extensions to RB have been proposedand implemented to measure specific gate error via in-terleaving [27] purity [28 29] and leakage [30 31]
RB is by construction designed to address fidelitiesin multiqubit systems in two ways For one RB over
2
the full n-qubit space can be performed by construct-ing sequences from the n-qubit Clifford group Addi-tionally the n-qubit space can be subdivided into setsof qubits ni and ni-qubit RB performed in each sub-set simultaneously [32] Both methods give metrics offidelity in the n-qubit space Despite the availability ofthese two methods there has been no demonstration ofRB with n gt 2 since it is viewed as sufficient to charac-terize only the primitive gateset Here we show for thefirst time a variety of three-qubit RB combinations in athree-qubit fixed-frequency superconducting device Forall-to-all gate connectivity we measure a three-qubit errorper Clifford (3Q EPC) of 0106 which is well-predictedby the primitive gate errors from two-qubitone-qubit si-multaneous RB However we find a strong dependenceon calibration procedure that is not apparent a prioriFor one such calibration the error increases to 0302Importantly this increase in error is not predicted bya commensurate increase in the primitive gate errors asmeasured from simultaneous RB We also show the im-portance of connectivity in devices as the 3Q EPC in-creases to 0207 when we limit the device to have lineargate connectivity
Before describing our experiment in detail we first pro-vide a brief summary of the RB method a detailed dis-cussion of RB can be found in Ref [33] The main idea isto construct a m-length sequence of random n-qubit Clif-ford gates
prodmminus1
i Cni = Cnmminus1 which is appended by
the inverse of the sequence Cminus1
nmminus1 Such an inverse is ef-
ficiently calculated by the Gottesman-Knill theorem [34]Starting in the state |0〉otimesn and applying the full sequenceof Clifford gates we then measure the population in |0〉of each qubit after the sequence This procedure is re-peated l times for different random sequences which inthe limit of large l twirls the error map to a depolariz-ing error map so that once averaged the results fit toan exponential decay Aαm + B where r = 2
nminus1
2n(1 minus α)
is the average error over the Clifford gates (for a widevariety of noise models [35ndash37]) State preparation andmeasurement errors do not affect the decay constant butare captured in the A and B fit parameters The numberof gates in the Clifford group grows superexponentiallywith the number of qubits such that there are 24 one-qubit gates 11520 two-qubit gates and 92897280 three-qubit gates [38] However the method only requires fairsampling from this set and each gate is constructed fromour set of primitive gates such that the average EPC canbe expressed into an one-qubit error per gate (1Q EPG)and two-qubit error per gate (2Q EPG) The exact num-ber of 1Q and 2Q gates depends on the basis used Inthis work our 2Q gate is a controlled NOT (CNOTij)where i is the control and j is the target We generateour 1Q and 2Q Clifford gates using the set of 1Q gates
IXπ2 Xminusπ2 Yπ2 Yminusπ2 where Pθ = eminusiθ2P Withthis gate set there are 22083 1Q primitive gates per 1Q
FIG 1 (Color Online) (a) Schematic of the experi-mental setup and connectivity of the CNOT 2Q gates(controlrarrtarget) (b) 1Q simultaneous RB [0] [1] [2] (c)2Q-1Q simultaneous RB [0 1] [2] and (c) 3Q RB [0 1 2]Under each is a sample (b) 1Q (c) 2Q and (d) 3Q Cliffordgate
Clifford and 15 CNOT gates and 122167 1Q gates per2Q Clifford To generate the 3Q Cliffords we use the setof 1Q gates Xπ2 Xminusπ2 Yminusπ2 plus arbitrary Z rota-tions which are software defined [18] This is the set usedby the QISKit compiler [39] that optimizes the numberof 1Q gates per Clifford For all-to-all connectivity thereare 35 CNOT gates and 116 1Q gates (counting onlyX and Y ) We use the QISKIT compiler to change theconnectivity by removing one of the CNOT gates whichresults in an average of 77 CNOT gates and 184 1Qgates per 3Q Clifford Sample 1Q 2Q and 3Q Cliffordsare shown in Fig 1
In the case of multiqubit systems RB may be per-formed on the full n-qubits (as detailed above) or onsubsets of the system For example it is common to per-form 2Q RB on the subset of two-qubits defining a CNOTgate while the other qubits are quiescent As explained in
3
Ref [32] this RB data will not necessarily decay exponen-tially because the other qubit subspaces are not twirledat best this provides an upper bound fidelity Subsetsare more rigorously characterized by simultaneous RBwhich also measures some level of crosstalk error sinceall qubits are active Herein we will use the notation[i j] [k] to denote benchmarking where the mth setof nm qubits is performing independent nm-qubit RBFor example [0] [1 2] would indicate 1Q RB on qubit0 and 2Q RB on qubits 1 and 2 The different combina-tions for three-qubits are shown in Fig 1
To test 3Q RB we use a device comprising ofthree fixed-frequency superconducting transmon qubits(Q0Q1Q2) of frequencies (535352915237) GHz cou-pled to a common 617GHz bus resonator Each qubithas a dedicated resonator which is used to readout thestate of the qubit and through which microwave drives for1Q and 2Q gates are applied Our 1Q gates are 448 nswide DRAG shaped microwave pulses [40] at the qubitfrequency Our 2Q gates are Gaussian smoothed squaremicrowave pulses applied to a qubit (the control) at thefrequency of one of the other qubits (the target) Thisactivates a cross-resonance interaction which can tunedto build a composite pulse CNOT gate of 240 ns detailsare found in Ref [41] A schematic of the device andCNOT connectivity is shown in Fig 1
For our three-qubit system we consider 8 pos-sible RB combinations simultaneous 1Q RB([0] [1] [2]) separate 2Q RB ([0 1][0 2][1 2])simultaneous 2Q RB and 1Q RB (2Q-1Q RB)([0 1] [2][0 2] [1][1 2] [0]) and finally 3QRB ([0 1 2]) For each combination we performl = 30 averages (except for separate 2Q RB wherel = 20) For simultaneous RB we attempt to matchthe sequence lengths on the different subsystems Thisoccurs naturally (on average) for simultaneous 1Q RBbecause all the 1Q gates are the same width For 2Q-1Qsimultaneous RB we use a fixed ratio of 9 between thenumber of 1Q Clifford gates and 2Q Clifford gates Aspreviously discussed we measure 3Q RB with all-to-alland limited gate connectivity We perform these RBsequences under two different calibration proceduresIn procedure A we calibrate the 1Q gate parameterssimultaneously eg qubit frequency pulse amplitudeand drag amplitude In procedure B we calibrate the 1Qgate parameters individually In both cases we calibratethe 2Q gates separately To give a sense of the typesof curves produced from 1Q 2Q and 3Q RB a subsetof the data from calibration A is shown in Fig 2 Theerrors from the full RB set and for both calibrations aresummarized in Table I
The data from Table I clearly demonstrate the differ-ence between benchmarking 1Q and 2Q gates separatelyversus together in [i] [j k] RB (2Q-1Q simultaneousRB) Almost all errors from 2Q-1Q RB are worse whichis consistent with increased crosstalk There is one ex-
FIG 2 (Color Online) RB Sequences for Qubit 0 for calibra-tion A Black lines are exponential fits to the data and thegray points are from the individual seeds Red squares (bluediamonds) are the averages over these seeds for the light gray(dark gray) points (a) 1Q RB from simultaneous 1Q (redsquares) and 2Q-1Q RB (blue diamonds) (b) 2Q RB for the01 pair performed in isolation (red square) and simultane-ously with 1Q RB on Q2 (blue diamonds) The RB fidelityfor the simultaneous is slightly worse (c) 3Q RB for all-to-all connectivity (red squares) and for limited connectivity(blue squares) Limited connectivity is achieved by removingCNOT12
ception CNOT12 for calibration A which decreases from31times10minus2 to 215times10minus2 This highlights the differencebetween the calibration procedures mainly that they re-sult in different calibrated values for the qubit frequencyThe qubit frequencies in calibration A are shifted bythe average ZZ interaction between pairs (ZZ01=20 kHzZZ02=352 kHz and ZZ12=114 kHz) Since the ZZ02 shiftis calibrated into the frequency of Q2 for calibration Athere is a Z error when benchmarking CNOT12 if Q0 isin the ground state the opposite is true for calibrationB and so the standalone CNOT12 RB error is very low
4
Cal A Cal B
T1 [295039] micros [424735] micros
T2 [397559] micros [617446] micros
1Q EPC Coherence Limit [653544]times10minus4 [423654]times10minus4
1Q EPG from [0] [1] [2] RB [112(2)086(1)122(2)]times10minus3 [140(5)081(1)168(5)]times10minus3
1Q EPG from [i] [j k] RB [139(3)092(2)133(2)]times10minus3 [165(4)095(2)152(3)]times10minus3
2Q EPC Coherence Limit [675]times10minus3 [566]times10minus3
2Q EPG from [i j] RB [16(1)155(10)31(2)]times10minus2 [093(7)29(1)13(1)]times10minus2
2Q EPG from [i j] [k] RB [22(1)205(10)215(10)]times10minus2 [255(10)44(2)46(2)]times10minus2
3Q EPC from [0 1 2] RB (all-to-all) 0106(2) 0302(6)
3Q EPC from [0 1 2] RB (omit CNOT12) 0207(3) NA
TABLE I EPG (error per gate) and EPC (error per Clifford) from different RB experiments in [Q0Q1Q2] order for 1Q (one-qubit) EPG and in order [CNOT01 CNOT02 CNOT12] for the 2Q (two-qubit) EPG 1Q EPG is the error per gate averagedover the set indicated in the main text 2Q EPG is calculated from the 2Q EPC assuming the 1Q EPG from [0] [1] [2]benchmarking There is no measure of 3Q EPC omitting CNOT12 for calibration B because the error was too high to properlyfit the data to an exponential decay curve The coherence limited errors are calculated assuming only errors from T1 and T2Variability in T1 and T2 between the calibrations is due to drift over the approximately three days between experiments Errorsreflect the uncertainty in the fit parameters
Cal A Cal B
All-to-All OmitCNOT12
All-to-All
3Q EPC from RB 0106 0207 0302
Coherence Limit 0037 008 0035
3Q EPC Predictedfrom [i] [j k] RB
0111 0220 0169
TABLE II Predicted 3Q EPC from 1Q and 2Q EPG numberslisted in Table I by applying Eqn 1 See the main text for adetailed dicussion of the calculation
(13times10minus2) Although there is only a subtle differencebetween the calibration procedures there is a large dif-ference between the 3Q RB errors illustrating how 3Q RBcan be a sensitive probe of such calibration procedureson algorithmic fidelity Overall calibrating the averageZZ into the qubit frequencies maximizes 3Q fidelity TheTable I data also show the importance of connectivity asomitting one of the CNOTs causes the algorithmic errorto increase appreciablyOne of the main questions about 3Q RB is how much
new information does it convey ie can 3Q errors bepredicted from 1Q and 2Q EPG since the 3Q Cliffordgates are built from the set of one and two-qubit gatesTo answer this question we calculate the predicted 3QEPC as
EPC = 1minusΠi=010212(1minus ǫ2i)N23 middot
Πj=012(1minus ǫ1j)N13 (1)
where N2 (N1) is the number of 2Q (1Q) gates per 3QClifford and ǫ1(ǫ2) is the 1Q (2Q) EPG Eqn 1 is for all-to-all connectivity for the limited connectivity we omitone of the CNOTs in the first product and change the
