arXiv:cond-mat/0508729v1 [cond-mat.supr-con] 30 Aug 2005 g aibeculn schemes Variable-coupling qubits (g) SCT of coupling Phase qubits charge of Coupling oscillators via qubits Coupling charge (f) of qubits coupling JJ JJ single (d) of coupling qubits Capacitive flux (c) of coupling Inductive (b) qubits two principles General for (a) schemes coupling Physical IX. detection Threshold oscillator (e) coupled SET via with Measurement qubit (d) charge of Measurement (c) measurement qubit how? Direct and (b) when why, quan- Readout: (a) of measurement information and tum read-out Qubit VIII. (PCQ) qubits qubit Potential current (e) persistent - SQUID 3-junction rf-SQUID qubits Flux (SCT) (d) Transistor pair Cooper (SCB) Single Box pair Cooper Single qubits Charge qubit (b) (JJ) junction Josephson (a) qubits Basic circuits VII. superconducting Quantum Box VI. pair Cooper Single (d) dc-SQUID (c) rf-SQUID junction (b) Josephson biased Current circuits (a) superconducting Classical systems V. qubit of oscillation Decoherence Rabi (g) and perturbation Harmonic (e) switching precession Adiabatic and (d) switching sphere sudden Bloch dc-pulses, the (c) on evolution State (b) state two-level The systems two-level (a) of Dynamics preparation IV. state and Readout (d) gates and Operations (c) entanglement and processing Qubits information (b) quantum for computation Conditions quantum (a) of Basics qubits III. and computers Nanotechnology, II. Introduction I. odcigeetia icisfrqatmifrainpr information quantum for circuits electrical conducting hspprgvsa nrdcint h hsc n principl and physics the to introduction an gives paper This uecnutn unu icis uisadComputing and Qubits Circuits, Quantum Superconducting al fcontents of Table eateto irtcnlg n aocec MC2, - Nanoscience and Microtechnology of Department hlesUiest fTechnology, of University Chalmers .Wni n ..Shumeiko V.S. and Wendin G. E426Gtebr,Sweden Gothenburg, SE-41296 Dtd eray2 2008) 2, February (Dated: References Glossary perspectives correction and error Conclusion quantum XV. and encoding transfer Qubit entanglement (d) and buses Qubit (c) multi- Teleportation (b) with measurements Bell engineering (a) systems state JJ qubit Quantum qubits phase XIV. JJ coupled Capacitively qubits (c) charge coupled coupled Inductively qubits (b) charge of coupled Capacitively (a) manipulation systems strip-line two-qubit approaches Experimental Yale a and Delft XIII. to the of coupled a Comparison (d) to qubit coupled charge-phase resonator qubit Yale flux (c) current oscillator persistent quantum Delft (b) quan- discussion General to (a) coupled qubits oscillators tum with Experiments XII. qubit Charge-phase (e) qubits qubit Flux junction (d) Josephson current-biased NIST procedures (c) measurement and Operation (b) detectors readout Readout (a) and qubits single decoherence devices with and noise Experiments environment: XI. the of Effects coupling (e) yy qubits, coupling Two xx (d) transverse qubits, Two point (c) degeneracy point the degeneracy at the Biasing from coupling away zz far transverse Biasing Ising-type qubits, Two (b) formulation N-qubit systems General (a) multi-qubit resonator of a Dynamics via X. coupled qubits Two (h) coupling capacitive Variable coupling phase Variable coupling Josephson Variable coupling inductive Variable ocessing. so prto fqatzdsuper- quantized of operation of es 2
I. INTRODUCTION tance of quantum computers on any time scale, but there is no doubt that the research will be a powerful driver of The first demonstration of oscillation of a supercon- the development of solid-state quantum state engineering ducting qubit by Nakamura et al. in 19991 can be and quantum technology, e.g. performing measurements said to represent the ”tip of the iceberg”: it rests on ”at the edge of the impossible”. a huge volume of advanced research on Josephson junc- This article aims at describing the inner workings of tions (JJ) and circuits developed during the last 25 superconducting Josephson junction (JJ) circuits, how years. Some of this work has concerned fundamen- these can form two-level systems acting as qubits, and tal research on Josephson junctions and superconduct- how they can be coupled together to multi-qubit net- ing quantum interferometers (SQUIDs) aimed at under- works. Since the field of experimental qubit applications standing macroscopic quantum coherence (MQC)2,3,4,5, is only five years old, it is not even clear if the field repre- providing the foundation of the persistent current flux sents an emerging technology for computers. Neverthe- qubit6,7,8. However, there has also been intense research less, the JJ-technology is presently the only example of a aimed at developing superconducting flux-based digital working solid state qubit with long coherence time, with electronics and computers9,10,11. Moreover, in the 1990’s demonstrated two-qubit gate operation and readout, and the single-Cooper-pair box/transistor (SCB, SCT)12, was with potential for scalability. This makes it worthwile to developed