power N23 to N22 The values discussed previously forN1 and N2 did not consider the finite duration of gatesIn reality there will be idle periods on some qubits (eg aCNOT01 gate must wait until any prior gates on qubits 0or 1 are finished) Taking these timing issues into accountand characterizing idle periods as one-qubit gates N1 =347 (N1 = 679) for all-to-all (limited) connectivity Thisis the number used for predicting the 3Q EPC
For the 1Q and 2Q EPG we use two sets of numbersfrom Table I and compare the predicted to measured 3QEPC as shown in Table II The first set are the 1Q and2Q EPG numbers given by the coherence limit whichis a lower bound on the error Unsurprisingly the mea-sured 3Q EPC is much higher than the error predicted bythe coherence limit indicating the the majority of errorsare due to unwanted and uncompensated terms in theHamiltonian such as crosstalk The second set of num-bers is from 2Q-1Q simultaneous RB which should bethe most accurate measure of primitive gate errors In-deed for calibration A the estimate of the 3Q EPC from2Q-1Q RB is accurate for both all-to-all and limited con-nectivity Howevever in the case of calibration B thereis very little agreement between the predicted and mea-sured 3Q EPC demonstrating the utility of 3Q RB asa unique measurement of multiqubit fidelity sensitive tosubtle errors that are not fully revealed by benchmarkingthe primitive gates
In the system studied here the dominant crosstalkerror is due to unwanted ZZ interactions By calibrat-ing the average ZZ shift into the qubit frequencies (cal-ibration A) this error is mitigated to the point that theremaining error is dominated by stochastic terms thatequally affect 2Q-1Q simultaneous and 3Q RB Howeverwhen the ZZ shifts are not compensated (calibration B)their effect depends on the structure of the gate and so
5
different RB processes can measure different errors Theerrors measured from 2Q-1Q RB are lower because thereis an aspect of dynamical decoupling that works to can-cel the ZZ terms For example the qubit performing 1QRB changes directions on the Bloch sphere every approx-imately 50 ns independent of the 2Q Clifford gates ap-plied to the other two qubits However the full 3Q Clif-ford gates have idle periods on the spectator qubits whilethe other qubits perform the 2Q gate (this is schemati-cally illustrated in Fig 1 d) As such there is build upof the coherent errors during this time The structureof the 3Q Clifford gate is not unique (our Clifford gatesminimize the total number of primitive gates) and dif-ferent constructions may amplify or attenuate differenterror terms investigating such constructions in detail isleft for future study
In conclusion we demonstrate for the first time 3QRB and subset 2Q-1Q simultaneous RB Although thereis no true primitive three-qubit gate 3Q RB measures afidelity that is not captured by the one- and two-qubitgate metrics As systems continue to increase in sizeand crosstalk terms dominate error metrics such as 3QRB will play an important role in benchmarking the truealgorithmic fidelity of these large systems
We thank Firat Solgun Markus Brink Sami Rosen-blatt and George Keefe for modeling and fabricatingthe device We thank Lev Bishop Andrew CrossEaswar Magesan and Antonio Corcoles for discussionsand manuscript comments We thank Christopher Woodand Sergey Bravyi for help generating the Clifford gatesThis work was supported by the Army Research Officeunder contract W911NF-14-1-0124
lowast dcmckayusibmcom[1] Isaac L Chuang and MA Nielsen ldquoPrescription for ex-
perimental determination of the dynamics of a quantumblack boxrdquo J Mod Opt 44 2455 (1997)
[2] Michael A Nielsen and Isaac L Chuang Quantum Com-
putation and Quantum Information (Cambridge 2000)[3] Alexandre Blais Jay Gambetta A Wallraff D I Schus-
ter S M Girvin M H Devoret and R J SchoelkopfldquoQuantum-information processing with circuit quantumelectrodynamicsrdquo Phys Rev A 75 032329 (2007)
[4] Austin G Fowler Matteo Mariantoni John M Mar-tinis and Andrew N Cleland ldquoSurface codesTowards practical large-scale quantum computationrdquoPhys Rev A 86 032324 (2012)
[5] Maika Takita Andrew W Cross A D CorcolesJerry M Chow and Jay M GambettaldquoExperimental demonstration of fault-tolerantstate preparation with superconducting qubitsrdquoPhys Rev Lett 119 180501 (2017)
[6] Norbert M Linke Dmitri Maslov Martin RoettelerShantanu Debnath Caroline Figgatt Kevin A Lands-man Kenneth Wright and Christopher Monroe ldquoEx-perimental comparison of two quantum computing ar-
chitecturesrdquo Proc Nat Acad Sci 114 3305 (2017)[7] R Barends J Kelly A Megrant A Veitia
D Sank E Jeffrey T C White J Mutus A GFowler B Campbell Y Chen Z Chen B ChiaroA Dunsworth C Neill P OMalley P RoushanA Vainsencher J Wenner A N Korotkov A N Cle-land and John M Martinis ldquoSuperconducting quantumcircuits at the surface code threshold for fault tolerancerdquoNature 508 500 (2014)
[8] Steven T Flammia and Yi-Kai Liu ldquoDirect fi-delity estimation from few pauli measurementsrdquoPhys Rev Lett 106 230501 (2011)
[9] Marcus P da Silva Olivier Landon-Cardinaland David Poulin ldquoPractical characteriza-tion of quantum devices without tomographyrdquoPhys Rev Lett 107 210404 (2011)
[10] David Gross Yi-Kai Liu Steven T Flam-mia Stephen Becker and Jens Eisert ldquoQuan-tum state tomography via compressed sensingrdquoPhys Rev Lett 105 150401 (2010)
[11] Marcus Cramer Martin B Plenio Steven T Flam-mia Rolando Somma David Gross Stephen DBartlett Olivier Landon-Cardinal David Poulin andYi-Kai Liu ldquoEfficient quantum state tomographyrdquoNat Comm 1 149 (2010)
[12] Osama Moussa Marcus P da Silva Colm A Ryan andRaymond Laflamme ldquoPractical experimental certifica-tion of computational quantum gates using a twirlingprocedurerdquo Phys Rev Lett 109 070504 (2012)
[13] Dawei Lu Hang Li Denis-Alexandre Trottier JunLi Aharon Brodutch Anthony P Krismanich AhmadGhavami Gary I Dmitrienko Guilu Long JonathanBaugh and Raymond Laflamme ldquoExperimental estima-tion of average fidelity of a clifford gate on a 7-qubit quan-tum processorrdquo Phys Rev Lett 114 140505 (2015)
[14] C A Riofro D Gross S T Flammia T MonzD Nigg R Blatt and J Eisert ldquoExperimental quan-tum compressed sensing for a seven-qubit systemrdquoNat Comm 8 15305 (2017)
[15] B P Lanyon C Maier M Holzpfel T BaumgratzC Hempel P Jurcevic I Dhand A S Buyskikh A JDaley M Cramer M B Plenio R Blatt and C FRoos ldquoEfficient tomography of a quantum many-bodysystemrdquo Nature Physics XXX XXX (2017)
[16] E Knill D Leibfried R Reichle J Britton R BBlakestad J D Jost C Langer R Ozeri S Sei-delin and D J Wineland ldquoRandomized benchmarkingof quantum gatesrdquo Phys Rev A 77 012307 (2008)
[17] Easwar Magesan J M Gambetta and Joseph EmersonldquoScalable and robust randomized benchmarking of quan-tum processesrdquo Phys Rev Lett 106 180504 (2011)
[18] David C McKay Christopher J Wood SarahSheldon Jerry M Chow and Jay M Gam-betta ldquoEfficient z gates for quantum computingrdquoPhys Rev A 96 022330 (2017)
[19] J M Chow J M Gambetta L Tornberg Jens KochLev S Bishop A A Houck B R Johnson L FrunzioS M Girvin and R J Schoelkopf ldquoRandomized bench-marking and process tomography for gate errors in asolid-state qubitrdquo Phys Rev Lett 102 090502 (2009)
[20] A D Corcoles Jay M Gambetta Jerry M ChowJohn A Smolin Matthew Ware Joel Strand B L TPlourde and M Steffen ldquoProcess verification of two-qubit quantum gates by randomized benchmarkingrdquo
6
Phys Rev A 87 030301 (2013)[21] J P Gaebler T R Tan Y Lin Y Wan R Bowler
A C Keith S Glancy K Coakley E Knill D Leibfriedand D J Wineland ldquoHigh-fidelity universal gate set for9Be
+ion qubitsrdquo Phys Rev Lett 117 060505 (2016)
[22] C J Ballance T P Harty N M Linke M ASepiol and D M Lucas ldquoHigh-fidelity quan-tum logic gates using trapped-ion hyperfine qubitsrdquoPhys Rev Lett 117 060504 (2016)
[23] J P Gaebler A M Meier T R Tan R Bowler Y LinD Hanneke J D Jost J P Home E Knill D Leibfriedand D J Wineland ldquoRandomized benchmarking of mul-tiqubit gatesrdquo Phys Rev Lett 108 260503 (2012)
[24] S Olmschenk R Chicireanu K D Nelson and J VPorto ldquoRandomized benchmarking of atomic qubits inan optical latticerdquo New J Phys 12 113007 (2010)
[25] C A Ryan M Laforest and R Laflamme ldquoRan-domized benchmarking of single- and multi-qubit con-trol in liquid-state nmr quantum information process-ingrdquo New J Phys 11 013034 (2009)
[26] M Veldhorst J C C Hwang C H Yang A W Leen-stra B de Ronde J P Dehollain J T Muhonen F EHudson K M Itoh A Morello and A S DzurakldquoAn addressable quantum dot qubit with fault-tolerantcontrol-fidelityrdquo Nat Nano 9 981 (2014)
[27] Easwar Magesan Jay M Gambetta B R John-son Colm A Ryan Jerry M Chow Seth TMerkel Marcus P da Silva George A KeefeMary B Rothwell Thomas A Ohki Mark B Ketchenand M Steffen ldquoEfficient measurement of quantumgate error by interleaved randomized benchmarkingrdquoPhys Rev Lett 109 080505 (2012)
[28] David C McKay Stefan Filipp Antonio MezzacapoEaswar Magesan Jerry M Chow and Jay M Gam-betta ldquoUniversal gate for fixed-frequency qubits via atunable busrdquo Phys Rev Applied 6 064007 (2016)
[29] Joel Wallman Chris Granade Robin Harper andSteven T Flammia ldquoEstimating the coherence of noiserdquo
New J Phys 17 113020 (2015)[30] Christopher J Wood and Jay M Gambetta ldquoQuantifi-
cation and characterization of leakage errorsrdquo (2017)arxiv170403081
[31] Joel J Wallman Marie Barnhill and JosephEmerson ldquoRobust characterization of leakage errorsrdquoNew J Phys 18 043021 (2016)
[32] Jay M Gambetta A D Corcoles S T Merkel B RJohnson John A Smolin Jerry M Chow Colm A RyanChad Rigetti S Poletto Thomas A Ohki Mark BKetchen and M Steffen ldquoCharacterization of ad-dressability by simultaneous randomized benchmarkingrdquoPhys Rev Lett 109 240504 (2012)
[33] Easwar Magesan Jay M Gambetta and Joseph Emer-son ldquoCharacterizing quantum gates via randomizedbenchmarkingrdquo Phys Rev A 85 042311 (2012)
[34] Daniel Gottesman ldquoThe heisenberg representation ofquantum computersrdquo (1998) arxivquant-ph9807006
[35] Jeffrey M Epstein Andrew W Cross EaswarMagesan and Jay M Gambetta ldquoInvestigatingthe limits of randomized benchmarking protocolsrdquoPhys Rev A 89 062321 (2014)
[36] Joel J Wallman ldquoRandomized benchmarking with gate-dependent noiserdquo (2017) arxiv170309835
[37] Timothy Proctor Kenneth Rudinger Kevin Young
Mohan Sarovar and Robin Blume-Kohout ldquoWhatrandomized benchmarking actually measuresrdquoPhys Rev Lett 119 130502 (2017)
[38] Maris Ozols ldquoClifford grouprdquo (2008)[39] ldquoQISKIT SDKrdquo Online (2017)[40] F Motzoi J M Gambetta P Rebentrost
and F K Wilhelm ldquoSimple pulses for elimi-nation of leakage in weakly nonlinear qubitsrdquoPhys Rev Lett 103 110501 (2009)
[41] Sarah Sheldon Easwar Magesan Jerry M Chowand Jay M Gambetta ldquoProcedure for systematicallytuning up cross-talk in the cross-resonance gaterdquoPhys Rev A 93 060302 (2016)
2
the full n-qubit space can be performed by construct-ing sequences from the n-qubit Clifford group Addi-tionally the n-qubit space can be subdivided into setsof qubits ni and ni-qubit RB performed in each sub-set simultaneously [32] Both methods give metrics offidelity in the n-qubit space Despite the availability ofthese two methods there has been no demonstration ofRB with n gt 2 since it is viewed as sufficient to charac-terize only the primitive gateset Here we show for thefirst time a variety of three-qubit RB combinations in athree-qubit fixed-frequency superconducting device Forall-to-all gate connectivity we measure a three-qubit errorper Clifford (3Q EPC) of 0106 which is well-predictedby the primitive gate errors from two-qubitone-qubit si-multaneous RB However we find a strong dependenceon calibration procedure that is not apparent a prioriFor one such calibration the error increases to 0302Importantly this increase in error is not predicted bya commensurate increase in the primitive gate errors asmeasured from simultaneous RB We also show the im-portance of connectivity in devices as the 3Q EPC in-creases to 0207 when we limit the device to have lineargate connectivity
Before describing our experiment in detail we first pro-vide a brief summary of the RB method a detailed dis-cussion of RB can be found in Ref [33] The main idea isto construct a m-length sequence of random n-qubit Clif-ford gates
prodmminus1
i Cni = Cnmminus1 which is appended by
the inverse of the sequence Cminus1
nmminus1 Such an inverse is ef-
ficiently calculated by the Gottesman-Knill theorem [34]Starting in the state |0〉otimesn and applying the full sequenceof Clifford gates we then measure the population in |0〉of each qubit after the sequence This procedure is re-peated l times for different random sequences which inthe limit of large l twirls the error map to a depolariz-ing error map so that once averaged the results fit toan exponential decay Aαm + B where r = 2
nminus1
2n(1 minus α)
is the average error over the Clifford gates (for a widevariety of noise models [35ndash37]) State preparation andmeasurement errors do not affect the decay constant butare captured in the A and B fit parameters The numberof gates in the Clifford group grows superexponentiallywith the number of qubits such that there are 24 one-qubit gates 11520 two-qubit gates and 92897280 three-qubit gates [38] However the method only requires fairsampling from this set and each gate is constructed fromour set of primitive gates such that the average EPC canbe expressed into an one-qubit error per gate (1Q EPG)and two-qubit error per gate (2Q EPG) The exact num-ber of 1Q and 2Q gates depends on the basis used Inthis work our 2Q gate is a controlled NOT (CNOTij)where i is the control and j is the target We generateour 1Q and 2Q Clifford gates using the set of 1Q gates
IXπ2 Xminusπ2 Yπ2 Yminusπ2 where Pθ = eminusiθ2P Withthis gate set there are 22083 1Q primitive gates per 1Q
FIG 1 (Color Online) (a) Schematic of the experi-mental setup and connectivity of the CNOT 2Q gates(controlrarrtarget) (b) 1Q simultaneous RB [0] [1] [2] (c)2Q-1Q simultaneous RB [0 1] [2] and (c) 3Q RB [0 1 2]Under each is a sample (b) 1Q (c) 2Q and (d) 3Q Cliffordgate
Clifford and 15 CNOT gates and 122167 1Q gates per2Q Clifford To generate the 3Q Cliffords we use the setof 1Q gates Xπ2 Xminusπ2 Yminusπ2 plus arbitrary Z rota-tions which are software defined [18] This is the set usedby the QISKit compiler [39] that optimizes the numberof 1Q gates per Clifford For all-to-all connectivity thereare 35 CNOT gates and 116 1Q gates (counting onlyX and Y ) We use the QISKIT compiler to change theconnectivity by removing one of the CNOT gates whichresults in an average of 77 CNOT gates and 184 1Qgates per 3Q Clifford Sample 1Q 2Q and 3Q Cliffordsare shown in Fig 1
In the case of multiqubit systems RB may be per-formed on the full n-qubits (as detailed above) or onsubsets of the system For example it is common to per-form 2Q RB on the subset of two-qubits defining a CNOTgate while the other qubits are quiescent As explained in
3
Ref [32] this RB data will not necessarily decay exponen-tially because the other qubit subspaces are not twirledat best this provides an upper bound fidelity Subsetsare more rigorously characterized by simultaneous RBwhich also measures some level of crosstalk error sinceall qubits are active Herein we will use the notation[i j] [k] to denote benchmarking where the mth setof nm qubits is performing independent nm-qubit RBFor example [0] [1 2] would indicate 1Q RB on qubit0 and 2Q RB on qubits 1 and 2 The different combina-tions for three-qubits are shown in Fig 1
To test 3Q RB we use a device comprising ofthree fixed-frequency superconducting transmon qubits(Q0Q1Q2) of frequencies (535352915237) GHz cou-pled to a common 617GHz bus resonator Each qubithas a dedicated resonator which is used to readout thestate of the qubit and through which microwave drives for1Q and 2Q gates are applied Our 1Q gates are 448 nswide DRAG shaped microwave pulses [40] at the qubitfrequency Our 2Q gates are Gaussian smoothed squaremicrowave pulses applied to a qubit (the control) at thefrequency of one of the other qubits (the target) Thisactivates a cross-resonance interaction which can tunedto build a composite pulse CNOT gate of 240 ns detailsare found in Ref [41] A schematic of the device andCNOT connectivity is shown in Fig 1
For our three-qubit system we consider 8 pos-sible RB combinations simultaneous 1Q RB([0] [1] [2]) separate 2Q RB ([0 1][0 2][1 2])simultaneous 2Q RB and 1Q RB (2Q-1Q RB)([0 1] [2][0 2] [1][1 2] [0]) and finally 3QRB ([0 1 2]) For each combination we performl = 30 averages (except for separate 2Q RB wherel = 20) For simultaneous RB we attempt to matchthe sequence lengths on the different subsystems Thisoccurs naturally (on average) for simultaneous 1Q RBbecause all the 1Q gates are the same width For 2Q-1Qsimultaneous RB we use a fixed ratio of 9 between thenumber of 1Q Clifford gates and 2Q Clifford gates Aspreviously discussed we measure 3Q RB with all-to-alland limited gate connectivity We perform these RBsequences under two different calibration proceduresIn procedure A we calibrate the 1Q gate parameterssimultaneously eg qubit frequency pulse amplitudeand drag amplitude In procedure B we calibrate the 1Qgate parameters individually In both cases we calibratethe 2Q gates separately To give a sense of the typesof curves produced from 1Q 2Q and 3Q RB a subsetof the data from calibration A is shown in Fig 2 Theerrors from the full RB set and for both calibrations aresummarized in Table I
The data from Table I clearly demonstrate the differ-ence between benchmarking 1Q and 2Q gates separatelyversus together in [i] [j k] RB (2Q-1Q simultaneousRB) Almost all errors from 2Q-1Q RB are worse whichis consistent with increased crosstalk There is one ex-
FIG 2 (Color Online) RB Sequences for Qubit 0 for calibra-tion A Black lines are exponential fits to the data and thegray points are from the individual seeds Red squares (bluediamonds) are the averages over these seeds for the light gray(dark gray) points (a) 1Q RB from simultaneous 1Q (redsquares) and 2Q-1Q RB (blue diamonds) (b) 2Q RB for the01 pair performed in isolation (red square) and simultane-ously with 1Q RB on Q2 (blue diamonds) The RB fidelityfor the simultaneous is slightly worse (c) 3Q RB for all-to-all connectivity (red squares) and for limited connectivity(blue squares) Limited connectivity is achieved by removingCNOT12
ception CNOT12 for calibration A which decreases from31times10minus2 to 215times10minus2 This highlights the differencebetween the calibration procedures mainly that they re-sult in different calibrated values for the qubit frequencyThe qubit frequencies in calibration A are shifted bythe average ZZ interaction between pairs (ZZ01=20 kHzZZ02=352 kHz and ZZ12=114 kHz) Since the ZZ02 shiftis calibrated into the frequency of Q2 for calibration Athere is a Z error when benchmarking CNOT12 if Q0 isin the ground state the opposite is true for calibrationB and so the standalone CNOT12 RB error is very low
4
Cal A Cal B
T1 [295039] micros [424735] micros
T2 [397559] micros [617446] micros
1Q EPC Coherence Limit [653544]times10minus4 [423654]times10minus4
1Q EPG from [0] [1] [2] RB [112(2)086(1)122(2)]times10minus3 [140(5)081(1)168(5)]times10minus3
1Q EPG from [i] [j k] RB [139(3)092(2)133(2)]times10minus3 [165(4)095(2)152(3)]times10minus3
2Q EPC Coherence Limit [675]times10minus3 [566]times10minus3
2Q EPG from [i j] RB [16(1)155(10)31(2)]times10minus2 [093(7)29(1)13(1)]times10minus2
2Q EPG from [i j] [k] RB [22(1)205(10)215(10)]times10minus2 [255(10)44(2)46(2)]times10minus2
3Q EPC from [0 1 2] RB (all-to-all) 0106(2) 0302(6)
3Q EPC from [0 1 2] RB (omit CNOT12) 0207(3) NA
TABLE I EPG (error per gate) and EPC (error per Clifford) from different RB experiments in [Q0Q1Q2] order for 1Q (one-qubit) EPG and in order [CNOT01 CNOT02 CNOT12] for the 2Q (two-qubit) EPG 1Q EPG is the error per gate averagedover the set indicated in the main text 2Q EPG is calculated from the 2Q EPC assuming the 1Q EPG from [0] [1] [2]benchmarking There is no measure of 3Q EPC omitting CNOT12 for calibration B because the error was too high to properlyfit the data to an exponential decay curve The coherence limited errors are calculated assuming only errors from T1 and T2Variability in T1 and T2 between the calibrations is due to drift over the approximately three days between experiments Errorsreflect the uncertainty in the fit parameters
Cal A Cal B
All-to-All OmitCNOT12
All-to-All
3Q EPC from RB 0106 0207 0302
Coherence Limit 0037 008 0035
3Q EPC Predictedfrom [i] [j k] RB
0111 0220 0169
TABLE II Predicted 3Q EPC from 1Q and 2Q EPG numberslisted in Table I by applying Eqn 1 See the main text for adetailed dicussion of the calculation
(13times10minus2) Although there is only a subtle differencebetween the calibration procedures there is a large dif-ference between the 3Q RB errors illustrating how 3Q RBcan be a sensitive probe of such calibration procedureson algorithmic fidelity Overall calibrating the averageZZ into the qubit frequencies maximizes 3Q fidelity TheTable I data also show the importance of connectivity asomitting one of the CNOTs causes the algorithmic errorto increase appreciablyOne of the main questions about 3Q RB is how much
new information does it convey ie can 3Q errors bepredicted from 1Q and 2Q EPG since the 3Q Cliffordgates are built from the set of one and two-qubit gatesTo answer this question we calculate the predicted 3QEPC as
EPC = 1minusΠi=010212(1minus ǫ2i)N23 middot
Πj=012(1minus ǫ1j)N13 (1)
where N2 (N1) is the number of 2Q (1Q) gates per 3QClifford and ǫ1(ǫ2) is the 1Q (2Q) EPG Eqn 1 is for all-to-all connectivity for the limited connectivity we omitone of the CNOTs in the first product and change the
power N23 to N22 The values discussed previously forN1 and N2 did not consider the finite duration of gatesIn reality there will be idle periods on some qubits (eg aCNOT01 gate must wait until any prior gates on qubits 0or 1 are finished) Taking these timing issues into accountand characterizing idle periods as one-qubit gates N1 =347 (N1 = 679) for all-to-all (limited) connectivity Thisis the number used for predicting the 3Q EPC