experimentally and used to demonstrate the describe this system in some detail. quantization of Cooper pairs on a small superconducting It needs to be said, however, that much of the basic island13, which is the foundation of the charge qubit1,14. theory for coupled JJ-qubits was worked out well ahead 14,41,42 Since then there has been a steady of experiment , defining and elaborating basic op- development15,16,17,18,19, with observation of microwave- eration and coupling schemes. We recommend the reader induced Rabi oscillation of the two-level populations to take a good look at the excellent research and re- 42 in charge21,22,23, and flux24,25,26,27 qubits and dc- view paper by Makhlin et al. which describes the basic pulse driven oscillation of charge qubits with rf-SET principles of a multi-JJ-qubit information processor, in- detection28. An important step is the development of the cluding essential schemes for qubit-qubit coupling. The charge-phase qubit, a hybrid version of the charge qubit ambition of the present article is to provide a both in- consisting of an SCT in a superconducting loop21,22, troductory and in-depth overview of essential Josephson demonstrating Rabi oscillations with very long coherence junction quantum circuits, discuss basic issues of readout time, of the order of 1 µs, allowing a large set of basic and measurement, and connect to the recent experimen- and advanced (”NMR-like”) one-qubit operations (gates) tal progress with JJ-based qubits for quantum informa- to be performed23. In addition, coherent oscillations tion processing (JJ-QIP). have been demonstrated in the ”simplest” JJ qubits of them all, namely a single Josephson junction30,31,32,33, or a two-JJ dc-SQUID34, where the qubit is formed by II. NANOTECHNOLOGY, COMPUTERS AND the two lowest states in the periodic potential of the JJ QUBITS itself. Although a powerful JJ-based quantum computer with The scaling down of microelectronics into the nanome- hundreds of qubits remains a distant goal, systems with ter range will inevitably make quantum effects like tun- 5-10 qubits will be built and tested by, say, 2010. Pair- neling and wave propagation important. This will even- wise coupling of qubits for two-qubit gate operations is tually impede the functioning of classical transistor com- then an essential task, and a few experiments with cou- ponents, but will also open up new opportunities for pled JJ-qubits with fixed capacitive or inductive cou- multi-terminal components and logic circuits built on e.g. plings have been reported35,36,37,38,39,40, in particular the resonant tunneling, ballistic transport, single electronics, first realization of a controlled-NOT gate with two cou- etc. pled SCBs36, used together with a one-qubit Hadamard There are two main branches of fundamentally differ- gate to generate an entangled two-qubit state. ent computer architectures, namely logically irreversible For scalability, and simple operation, the ability to con- and logically reversible. Ordinary computers are irre- trol qubit couplings, e.g. switching them on and off, will versible because they dissipate both energy and infor- be essential. So far, experiments on coupled JJ qubits mation. Even if CMOS circuits in principle only draw have been performed without direct physical control of current when switching, this is nevertheless the source the qubit coupling, but there are many proposed schemes of intense local heat generation, threatening to burn up for two(multi)-qubit gates based on fixed or controllable future processor chips. The energy dissipation can in physical qubit-qubit couplings or tunings of qubits and principle be reduced by going to single-electron devices, bus resonators. superconducting electronics and quantum devices, but All of the JJ-circuit devices introduced above are this does not alter the fact that the information process- based on nanoscale science and technology and represent ing is logically irreversible. In the simplest case of an emerging technologies for quantum engineering and, at AND gate, two incoming bit lines only result in a single best, information processing. One may debate the impor- bit output, which means that one bit is in practice erased 3
(initialized to zero) and the heat irreversibly dissipated to the environment. The use of quantum-effect devices does not change the fact that we are dealing with com- puters where each gate is logically irreversible and where discarded information constantly is erased and turned into heat. A computer with quantum device components therefore does not make a quantum computer. A quantum information processor has to be built on fundamentally reversible schemes with reversible gates where no information is discarded, and where all inter- nal processes in the components are elastic. This issue is connected with the problem of the minimum energy FIG. 1: 3-junction persistent current flux qubit (PCQ) (inner needed for performing a calculation43 (connected with loop) surrounded by a 2-junction SQUID. Courtesy of