For the 1Q and 2Q EPG we use two sets of numbersfrom Table I and compare the predicted to measured 3QEPC as shown in Table II The first set are the 1Q and2Q EPG numbers given by the coherence limit whichis a lower bound on the error Unsurprisingly the mea-sured 3Q EPC is much higher than the error predicted bythe coherence limit indicating the the majority of errorsare due to unwanted and uncompensated terms in theHamiltonian such as crosstalk The second set of num-bers is from 2Q-1Q simultaneous RB which should bethe most accurate measure of primitive gate errors In-deed for calibration A the estimate of the 3Q EPC from2Q-1Q RB is accurate for both all-to-all and limited con-nectivity Howevever in the case of calibration B thereis very little agreement between the predicted and mea-sured 3Q EPC demonstrating the utility of 3Q RB asa unique measurement of multiqubit fidelity sensitive tosubtle errors that are not fully revealed by benchmarkingthe primitive gates
In the system studied here the dominant crosstalkerror is due to unwanted ZZ interactions By calibrat-ing the average ZZ shift into the qubit frequencies (cal-ibration A) this error is mitigated to the point that theremaining error is dominated by stochastic terms thatequally affect 2Q-1Q simultaneous and 3Q RB Howeverwhen the ZZ shifts are not compensated (calibration B)their effect depends on the structure of the gate and so
5
different RB processes can measure different errors Theerrors measured from 2Q-1Q RB are lower because thereis an aspect of dynamical decoupling that works to can-cel the ZZ terms For example the qubit performing 1QRB changes directions on the Bloch sphere every approx-imately 50 ns independent of the 2Q Clifford gates ap-plied to the other two qubits However the full 3Q Clif-ford gates have idle periods on the spectator qubits whilethe other qubits perform the 2Q gate (this is schemati-cally illustrated in Fig 1 d) As such there is build upof the coherent errors during this time The structureof the 3Q Clifford gate is not unique (our Clifford gatesminimize the total number of primitive gates) and dif-ferent constructions may amplify or attenuate differenterror terms investigating such constructions in detail isleft for future study
In conclusion we demonstrate for the first time 3QRB and subset 2Q-1Q simultaneous RB Although thereis no true primitive three-qubit gate 3Q RB measures afidelity that is not captured by the one- and two-qubitgate metrics As systems continue to increase in sizeand crosstalk terms dominate error metrics such as 3QRB will play an important role in benchmarking the truealgorithmic fidelity of these large systems
We thank Firat Solgun Markus Brink Sami Rosen-blatt and George Keefe for modeling and fabricatingthe device We thank Lev Bishop Andrew CrossEaswar Magesan and Antonio Corcoles for discussionsand manuscript comments We thank Christopher Woodand Sergey Bravyi for help generating the Clifford gatesThis work was supported by the Army Research Officeunder contract W911NF-14-1-0124
lowast dcmckayusibmcom[1] Isaac L Chuang and MA Nielsen ldquoPrescription for ex-
perimental determination of the dynamics of a quantumblack boxrdquo J Mod Opt 44 2455 (1997)
[2] Michael A Nielsen and Isaac L Chuang Quantum Com-
putation and Quantum Information (Cambridge 2000)[3] Alexandre Blais Jay Gambetta A Wallraff D I Schus-
ter S M Girvin M H Devoret and R J SchoelkopfldquoQuantum-information processing with circuit quantumelectrodynamicsrdquo Phys Rev A 75 032329 (2007)
[4] Austin G Fowler Matteo Mariantoni John M Mar-tinis and Andrew N Cleland ldquoSurface codesTowards practical large-scale quantum computationrdquoPhys Rev A 86 032324 (2012)
[5] Maika Takita Andrew W Cross A D CorcolesJerry M Chow and Jay M GambettaldquoExperimental demonstration of fault-tolerantstate preparation with superconducting qubitsrdquoPhys Rev Lett 119 180501 (2017)
[6] Norbert M Linke Dmitri Maslov Martin RoettelerShantanu Debnath Caroline Figgatt Kevin A Lands-man Kenneth Wright and Christopher Monroe ldquoEx-perimental comparison of two quantum computing ar-
chitecturesrdquo Proc Nat Acad Sci 114 3305 (2017)[7] R Barends J Kelly A Megrant A Veitia
D Sank E Jeffrey T C White J Mutus A GFowler B Campbell Y Chen Z Chen B ChiaroA Dunsworth C Neill P OMalley P RoushanA Vainsencher J Wenner A N Korotkov A N Cle-land and John M Martinis ldquoSuperconducting quantumcircuits at the surface code threshold for fault tolerancerdquoNature 508 500 (2014)
[8] Steven T Flammia and Yi-Kai Liu ldquoDirect fi-delity estimation from few pauli measurementsrdquoPhys Rev Lett 106 230501 (2011)
[9] Marcus P da Silva Olivier Landon-Cardinaland David Poulin ldquoPractical characteriza-tion of quantum devices without tomographyrdquoPhys Rev Lett 107 210404 (2011)
[10] David Gross Yi-Kai Liu Steven T Flam-mia Stephen Becker and Jens Eisert ldquoQuan-tum state tomography via compressed sensingrdquoPhys Rev Lett 105 150401 (2010)
[11] Marcus Cramer Martin B Plenio Steven T Flam-mia Rolando Somma David Gross Stephen DBartlett Olivier Landon-Cardinal David Poulin andYi-Kai Liu ldquoEfficient quantum state tomographyrdquoNat Comm 1 149 (2010)
[12] Osama Moussa Marcus P da Silva Colm A Ryan andRaymond Laflamme ldquoPractical experimental certifica-tion of computational quantum gates using a twirlingprocedurerdquo Phys Rev Lett 109 070504 (2012)
[13] Dawei Lu Hang Li Denis-Alexandre Trottier JunLi Aharon Brodutch Anthony P Krismanich AhmadGhavami Gary I Dmitrienko Guilu Long JonathanBaugh and Raymond Laflamme ldquoExperimental estima-tion of average fidelity of a clifford gate on a 7-qubit quan-tum processorrdquo Phys Rev Lett 114 140505 (2015)
[14] C A Riofro D Gross S T Flammia T MonzD Nigg R Blatt and J Eisert ldquoExperimental quan-tum compressed sensing for a seven-qubit systemrdquoNat Comm 8 15305 (2017)
[15] B P Lanyon C Maier M Holzpfel T BaumgratzC Hempel P Jurcevic I Dhand A S Buyskikh A JDaley M Cramer M B Plenio R Blatt and C FRoos ldquoEfficient tomography of a quantum many-bodysystemrdquo Nature Physics XXX XXX (2017)
[16] E Knill D Leibfried R Reichle J Britton R BBlakestad J D Jost C Langer R Ozeri S Sei-delin and D J Wineland ldquoRandomized benchmarkingof quantum gatesrdquo Phys Rev A 77 012307 (2008)
[17] Easwar Magesan J M Gambetta and Joseph EmersonldquoScalable and robust randomized benchmarking of quan-tum processesrdquo Phys Rev Lett 106 180504 (2011)
[18] David C McKay Christopher J Wood SarahSheldon Jerry M Chow and Jay M Gam-betta ldquoEfficient z gates for quantum computingrdquoPhys Rev A 96 022330 (2017)
[19] J M Chow J M Gambetta L Tornberg Jens KochLev S Bishop A A Houck B R Johnson L FrunzioS M Girvin and R J Schoelkopf ldquoRandomized bench-marking and process tomography for gate errors in asolid-state qubitrdquo Phys Rev Lett 102 090502 (2009)
[20] A D Corcoles Jay M Gambetta Jerry M ChowJohn A Smolin Matthew Ware Joel Strand B L TPlourde and M Steffen ldquoProcess verification of two-qubit quantum gates by randomized benchmarkingrdquo
6
Phys Rev A 87 030301 (2013)[21] J P Gaebler T R Tan Y Lin Y Wan R Bowler
A C Keith S Glancy K Coakley E Knill D Leibfriedand D J Wineland ldquoHigh-fidelity universal gate set for9Be
+ion qubitsrdquo Phys Rev Lett 117 060505 (2016)
[22] C J Ballance T P Harty N M Linke M ASepiol and D M Lucas ldquoHigh-fidelity quan-tum logic gates using trapped-ion hyperfine qubitsrdquoPhys Rev Lett 117 060504 (2016)
[23] J P Gaebler A M Meier T R Tan R Bowler Y LinD Hanneke J D Jost J P Home E Knill D Leibfriedand D J Wineland ldquoRandomized benchmarking of mul-tiqubit gatesrdquo Phys Rev Lett 108 260503 (2012)
[24] S Olmschenk R Chicireanu K D Nelson and J VPorto ldquoRandomized benchmarking of atomic qubits inan optical latticerdquo New J Phys 12 113007 (2010)
[25] C A Ryan M Laforest and R Laflamme ldquoRan-domized benchmarking of single- and multi-qubit con-trol in liquid-state nmr quantum information process-ingrdquo New J Phys 11 013034 (2009)
[26] M Veldhorst J C C Hwang C H Yang A W Leen-stra B de Ronde J P Dehollain J T Muhonen F EHudson K M Itoh A Morello and A S DzurakldquoAn addressable quantum dot qubit with fault-tolerantcontrol-fidelityrdquo Nat Nano 9 981 (2014)
[27] Easwar Magesan Jay M Gambetta B R John-son Colm A Ryan Jerry M Chow Seth TMerkel Marcus P da Silva George A KeefeMary B Rothwell Thomas A Ohki Mark B Ketchenand M Steffen ldquoEfficient measurement of quantumgate error by interleaved randomized benchmarkingrdquoPhys Rev Lett 109 080505 (2012)
[28] David C McKay Stefan Filipp Antonio MezzacapoEaswar Magesan Jerry M Chow and Jay M Gam-betta ldquoUniversal gate for fixed-frequency qubits via atunable busrdquo Phys Rev Applied 6 064007 (2016)
[29] Joel Wallman Chris Granade Robin Harper andSteven T Flammia ldquoEstimating the coherence of noiserdquo
New J Phys 17 113020 (2015)[30] Christopher J Wood and Jay M Gambetta ldquoQuantifi-
cation and characterization of leakage errorsrdquo (2017)arxiv170403081
[31] Joel J Wallman Marie Barnhill and JosephEmerson ldquoRobust characterization of leakage errorsrdquoNew J Phys 18 043021 (2016)
[32] Jay M Gambetta A D Corcoles S T Merkel B RJohnson John A Smolin Jerry M Chow Colm A RyanChad Rigetti S Poletto Thomas A Ohki Mark BKetchen and M Steffen ldquoCharacterization of ad-dressability by simultaneous randomized benchmarkingrdquoPhys Rev Lett 109 240504 (2012)
[33] Easwar Magesan Jay M Gambetta and Joseph Emer-son ldquoCharacterizing quantum gates via randomizedbenchmarkingrdquo Phys Rev A 85 042311 (2012)
[34] Daniel Gottesman ldquoThe heisenberg representation ofquantum computersrdquo (1998) arxivquant-ph9807006
[35] Jeffrey M Epstein Andrew W Cross EaswarMagesan and Jay M Gambetta ldquoInvestigatingthe limits of randomized benchmarking protocolsrdquoPhys Rev A 89 062321 (2014)
[36] Joel J Wallman ldquoRandomized benchmarking with gate-dependent noiserdquo (2017) arxiv170309835
[37] Timothy Proctor Kenneth Rudinger Kevin Young
Mohan Sarovar and Robin Blume-Kohout ldquoWhatrandomized benchmarking actually measuresrdquoPhys Rev Lett 119 130502 (2017)
[38] Maris Ozols ldquoClifford grouprdquo (2008)[39] ldquoQISKIT SDKrdquo Online (2017)[40] F Motzoi J M Gambetta P Rebentrost
and F K Wilhelm ldquoSimple pulses for elimi-nation of leakage in weakly nonlinear qubitsrdquoPhys Rev Lett 103 110501 (2009)
[41] Sarah Sheldon Easwar Magesan Jerry M Chowand Jay M Gambetta ldquoProcedure for systematicallytuning up cross-talk in the cross-resonance gaterdquoPhys Rev A 93 060302 (2016)
3
Ref [32] this RB data will not necessarily decay exponen-tially because the other qubit subspaces are not twirledat best this provides an upper bound fidelity Subsetsare more rigorously characterized by simultaneous RBwhich also measures some level of crosstalk error sinceall qubits are active Herein we will use the notation[i j] [k] to denote benchmarking where the mth setof nm qubits is performing independent nm-qubit RBFor example [0] [1 2] would indicate 1Q RB on qubit0 and 2Q RB on qubits 1 and 2 The different combina-tions for three-qubits are shown in Fig 1
To test 3Q RB we use a device comprising ofthree fixed-frequency superconducting transmon qubits(Q0Q1Q2) of frequencies (535352915237) GHz cou-pled to a common 617GHz bus resonator Each qubithas a dedicated resonator which is used to readout thestate of the qubit and through which microwave drives for1Q and 2Q gates are applied Our 1Q gates are 448 nswide DRAG shaped microwave pulses [40] at the qubitfrequency Our 2Q gates are Gaussian smoothed squaremicrowave pulses applied to a qubit (the control) at thefrequency of one of the other qubits (the target) Thisactivates a cross-resonance interaction which can tunedto build a composite pulse CNOT gate of 240 ns detailsare found in Ref [41] A schematic of the device andCNOT connectivity is shown in Fig 1
For our three-qubit system we consider 8 pos-sible RB combinations simultaneous 1Q RB([0] [1] [2]) separate 2Q RB ([0 1][0 2][1 2])simultaneous 2Q RB and 1Q RB (2Q-1Q RB)([0 1] [2][0 2] [1][1 2] [0]) and finally 3QRB ([0 1 2]) For each combination we performl = 30 averages (except for separate 2Q RB wherel = 20) For simultaneous RB we attempt to matchthe sequence lengths on the different subsystems Thisoccurs naturally (on average) for simultaneous 1Q RBbecause all the 1Q gates are the same width For 2Q-1Qsimultaneous RB we use a fixed ratio of 9 between thenumber of 1Q Clifford gates and 2Q Clifford gates Aspreviously discussed we measure 3Q RB with all-to-alland limited gate connectivity We perform these RBsequences under two different calibration proceduresIn procedure A we calibrate the 1Q gate parameterssimultaneously eg qubit frequency pulse amplitudeand drag amplitude In procedure B we calibrate the 1Qgate parameters individually In both cases we calibratethe 2Q gates separately To give a sense of the typesof curves produced from 1Q 2Q and 3Q RB a subsetof the data from calibration A is shown in Fig 2 Theerrors from the full RB set and for both calibrations aresummarized in Table I
The data from Table I clearly demonstrate the differ-ence between benchmarking 1Q and 2Q gates separatelyversus together in [i] [j k] RB (2Q-1Q simultaneousRB) Almost all errors from 2Q-1Q RB are worse whichis consistent with increased crosstalk There is one ex-
FIG 2 (Color Online) RB Sequences for Qubit 0 for calibra-tion A Black lines are exponential fits to the data and thegray points are from the individual seeds Red squares (bluediamonds) are the averages over these seeds for the light gray(dark gray) points (a) 1Q RB from simultaneous 1Q (redsquares) and 2Q-1Q RB (blue diamonds) (b) 2Q RB for the01 pair performed in isolation (red square) and simultane-ously with 1Q RB on Q2 (blue diamonds) The RB fidelityfor the simultaneous is slightly worse (c) 3Q RB for all-to-all connectivity (red squares) and for limited connectivity(blue squares) Limited connectivity is achieved by removingCNOT12
ception CNOT12 for calibration A which decreases from31times10minus2 to 215times10minus2 This highlights the differencebetween the calibration procedures mainly that they re-sult in different calibrated values for the qubit frequencyThe qubit frequencies in calibration A are shifted bythe average ZZ interaction between pairs (ZZ01=20 kHzZZ02=352 kHz and ZZ12=114 kHz) Since the ZZ02 shiftis calibrated into the frequency of Q2 for calibration Athere is a Z error when benchmarking CNOT12 if Q0 isin the ground state the opposite is true for calibrationB and so the standalone CNOT12 RB error is very low
4
Cal A Cal B
T1 [295039] micros [424735] micros
T2 [397559] micros [617446] micros
1Q EPC Coherence Limit [653544]times10minus4 [423654]times10minus4
1Q EPG from [0] [1] [2] RB [112(2)086(1)122(2)]times10minus3 [140(5)081(1)168(5)]times10minus3
1Q EPG from [i] [j k] RB [139(3)092(2)133(2)]times10minus3 [165(4)095(2)152(3)]times10minus3
2Q EPC Coherence Limit [675]times10minus3 [566]times10minus3
2Q EPG from [i j] RB [16(1)155(10)31(2)]times10minus2 [093(7)29(1)13(1)]times10minus2
2Q EPG from [i j] [k] RB [22(1)205(10)215(10)]times10minus2 [255(10)44(2)46(2)]times10minus2
3Q EPC from [0 1 2] RB (all-to-all) 0106(2) 0302(6)
3Q EPC from [0 1 2] RB (omit CNOT12) 0207(3) NA
TABLE I EPG (error per gate) and EPC (error per Clifford) from different RB experiments in [Q0Q1Q2] order for 1Q (one-qubit) EPG and in order [CNOT01 CNOT02 CNOT12] for the 2Q (two-qubit) EPG 1Q EPG is the error per gate averagedover the set indicated in the main text 2Q EPG is calculated from the 2Q EPC assuming the 1Q EPG from [0] [1] [2]benchmarking There is no measure of 3Q EPC omitting CNOT12 for calibration B because the error was too high to properlyfit the data to an exponential decay curve The coherence limited errors are calculated assuming only errors from T1 and T2Variability in T1 and T2 between the calibrations is due to drift over the approximately three days between experiments Errorsreflect the uncertainty in the fit parameters
Cal A Cal B
All-to-All OmitCNOT12
All-to-All
3Q EPC from RB 0106 0207 0302
Coherence Limit 0037 008 0035
3Q EPC Predictedfrom [i] [j k] RB
0111 0220 0169
TABLE II Predicted 3Q EPC from 1Q and 2Q EPG numberslisted in Table I by applying Eqn 1 See the main text for adetailed dicussion of the calculation
(13times10minus2) Although there is only a subtle differencebetween the calibration procedures there is a large dif-ference between the 3Q RB errors illustrating how 3Q RBcan be a sensitive probe of such calibration procedureson algorithmic fidelity Overall calibrating the averageZZ into the qubit frequencies maximizes 3Q fidelity TheTable I data also show the importance of connectivity asomitting one of the CNOTs causes the algorithmic errorto increase appreciablyOne of the main questions about 3Q RB is how much
new information does it convey ie can 3Q errors bepredicted from 1Q and 2Q EPG since the 3Q Cliffordgates are built from the set of one and two-qubit gatesTo answer this question we calculate the predicted 3QEPC as
EPC = 1minusΠi=010212(1minus ǫ2i)N23 middot
Πj=012(1minus ǫ1j)N13 (1)
where N2 (N1) is the number of 2Q (1Q) gates per 3QClifford and ǫ1(ǫ2) is the 1Q (2Q) EPG Eqn 1 is for all-to-all connectivity for the limited connectivity we omitone of the CNOTs in the first product and change the
power N23 to N22 The values discussed previously forN1 and N2 did not consider the finite duration of gatesIn reality there will be idle periods on some qubits (eg aCNOT01 gate must wait until any prior gates on qubits 0or 1 are finished) Taking these timing issues into accountand characterizing idle periods as one-qubit gates N1 =347 (N1 = 679) for all-to-all (limited) connectivity Thisis the number used for predicting the 3Q EPC
For the 1Q and 2Q EPG we use two sets of numbersfrom Table I and compare the predicted to measured 3QEPC as shown in Table II The first set are the 1Q and2Q EPG numbers given by the coherence limit whichis a lower bound on the error Unsurprisingly the mea-sured 3Q EPC is much higher than the error predicted bythe coherence limit indicating the the majority of errorsare due to unwanted and uncompensated terms in theHamiltonian such as crosstalk The second set of num-bers is from 2Q-1Q simultaneous RB which should bethe most accurate measure of primitive gate errors In-deed for calibration A the estimate of the 3Q EPC from2Q-1Q RB is accurate for both all-to-all and limited con-nectivity Howevever in the case of calibration B thereis very little agreement between the predicted and mea-sured 3Q EPC demonstrating the utility of 3Q RB asa unique measurement of multiqubit fidelity sensitive tosubtle errors that are not fully revealed by benchmarkingthe primitive gates
In the system studied here the dominant crosstalkerror is due to unwanted ZZ interactions By calibrat-ing the average ZZ shift into the qubit frequencies (cal-ibration A) this error is mitigated to the point that theremaining error is dominated by stochastic terms thatequally affect 2Q-1Q simultaneous and 3Q RB Howeverwhen the ZZ shifts are not compensated (calibration B)their effect depends on the structure of the gate and so
5
different RB processes can measure different errors Theerrors measured from 2Q-1Q RB are lower because thereis an aspect of dynamical decoupling that works to can-cel the ZZ terms For example the qubit performing 1QRB changes directions on the Bloch sphere every approx-imately 50 ns independent of the 2Q Clifford gates ap-plied to the other two qubits However the full 3Q Clif-ford gates have idle periods on the spectator qubits whilethe other qubits perform the 2Q gate (this is schemati-cally illustrated in Fig 1 d) As such there is build upof the coherent errors during this time The structureof the 3Q Clifford gate is not unique (our Clifford gatesminimize the total number of primitive gates) and dif-ferent constructions may amplify or attenuate differenterror terms investigating such constructions in detail isleft for future study
In conclusion we demonstrate for the first time 3QRB and subset 2Q-1Q simultaneous RB Although thereis no true primitive three-qubit gate 3Q RB measures afidelity that is not captured by the one- and two-qubitgate metrics As systems continue to increase in sizeand crosstalk terms dominate error metrics such as 3QRB will play an important role in benchmarking the truealgorithmic fidelity of these large systems
We thank Firat Solgun Markus Brink Sami Rosen-blatt and George Keefe for modeling and fabricatingthe device We thank Lev Bishop Andrew CrossEaswar Magesan and Antonio Corcoles for discussionsand manuscript comments We thank Christopher Woodand Sergey Bravyi for help generating the Clifford gatesThis work was supported by the Army Research Officeunder contract W911NF-14-1-0124
lowast dcmckayusibmcom[1] Isaac L Chuang and MA Nielsen ldquoPrescription for ex-
perimental determination of the dynamics of a quantumblack boxrdquo J Mod Opt 44 2455 (1997)
[2] Michael A Nielsen and Isaac L Chuang Quantum Com-
putation and Quantum Information (Cambridge 2000)[3] Alexandre Blais Jay Gambetta A Wallraff D I Schus-
ter S M Girvin M H Devoret and R J SchoelkopfldquoQuantum-information processing with circuit quantumelectrodynamicsrdquo Phys Rev A 75 032329 (2007)
[4] Austin G Fowler Matteo Mariantoni John M Mar-tinis and Andrew N Cleland ldquoSurface codesTowards practical large-scale quantum computationrdquoPhys Rev A 86 032324 (2012)
[5] Maika Takita Andrew W Cross A D CorcolesJerry M Chow and Jay M GambettaldquoExperimental demonstration of fault-tolerantstate preparation with superconducting qubitsrdquoPhys Rev Lett 119 180501 (2017)
[6] Norbert M Linke Dmitri Maslov Martin RoettelerShantanu Debnath Caroline Figgatt Kevin A Lands-man Kenneth Wright and Christopher Monroe ldquoEx-perimental comparison of two quantum computing ar-
chitecturesrdquo Proc Nat Acad Sci 114 3305 (2017)[7] R Barends J Kelly A Megrant A Veitia
D Sank E Jeffrey T C White J Mutus A GFowler B Campbell Y Chen Z Chen B ChiaroA Dunsworth C Neill P OMalley P RoushanA Vainsencher J Wenner A N Korotkov A N Cle-land and John M Martinis ldquoSuperconducting quantumcircuits at the surface code threshold for fault tolerancerdquoNature 508 500 (2014)
[8] Steven T Flammia and Yi-Kai Liu ldquoDirect fi-delity estimation from few pauli measurementsrdquoPhys Rev Lett 106 230501 (2011)
[9] Marcus P da Silva Olivier Landon-Cardinaland David Poulin ldquoPractical characteriza-tion of quantum devices without tomographyrdquoPhys Rev Lett 107 210404 (2011)
[10] David Gross Yi-Kai Liu Steven T Flam-mia Stephen Becker and Jens Eisert ldquoQuan-tum state tomography via compressed sensingrdquoPhys Rev Lett 105 150401 (2010)
[11] Marcus Cramer Martin B Plenio Steven T Flam-mia Rolando Somma David Gross Stephen DBartlett Olivier Landon-Cardinal David Poulin andYi-Kai Liu ldquoEfficient quantum state tomographyrdquoNat Comm 1 149 (2010)
[12] Osama Moussa Marcus P da Silva Colm A Ryan andRaymond Laflamme ldquoPractical experimental certifica-tion of computational quantum gates using a twirlingprocedurerdquo Phys Rev Lett 109 070504 (2012)
[13] Dawei Lu Hang Li Denis-Alexandre Trottier JunLi Aharon Brodutch Anthony P Krismanich AhmadGhavami Gary I Dmitrienko Guilu Long JonathanBaugh and Raymond Laflamme ldquoExperimental estima-tion of average fidelity of a clifford gate on a 7-qubit quan-tum processorrdquo Phys Rev Lett 114 140505 (2015)
[14] C A Riofro D Gross S T Flammia T MonzD Nigg R Blatt and J Eisert ldquoExperimental quan-tum compressed sensing for a seven-qubit systemrdquoNat Comm 8 15305 (2017)
[15] B P Lanyon C Maier M Holzpfel T BaumgratzC Hempel P Jurcevic I Dhand A S Buyskikh A JDaley M Cramer M B Plenio R Blatt and C FRoos ldquoEfficient tomography of a quantum many-bodysystemrdquo Nature Physics XXX XXX (2017)
[16] E Knill D Leibfried R Reichle J Britton R BBlakestad J D Jost C Langer R Ozeri S Sei-delin and D J Wineland ldquoRandomized benchmarkingof quantum gatesrdquo Phys Rev A 77 012307 (2008)
[17] Easwar Magesan J M Gambetta and Joseph EmersonldquoScalable and robust randomized benchmarking of quan-tum processesrdquo Phys Rev Lett 106 180504 (2011)
[18] David C McKay Christopher J Wood SarahSheldon Jerry M Chow and Jay M Gam-betta ldquoEfficient z gates for quantum computingrdquoPhys Rev A 96 022330 (2017)
[19] J M Chow J M Gambetta L Tornberg Jens KochLev S Bishop A A Houck B R Johnson L FrunzioS M Girvin and R J Schoelkopf ldquoRandomized bench-marking and process tomography for gate errors in asolid-state qubitrdquo Phys Rev Lett 102 090502 (2009)
[20] A D Corcoles Jay M Gambetta Jerry M ChowJohn A Smolin Matthew Ware Joel Strand B L TPlourde and M Steffen ldquoProcess verification of two-qubit quantum gates by randomized benchmarkingrdquo
6
Phys Rev A 87 030301 (2013)[21] J P Gaebler T R Tan Y Lin Y Wan R Bowler
A C Keith S Glancy K Coakley E Knill D Leibfriedand D J Wineland ldquoHigh-fidelity universal gate set for9Be
+ion qubitsrdquo Phys Rev Lett 117 060505 (2016)
[22] C J Ballance T P Harty N M Linke M ASepiol and D M Lucas ldquoHigh-fidelity quan-tum logic gates using trapped-ion hyperfine qubitsrdquoPhys Rev Lett 117 060504 (2016)
[23] J P Gaebler A M Meier T R Tan R Bowler Y LinD Hanneke J D Jost J P Home E Knill D Leibfriedand D J Wineland ldquoRandomized benchmarking of mul-tiqubit gatesrdquo Phys Rev Lett 108 260503 (2012)
[24] S Olmschenk R Chicireanu K D Nelson and J VPorto ldquoRandomized benchmarking of atomic qubits inan optical latticerdquo New J Phys 12 113007 (2010)
[25] C A Ryan M Laforest and R Laflamme ldquoRan-domized benchmarking of single- and multi-qubit con-trol in liquid-state nmr quantum information process-ingrdquo New J Phys 11 013034 (2009)
[26] M Veldhorst J C C Hwang C H Yang A W Leen-stra B de Ronde J P Dehollain J T Muhonen F EHudson K M Itoh A Morello and A S DzurakldquoAn addressable quantum dot qubit with fault-tolerantcontrol-fidelityrdquo Nat Nano 9 981 (2014)
[27] Easwar Magesan Jay M Gambetta B R John-son Colm A Ryan Jerry M Chow Seth TMerkel Marcus P da Silva George A KeefeMary B Rothwell Thomas A Ohki Mark B Ketchenand M Steffen ldquoEfficient measurement of quantumgate error by interleaved randomized benchmarkingrdquoPhys Rev Lett 109 080505 (2012)
[28] David C McKay Stefan Filipp Antonio MezzacapoEaswar Magesan Jerry M Chow and Jay M Gam-betta ldquoUniversal gate for fixed-frequency qubits via atunable busrdquo Phys Rev Applied 6 064007 (2016)
[29] Joel Wallman Chris Granade Robin Harper andSteven T Flammia ldquoEstimating the coherence of noiserdquo
New J Phys 17 113020 (2015)[30] Christopher J Wood and Jay M Gambetta ldquoQuantifi-
cation and characterization of leakage errorsrdquo (2017)arxiv170403081
[31] Joel J Wallman Marie Barnhill and JosephEmerson ldquoRobust characterization of leakage errorsrdquoNew J Phys 18 043021 (2016)
[32] Jay M Gambetta A D Corcoles S T Merkel B RJohnson John A Smolin Jerry M Chow Colm A RyanChad Rigetti S Poletto Thomas A Ohki Mark BKetchen and M Steffen ldquoCharacterization of ad-dressability by simultaneous randomized benchmarkingrdquoPhys Rev Lett 109 240504 (2012)
[33] Easwar Magesan Jay M Gambetta and Joseph Emer-son ldquoCharacterizing quantum gates via randomizedbenchmarkingrdquo Phys Rev A 85 042311 (2012)
[34] Daniel Gottesman ldquoThe heisenberg representation ofquantum computersrdquo (1998) arxivquant-ph9807006
[35] Jeffrey M Epstein Andrew W Cross EaswarMagesan and Jay M Gambetta ldquoInvestigatingthe limits of randomized benchmarking protocolsrdquoPhys Rev A 89 062321 (2014)
[36] Joel J Wallman ldquoRandomized benchmarking with gate-dependent noiserdquo (2017) arxiv170309835
[37] Timothy Proctor Kenneth Rudinger Kevin Young
Mohan Sarovar and Robin Blume-Kohout ldquoWhatrandomized benchmarking actually measuresrdquoPhys Rev Lett 119 130502 (2017)
[38] Maris Ozols ldquoClifford grouprdquo (2008)[39] ldquoQISKIT SDKrdquo Online (2017)[40] F Motzoi J M Gambetta P Rebentrost
and F K Wilhelm ldquoSimple pulses for elimi-nation of leakage in weakly nonlinear qubitsrdquoPhys Rev Lett 103 110501 (2009)
[41] Sarah Sheldon Easwar Magesan Jerry M Chowand Jay M Gambetta ldquoProcedure for systematicallytuning up cross-talk in the cross-resonance gaterdquoPhys Rev A 93 060302 (2016)
4
Cal A Cal B
T1 [295039] micros [424735] micros
T2 [397559] micros [617446] micros
1Q EPC Coherence Limit [653544]times10minus4 [423654]times10minus4
1Q EPG from [0] [1] [2] RB [112(2)086(1)122(2)]times10minus3 [140(5)081(1)168(5)]times10minus3
1Q EPG from [i] [j k] RB [139(3)092(2)133(2)]times10minus3 [165(4)095(2)152(3)]times10minus3
2Q EPC Coherence Limit [675]times10minus3 [566]times10minus3
2Q EPG from [i j] RB [16(1)155(10)31(2)]times10minus2 [093(7)29(1)13(1)]times10minus2
2Q EPG from [i j] [k] RB [22(1)205(10)215(10)]times10minus2 [255(10)44(2)46(2)]times10minus2
3Q EPC from [0 1 2] RB (all-to-all) 0106(2) 0302(6)
3Q EPC from [0 1 2] RB (omit CNOT12) 0207(3) NA
TABLE I EPG (error per gate) and EPC (error per Clifford) from different RB experiments in [Q0Q1Q2] order for 1Q (one-qubit) EPG and in order [CNOT01 CNOT02 CNOT12] for the 2Q (two-qubit) EPG 1Q EPG is the error per gate averagedover the set indicated in the main text 2Q EPG is calculated from the 2Q EPC assuming the 1Q EPG from [0] [1] [2]benchmarking There is no measure of 3Q EPC omitting CNOT12 for calibration B because the error was too high to properlyfit the data to an exponential decay curve The coherence limited errors are calculated assuming only errors from T1 and T2Variability in T1 and T2 between the calibrations is due to drift over the approximately three days between experiments Errorsreflect the uncertainty in the fit parameters
Cal A Cal B
All-to-All OmitCNOT12
All-to-All
3Q EPC from RB 0106 0207 0302
Coherence Limit 0037 008 0035
3Q EPC Predictedfrom [i] [j k] RB
0111 0220 0169
TABLE II Predicted 3Q EPC from 1Q and 2Q EPG numberslisted in Table I by applying Eqn 1 See the main text for adetailed dicussion of the calculation
(13times10minus2) Although there is only a subtle differencebetween the calibration procedures there is a large dif-ference between the 3Q RB errors illustrating how 3Q RBcan be a sensitive probe of such calibration procedureson algorithmic fidelity Overall calibrating the averageZZ into the qubit frequencies maximizes 3Q fidelity TheTable I data also show the importance of connectivity asomitting one of the CNOTs causes the algorithmic errorto increase appreciablyOne of the main questions about 3Q RB is how much
new information does it convey ie can 3Q errors bepredicted from 1Q and 2Q EPG since the 3Q Cliffordgates are built from the set of one and two-qubit gatesTo answer this question we calculate the predicted 3QEPC as
EPC = 1minusΠi=010212(1minus ǫ2i)N23 middot
Πj=012(1minus ǫ1j)N13 (1)
where N2 (N1) is the number of 2Q (1Q) gates per 3QClifford and ǫ1(ǫ2) is the 1Q (2Q) EPG Eqn 1 is for all-to-all connectivity for the limited connectivity we omitone of the CNOTs in the first product and change the
power N23 to N22 The values discussed previously forN1 and N2 did not consider the finite duration of gatesIn reality there will be idle periods on some qubits (eg aCNOT01 gate must wait until any prior gates on qubits 0or 1 are finished) Taking these timing issues into accountand characterizing idle periods as one-qubit gates N1 =347 (N1 = 679) for all-to-all (limited) connectivity Thisis the number used for predicting the 3Q EPC
For the 1Q and 2Q EPG we use two sets of numbersfrom Table I and compare the predicted to measured 3QEPC as shown in Table II The first set are the 1Q and2Q EPG numbers given by the coherence limit whichis a lower bound on the error Unsurprisingly the mea-sured 3Q EPC is much higher than the error predicted bythe coherence limit indicating the the majority of errorsare due to unwanted and uncompensated terms in theHamiltonian such as crosstalk The second set of num-bers is from 2Q-1Q simultaneous RB which should bethe most accurate measure of primitive gate errors In-deed for calibration A the estimate of the 3Q EPC from2Q-1Q RB is accurate for both all-to-all and limited con-nectivity Howevever in the case of calibration B thereis very little agreement between the predicted and mea-sured 3Q EPC demonstrating the utility of 3Q RB asa unique measurement of multiqubit fidelity sensitive tosubtle errors that are not fully revealed by benchmarkingthe primitive gates
In the system studied here the dominant crosstalkerror is due to unwanted ZZ interactions By calibrat-ing the average ZZ shift into the qubit frequencies (cal-ibration A) this error is mitigated to the point that theremaining error is dominated by stochastic terms thatequally affect 2Q-1Q simultaneous and 3Q RB Howeverwhen the ZZ shifts are not compensated (calibration B)their effect depends on the structure of the gate and so
5
different RB processes can measure different errors Theerrors measured from 2Q-1Q RB are lower because thereis an aspect of dynamical decoupling that works to can-cel the ZZ terms For example the qubit performing 1QRB changes directions on the Bloch sphere every approx-imately 50 ns independent of the 2Q Clifford gates ap-plied to the other two qubits However the full 3Q Clif-ford gates have idle periods on the spectator qubits whilethe other qubits perform the 2Q gate (this is schemati-cally illustrated in Fig 1 d) As such there is build upof the coherent errors during this time The structureof the 3Q Clifford gate is not unique (our Clifford gatesminimize the total number of primitive gates) and dif-ferent constructions may amplify or attenuate differenterror terms investigating such constructions in detail isleft for future study
In conclusion we demonstrate for the first time 3QRB and subset 2Q-1Q simultaneous RB Although thereis no true primitive three-qubit gate 3Q RB measures afidelity that is not captured by the one- and two-qubitgate metrics As systems continue to increase in sizeand crosstalk terms dominate error metrics such as 3QRB will play an important role in benchmarking the truealgorithmic fidelity of these large systems
We thank Firat Solgun Markus Brink Sami Rosen-blatt and George Keefe for modeling and fabricatingthe device We thank Lev Bishop Andrew CrossEaswar Magesan and Antonio Corcoles for discussionsand manuscript comments We thank Christopher Woodand Sergey Bravyi for help generating the Clifford gatesThis work was supported by the Army Research Officeunder contract W911NF-14-1-0124
lowast dcmckayusibmcom[1] Isaac L Chuang and MA Nielsen ldquoPrescription for ex-
perimental determination of the dynamics of a quantumblack boxrdquo J Mod Opt 44 2455 (1997)
[2] Michael A Nielsen and Isaac L Chuang Quantum Com-
putation and Quantum Information (Cambridge 2000)[3] Alexandre Blais Jay Gambetta A Wallraff D I Schus-
ter S M Girvin M H Devoret and R J SchoelkopfldquoQuantum-information processing with circuit quantumelectrodynamicsrdquo Phys Rev A 75 032329 (2007)
[4] Austin G Fowler Matteo Mariantoni John M Mar-tinis and Andrew N Cleland ldquoSurface codesTowards practical large-scale quantum computationrdquoPhys Rev A 86 032324 (2012)
[5] Maika Takita Andrew W Cross A D CorcolesJerry M Chow and Jay M GambettaldquoExperimental demonstration of fault-tolerantstate preparation with superconducting qubitsrdquoPhys Rev Lett 119 180501 (2017)
[6] Norbert M Linke Dmitri Maslov Martin RoettelerShantanu Debnath Caroline Figgatt Kevin A Lands-man Kenneth Wright and Christopher Monroe ldquoEx-perimental comparison of two quantum computing ar-
chitecturesrdquo Proc Nat Acad Sci 114 3305 (2017)[7] R Barends J Kelly A Megrant A Veitia
D Sank E Jeffrey T C White J Mutus A GFowler B Campbell Y Chen Z Chen B ChiaroA Dunsworth C Neill P OMalley P RoushanA Vainsencher J Wenner A N Korotkov A N Cle-land and John M Martinis ldquoSuperconducting quantumcircuits at the surface code threshold for fault tolerancerdquoNature 508 500 (2014)
[8] Steven T Flammia and Yi-Kai Liu ldquoDirect fi-delity estimation from few pauli measurementsrdquoPhys Rev Lett 106 230501 (2011)
[9] Marcus P da Silva Olivier Landon-Cardinaland David Poulin ldquoPractical characteriza-tion of quantum devices without tomographyrdquoPhys Rev Lett 107 210404 (2011)
[10] David Gross Yi-Kai Liu Steven T Flam-mia Stephen Becker and Jens Eisert ldquoQuan-tum state tomography via compressed sensingrdquoPhys Rev Lett 105 150401 (2010)
[11] Marcus Cramer Martin B Plenio Steven T Flam-mia Rolando Somma David Gross Stephen DBartlett Olivier Landon-Cardinal David Poulin andYi-Kai Liu ldquoEfficient quantum state tomographyrdquoNat Comm 1 149 (2010)
[12] Osama Moussa Marcus P da Silva Colm A Ryan andRaymond Laflamme ldquoPractical experimental certifica-tion of computational quantum gates using a twirlingprocedurerdquo Phys Rev Lett 109 070504 (2012)
[13] Dawei Lu Hang Li Denis-Alexandre Trottier JunLi Aharon Brodutch Anthony P Krismanich AhmadGhavami Gary I Dmitrienko Guilu Long JonathanBaugh and Raymond Laflamme ldquoExperimental estima-tion of average fidelity of a clifford gate on a 7-qubit quan-tum processorrdquo Phys Rev Lett 114 140505 (2015)
[14] C A Riofro D Gross S T Flammia T MonzD Nigg R Blatt and J Eisert ldquoExperimental quan-tum compressed sensing for a seven-qubit systemrdquoNat Comm 8 15305 (2017)
[15] B P Lanyon C Maier M Holzpfel T BaumgratzC Hempel P Jurcevic I Dhand A S Buyskikh A JDaley M Cramer M B Plenio R Blatt and C FRoos ldquoEfficient tomography of a quantum many-bodysystemrdquo Nature Physics XXX XXX (2017)
[16] E Knill D Leibfried R Reichle J Britton R BBlakestad J D Jost C Langer R Ozeri S Sei-delin and D J Wineland ldquoRandomized benchmarkingof quantum gatesrdquo Phys Rev A 77 012307 (2008)
[17] Easwar Magesan J M Gambetta and Joseph EmersonldquoScalable and robust randomized benchmarking of quan-tum processesrdquo Phys Rev Lett 106 180504 (2011)
[18] David C McKay Christopher J Wood SarahSheldon Jerry M Chow and Jay M Gam-betta ldquoEfficient z gates for quantum computingrdquoPhys Rev A 96 022330 (2017)
[19] J M Chow J M Gambetta L Tornberg Jens KochLev S Bishop A A Houck B R Johnson L FrunzioS M Girvin and R J Schoelkopf ldquoRandomized bench-marking and process tomography for gate errors in asolid-state qubitrdquo Phys Rev Lett 102 090502 (2009)
[20] A D Corcoles Jay M Gambetta Jerry M ChowJohn A Smolin Matthew Ware Joel Strand B L TPlourde and M Steffen ldquoProcess verification of two-qubit quantum gates by randomized benchmarkingrdquo
6
Phys Rev A 87 030301 (2013)[21] J P Gaebler T R Tan Y Lin Y Wan R Bowler
A C Keith S Glancy K Coakley E Knill D Leibfriedand D J Wineland ldquoHigh-fidelity universal gate set for9Be
+ion qubitsrdquo Phys Rev Lett 117 060505 (2016)
[22] C J Ballance T P Harty N M Linke M ASepiol and D M Lucas ldquoHigh-fidelity quan-tum logic gates using trapped-ion hyperfine qubitsrdquoPhys Rev Lett 117 060504 (2016)
[23] J P Gaebler A M Meier T R Tan R Bowler Y LinD Hanneke J D Jost J P Home E Knill D Leibfriedand D J Wineland ldquoRandomized benchmarking of mul-tiqubit gatesrdquo Phys Rev Lett 108 260503 (2012)
[24] S Olmschenk R Chicireanu K D Nelson and J VPorto ldquoRandomized benchmarking of atomic qubits inan optical latticerdquo New J Phys 12 113007 (2010)
[25] C A Ryan M Laforest and R Laflamme ldquoRan-domized benchmarking of single- and multi-qubit con-trol in liquid-state nmr quantum information process-ingrdquo New J Phys 11 013034 (2009)
[26] M Veldhorst J C C Hwang C H Yang A W Leen-stra B de Ronde J P Dehollain J T Muhonen F EHudson K M Itoh A Morello and A S DzurakldquoAn addressable quantum dot qubit with fault-tolerantcontrol-fidelityrdquo Nat Nano 9 981 (2014)
[27] Easwar Magesan Jay M Gambetta B R John-son Colm A Ryan Jerry M Chow Seth TMerkel Marcus P da Silva George A KeefeMary B Rothwell Thomas A Ohki Mark B Ketchenand M Steffen ldquoEfficient measurement of quantumgate error by interleaved randomized benchmarkingrdquoPhys Rev Lett 109 080505 (2012)
[28] David C McKay Stefan Filipp Antonio MezzacapoEaswar Magesan Jerry M Chow and Jay M Gam-betta ldquoUniversal gate for fixed-frequency qubits via atunable busrdquo Phys Rev Applied 6 064007 (2016)
[29] Joel Wallman Chris Granade Robin Harper andSteven T Flammia ldquoEstimating the coherence of noiserdquo
New J Phys 17 113020 (2015)[30] Christopher J Wood and Jay M Gambetta ldquoQuantifi-
cation and characterization of leakage errorsrdquo (2017)arxiv170403081
[31] Joel J Wallman Marie Barnhill and JosephEmerson ldquoRobust characterization of leakage errorsrdquoNew J Phys 18 043021 (2016)
[32] Jay M Gambetta A D Corcoles S T Merkel B RJohnson John A Smolin Jerry M Chow Colm A RyanChad Rigetti S Poletto Thomas A Ohki Mark BKetchen and M Steffen ldquoCharacterization of ad-dressability by simultaneous randomized benchmarkingrdquoPhys Rev Lett 109 240504 (2012)
[33] Easwar Magesan Jay M Gambetta and Joseph Emer-son ldquoCharacterizing quantum gates via randomizedbenchmarkingrdquo Phys Rev A 85 042311 (2012)
[34] Daniel Gottesman ldquoThe heisenberg representation ofquantum computersrdquo (1998) arxivquant-ph9807006
[35] Jeffrey M Epstein Andrew W Cross EaswarMagesan and Jay M Gambetta ldquoInvestigatingthe limits of randomized benchmarking protocolsrdquoPhys Rev A 89 062321 (2014)
[36] Joel J Wallman ldquoRandomized benchmarking with gate-dependent noiserdquo (2017) arxiv170309835
[37] Timothy Proctor Kenneth Rudinger Kevin Young
Mohan Sarovar and Robin Blume-Kohout ldquoWhatrandomized benchmarking actually measuresrdquoPhys Rev Lett 119 130502 (2017)
[38] Maris Ozols ldquoClifford grouprdquo (2008)[39] ldquoQISKIT SDKrdquo Online (2017)[40] F Motzoi J M Gambetta P Rebentrost
and F K Wilhelm ldquoSimple pulses for elimi-nation of leakage in weakly nonlinear qubitsrdquoPhys Rev Lett 103 110501 (2009)
[41] Sarah Sheldon Easwar Magesan Jerry M Chowand Jay M Gambetta ldquoProcedure for systematicallytuning up cross-talk in the cross-resonance gaterdquoPhys Rev A 93 060302 (2016)
5
different RB processes can measure different errors Theerrors measured from 2Q-1Q RB are lower because thereis an aspect of dynamical decoupling that works to can-cel the ZZ terms For example the qubit performing 1QRB changes directions on the Bloch sphere every approx-imately 50 ns independent of the 2Q Clifford gates ap-plied to the other two qubits However the full 3Q Clif-ford gates have idle periods on the spectator qubits whilethe other qubits perform the 2Q gate (this is schemati-cally illustrated in Fig 1 d) As such there is build upof the coherent errors during this time The structureof the 3Q Clifford gate is not unique (our Clifford gatesminimize the total number of primitive gates) and dif-ferent constructions may amplify or attenuate differenterror terms investigating such constructions in detail isleft for future study
In conclusion we demonstrate for the first time 3QRB and subset 2Q-1Q simultaneous RB Although thereis no true primitive three-qubit gate 3Q RB measures afidelity that is not captured by the one- and two-qubitgate metrics As systems continue to increase in sizeand crosstalk terms dominate error metrics such as 3QRB will play an important role in benchmarking the truealgorithmic fidelity of these large systems
We thank Firat Solgun Markus Brink Sami Rosen-blatt and George Keefe for modeling and fabricatingthe device We thank Lev Bishop Andrew CrossEaswar Magesan and Antonio Corcoles for discussionsand manuscript comments We thank Christopher Woodand Sergey Bravyi for help generating the Clifford gatesThis work was supported by the Army Research Officeunder contract W911NF-14-1-0124
lowast dcmckayusibmcom[1] Isaac L Chuang and MA Nielsen ldquoPrescription for ex-
perimental determination of the dynamics of a quantumblack boxrdquo J Mod Opt 44 2455 (1997)
[2] Michael A Nielsen and Isaac L Chuang Quantum Com-
putation and Quantum Information (Cambridge 2000)[3] Alexandre Blais Jay Gambetta A Wallraff D I Schus-
ter S M Girvin M H Devoret and R J SchoelkopfldquoQuantum-information processing with circuit quantumelectrodynamicsrdquo Phys Rev A 75 032329 (2007)
[4] Austin G Fowler Matteo Mariantoni John M Mar-tinis and Andrew N Cleland ldquoSurface codesTowards practical large-scale quantum computationrdquoPhys Rev A 86 032324 (2012)
[5] Maika Takita Andrew W Cross A D CorcolesJerry M Chow and Jay M GambettaldquoExperimental demonstration of fault-tolerantstate preparation with superconducting qubitsrdquoPhys Rev Lett 119 180501 (2017)
[6] Norbert M Linke Dmitri Maslov Martin RoettelerShantanu Debnath Caroline Figgatt Kevin A Lands-man Kenneth Wright and Christopher Monroe ldquoEx-perimental comparison of two quantum computing ar-
chitecturesrdquo Proc Nat Acad Sci 114 3305 (2017)[7] R Barends J Kelly A Megrant A Veitia
D Sank E Jeffrey T C White J Mutus A GFowler B Campbell Y Chen Z Chen B ChiaroA Dunsworth C Neill P OMalley P RoushanA Vainsencher J Wenner A N Korotkov A N Cle-land and John M Martinis ldquoSuperconducting quantumcircuits at the surface code threshold for fault tolerancerdquoNature 508 500 (2014)
[8] Steven T Flammia and Yi-Kai Liu ldquoDirect fi-delity estimation from few pauli measurementsrdquoPhys Rev Lett 106 230501 (2011)
[9] Marcus P da Silva Olivier Landon-Cardinaland David Poulin ldquoPractical characteriza-tion of quantum devices without tomographyrdquoPhys Rev Lett 107 210404 (2011)
[10] David Gross Yi-Kai Liu Steven T Flam-mia Stephen Becker and Jens Eisert ldquoQuan-tum state tomography via compressed sensingrdquoPhys Rev Lett 105 150401 (2010)
[11] Marcus Cramer Martin B Plenio Steven T Flam-mia Rolando Somma David Gross Stephen DBartlett Olivier Landon-Cardinal David Poulin andYi-Kai Liu ldquoEfficient quantum state tomographyrdquoNat Comm 1 149 (2010)
[12] Osama Moussa Marcus P da Silva Colm A Ryan andRaymond Laflamme ldquoPractical experimental certifica-tion of computational quantum gates using a twirlingprocedurerdquo Phys Rev Lett 109 070504 (2012)
[13] Dawei Lu Hang Li Denis-Alexandre Trottier JunLi Aharon Brodutch Anthony P Krismanich AhmadGhavami Gary I Dmitrienko Guilu Long JonathanBaugh and Raymond Laflamme ldquoExperimental estima-tion of average fidelity of a clifford gate on a 7-qubit quan-tum processorrdquo Phys Rev Lett 114 140505 (2015)
[14] C A Riofro D Gross S T Flammia T MonzD Nigg R Blatt and J Eisert ldquoExperimental quan-tum compressed sensing for a seven-qubit systemrdquoNat Comm 8 15305 (2017)
[15] B P Lanyon C Maier M Holzpfel T BaumgratzC Hempel P Jurcevic I Dhand A S Buyskikh A JDaley M Cramer M B Plenio R Blatt and C FRoos ldquoEfficient tomography of a quantum many-bodysystemrdquo Nature Physics XXX XXX (2017)
[16] E Knill D Leibfried R Reichle J Britton R BBlakestad J D Jost C Langer R Ozeri S Sei-delin and D J Wineland ldquoRandomized benchmarkingof quantum gatesrdquo Phys Rev A 77 012307 (2008)
[17] Easwar Magesan J M Gambetta and Joseph EmersonldquoScalable and robust randomized benchmarking of quan-tum processesrdquo Phys Rev Lett 106 180504 (2011)
[18] David C McKay Christopher J Wood SarahSheldon Jerry M Chow and Jay M Gam-betta ldquoEfficient z gates for quantum computingrdquoPhys Rev A 96 022330 (2017)
[19] J M Chow J M Gambetta L Tornberg Jens KochLev S Bishop A A Houck B R Johnson L FrunzioS M Girvin and R J Schoelkopf ldquoRandomized bench-marking and process tomography for gate errors in asolid-state qubitrdquo Phys Rev Lett 102 090502 (2009)
[20] A D Corcoles Jay M Gambetta Jerry M ChowJohn A Smolin Matthew Ware Joel Strand B L TPlourde and M Steffen ldquoProcess verification of two-qubit quantum gates by randomized benchmarkingrdquo
6
Phys Rev A 87 030301 (2013)[21] J P Gaebler T R Tan Y Lin Y Wan R Bowler
A C Keith S Glancy K Coakley E Knill D Leibfriedand D J Wineland ldquoHigh-fidelity universal gate set for9Be
+ion qubitsrdquo Phys Rev Lett 117 060505 (2016)
[22] C J Ballance T P Harty N M Linke M ASepiol and D M Lucas ldquoHigh-fidelity quan-tum logic gates using trapped-ion hyperfine qubitsrdquoPhys Rev Lett 117 060504 (2016)
[23] J P Gaebler A M Meier T R Tan R Bowler Y LinD Hanneke J D Jost J P Home E Knill D Leibfriedand D J Wineland ldquoRandomized benchmarking of mul-tiqubit gatesrdquo Phys Rev Lett 108 260503 (2012)
[24] S Olmschenk R Chicireanu K D Nelson and J VPorto ldquoRandomized benchmarking of atomic qubits inan optical latticerdquo New J Phys 12 113007 (2010)
[25] C A Ryan M Laforest and R Laflamme ldquoRan-domized benchmarking of single- and multi-qubit con-trol in liquid-state nmr quantum information process-ingrdquo New J Phys 11 013034 (2009)
[26] M Veldhorst J C C Hwang C H Yang A W Leen-stra B de Ronde J P Dehollain J T Muhonen F EHudson K M Itoh A Morello and A S DzurakldquoAn addressable quantum dot qubit with fault-tolerantcontrol-fidelityrdquo Nat Nano 9 981 (2014)
[27] Easwar Magesan Jay M Gambetta B R John-son Colm A Ryan Jerry M Chow Seth TMerkel Marcus P da Silva George A KeefeMary B Rothwell Thomas A Ohki Mark B Ketchenand M Steffen ldquoEfficient measurement of quantumgate error by interleaved randomized benchmarkingrdquoPhys Rev Lett 109 080505 (2012)
[28] David C McKay Stefan Filipp Antonio MezzacapoEaswar Magesan Jerry M Chow and Jay M Gam-betta ldquoUniversal gate for fixed-frequency qubits via atunable busrdquo Phys Rev Applied 6 064007 (2016)
[29] Joel Wallman Chris Granade Robin Harper andSteven T Flammia ldquoEstimating the coherence of noiserdquo
New J Phys 17 113020 (2015)[30] Christopher J Wood and Jay M Gambetta ldquoQuantifi-
cation and characterization of leakage errorsrdquo (2017)arxiv170403081
[31] Joel J Wallman Marie Barnhill and JosephEmerson ldquoRobust characterization of leakage errorsrdquoNew J Phys 18 043021 (2016)
[32] Jay M Gambetta A D Corcoles S T Merkel B RJohnson John A Smolin Jerry M Chow Colm A RyanChad Rigetti S Poletto Thomas A Ohki Mark BKetchen and M Steffen ldquoCharacterization of ad-dressability by simultaneous randomized benchmarkingrdquoPhys Rev Lett 109 240504 (2012)
[33] Easwar Magesan Jay M Gambetta and Joseph Emer-son ldquoCharacterizing quantum gates via randomizedbenchmarkingrdquo Phys Rev A 85 042311 (2012)
[34] Daniel Gottesman ldquoThe heisenberg representation ofquantum computersrdquo (1998) arxivquant-ph9807006
[35] Jeffrey M Epstein Andrew W Cross EaswarMagesan and Jay M Gambetta ldquoInvestigatingthe limits of randomized benchmarking protocolsrdquoPhys Rev A 89 062321 (2014)
[36] Joel J Wallman ldquoRandomized benchmarking with gate-dependent noiserdquo (2017) arxiv170309835
[37] Timothy Proctor Kenneth Rudinger Kevin Young
Mohan Sarovar and Robin Blume-Kohout ldquoWhatrandomized benchmarking actually measuresrdquoPhys Rev Lett 119 130502 (2017)
[38] Maris Ozols ldquoClifford grouprdquo (2008)[39] ldquoQISKIT SDKrdquo Online (2017)[40] F Motzoi J M Gambetta P Rebentrost
and F K Wilhelm ldquoSimple pulses for elimi-nation of leakage in weakly nonlinear qubitsrdquoPhys Rev Lett 103 110501 (2009)
[41] Sarah Sheldon Easwar Magesan Jerry M Chowand Jay M Gambetta ldquoProcedure for systematicallytuning up cross-talk in the cross-resonance gaterdquoPhys Rev A 93 060302 (2016)
6
Phys Rev A 87 030301 (2013)[21] J P Gaebler T R Tan Y Lin Y Wan R Bowler
A C Keith S Glancy K Coakley E Knill D Leibfriedand D J Wineland ldquoHigh-fidelity universal gate set for9Be
+ion qubitsrdquo Phys Rev Lett 117 060505 (2016)
[22] C J Ballance T P Harty N M Linke M ASepiol and D M Lucas ldquoHigh-fidelity quan-tum logic gates using trapped-ion hyperfine qubitsrdquoPhys Rev Lett 117 060504 (2016)
[23] J P Gaebler A M Meier T R Tan R Bowler Y LinD Hanneke J D Jost J P Home E Knill D Leibfriedand D J Wineland ldquoRandomized benchmarking of mul-tiqubit gatesrdquo Phys Rev Lett 108 260503 (2012)
[24] S Olmschenk R Chicireanu K D Nelson and J VPorto ldquoRandomized benchmarking of atomic qubits inan optical latticerdquo New J Phys 12 113007 (2010)
[25] C A Ryan M Laforest and R Laflamme ldquoRan-domized benchmarking of single- and multi-qubit con-trol in liquid-state nmr quantum information process-ingrdquo New J Phys 11 013034 (2009)
[26] M Veldhorst J C C Hwang C H Yang A W Leen-stra B de Ronde J P Dehollain J T Muhonen F EHudson K M Itoh A Morello and A S DzurakldquoAn addressable quantum dot qubit with fault-tolerantcontrol-fidelityrdquo Nat Nano 9 981 (2014)
[27] Easwar Magesan Jay M Gambetta B R John-son Colm A Ryan Jerry M Chow Seth TMerkel Marcus P da Silva George A KeefeMary B Rothwell Thomas A Ohki Mark B Ketchenand M Steffen ldquoEfficient measurement of quantumgate error by interleaved randomized benchmarkingrdquoPhys Rev Lett 109 080505 (2012)
[28] David C McKay Stefan Filipp Antonio MezzacapoEaswar Magesan Jerry M Chow and Jay M Gam-betta ldquoUniversal gate for fixed-frequency qubits via atunable busrdquo Phys Rev Applied 6 064007 (2016)
[29] Joel Wallman Chris Granade Robin Harper andSteven T Flammia ldquoEstimating the coherence of noiserdquo
New J Phys 17 113020 (2015)[30] Christopher J Wood and Jay M Gambetta ldquoQuantifi-
cation and characterization of leakage errorsrdquo (2017)arxiv170403081
[31] Joel J Wallman Marie Barnhill and JosephEmerson ldquoRobust characterization of leakage errorsrdquoNew J Phys 18 043021 (2016)
[32] Jay M Gambetta A D Corcoles S T Merkel B RJohnson John A Smolin Jerry M Chow Colm A RyanChad Rigetti S Poletto Thomas A Ohki Mark BKetchen and M Steffen ldquoCharacterization of ad-dressability by simultaneous randomized benchmarkingrdquoPhys Rev Lett 109 240504 (2012)
[33] Easwar Magesan Jay M Gambetta and Joseph Emer-son ldquoCharacterizing quantum gates via randomizedbenchmarkingrdquo Phys Rev A 85 042311 (2012)
[34] Daniel Gottesman ldquoThe heisenberg representation ofquantum computersrdquo (1998) arxivquant-ph9807006
[35] Jeffrey M Epstein Andrew W Cross EaswarMagesan and Jay M Gambetta ldquoInvestigatingthe limits of randomized benchmarking protocolsrdquoPhys Rev A 89 062321 (2014)
[36] Joel J Wallman ldquoRandomized benchmarking with gate-dependent noiserdquo (2017) arxiv170309835
[37] Timothy Proctor Kenneth Rudinger Kevin Young
Mohan Sarovar and Robin Blume-Kohout ldquoWhatrandomized benchmarking actually measuresrdquoPhys Rev Lett 119 130502 (2017)
[38] Maris Ozols ldquoClifford grouprdquo (2008)[39] ldquoQISKIT SDKrdquo Online (2017)[40] F Motzoi J M Gambetta P Rebentrost
and F K Wilhelm ldquoSimple pulses for elimi-nation of leakage in weakly nonlinear qubitsrdquoPhys Rev Lett 103 110501 (2009)
[41] Sarah Sheldon Easwar Magesan Jerry M Chowand Jay M Gambetta ldquoProcedure for systematicallytuning up cross-talk in the cross-resonance gaterdquoPhys Rev A 93 060302 (2016)
![arXiv:1308.2928v1 [quant-ph] 13 Aug 2013Investigating the limits of randomized benchmarking protocols Je rey M. Epstein, Andrew W. Cross, Easwar Magesan, and Jay M. Gambetta IBM T.J.](https://static.fdocuments.in/doc/165x107/60cbbeae249a9117593d0de4/arxiv13082928v1-quant-ph-13-aug-2013-investigating-the-limits-of-randomized.jpg)
![arXiv:1005.1279v1 [cond-mat.mes-hall] 7 May 2010 · FIG. 2. Randomized benchmarking for qubit L with ˙= 1ns using (a) Gaussian pulses, and (b) additional Gaussian deriva-tive pulses](https://static.fdocuments.in/doc/165x107/5f57df9fcf00f13b69253b82/arxiv10051279v1-cond-matmes-hall-7-may-2010-fig-2-randomized-benchmarking.jpg)