J.E. the entropy change created by erasing the final result, Mooij, TU Delft. i.e. reading the result and then clearing the register). A reversible information processor can in principle be built by classical means using adiabatically switched networks of different kinds. The principles were investigated in the 1980’s and form a background for much of the work on quantum computation44,45,46,47,48,49,50. Recently there has been some very interesting development of reversible computers (see51,52,53,54 and references therein). How- ever, a reversible computer still does not make a quan- tum computer. What is characteristic for a quantum computer is that it is reversible and quantum coherent, meaning that one can build entangled non-classical multi- FIG. 2: Single Cooper pair box (SCB) (right) coupled to a qubit states. single-electron transistor (SET) (left) for readout. Courtesy One can broadly distinguish between microscopic, of P. Delsing, Chalmers. mesoscopic and macroscopic qubits. Microscopic effec- tive two-level systems are localized systems confined by natural or artificial constraining potentials. Natural sys- EoH68. This is really an atomic-like microscopic qubit: tems then typically are atomic or molecular impurities a thin film of liquid He is made to cover a Si surface, utilizing electronic charge, or electronic or nuclear spins. and electrons are bound by the image force above the He These systems may be implanted55,56 or naturally oc- surface, forming an electronic two-level system. Qubits curring due to the material growth process57. A related are laterally defined by electrostatic gate patterns in the type of impurity qubit involves endohedral fullerenes, i.e. Si substrate, which also defines circuits for qubit control, atoms implanted into C60 or similar cages58, to be placed qubit coupling and readout. at specific positions on a surface prepared for control and Macroscopic superconducting qubits - the subject of readout. this article - are based on electrical circuits containing Mesoscopic qubit systems typically involve geomet- Josephson junctions (JJ). Looking at two extreme exam- rically defined confining potentials like quantum dots. ples, the principle is actually very simple. In one limit, Quantum dots (QD) for qubits59 are usually made in the qubit is simply represented by the two rotation direc- semiconductor materials. One type of QD is a small tions of the persistent supercurrent of Cooper pairs in a natural or artificial semiconductor grain with quantized superconducting ring containing Josephson tunnel junc- electronic levels. The electronic excitations may be ex- tions (rf-SQUID)(flux qubit)6,7,8, shown below in Fig. 1 citonic (excitons,60,61 or biexcitons,62), charge-like63, or for the Delft 3-junction flux qubit, spin-like (e.g. singlet-triplet)64,65. Another type of QD In another limit, the qubit is represented by the pres- is geometrically defined in semiconductor 2DEG by elec- ence or absence of a Cooper pair on a small supercon- trostatic split-gate arrangements. Although the host ma- ducting island (Single Cooper pair Box, SCB, or Transis- terials are epitaxially layered semiconducting materials tor, SCT) (charge qubit)1,13,28,69, as illustrated in Fig. 2. (e.g.GaAs/AlGaAs), the ungated 2DEG electronic sys- Hybrid circuits21,70,71 can in principle be tuned between tem is metallic. A split-gate arrangement can then define these limits by varing the relations between the electro- a voltage-controlled system of quantum dots coupled to static charging energy EC and the Josephson tunneling metallic reservoir electrodes, creating a system for elec- energy EJ . tron charge and spin transport through quantum point All of these solid-state qubits have advantages and dis- contacts and (effective) two-level quantum dots64,65,66,67. advantages, and only systematic research and develop- There is also a potential qubit based on liquid He ment of multi-qubit systems will show the practical and superfluid technology, namely ”Electrons on Helium”, ultimate limitations of various systems. The supercon- 4 ducting systems have presently the undisputable advan- tage of acutally existing, showing Rabi oscillations and responding to one- and two-qubit gate operations. In fact, even an elementary SCB two-qubit entangling gate creating Bell-type states has been demonstrated very recently36. All of the non-superconducting qubits are so far, promising but still potential qubits. Several of the impurity electron spin qubits show impressive relax- ation lifetimes in bulk measurements, but it remains to demonstrate how to read out individual qubit spins.
III. BASICS OF QUANTUM COMPUTATION
A. Conditions for quantum information processing
72 DiVincenzo has formulated a set of rules and con- FIG. 3: The Bloch sphere. Points on the sphere correspond ditions that need to be fulfilled in order for quantum to the quantum states |ψi; in particular, the north and south computing to be possible: poles correspond to the computational basis states |0i and 1. Register of 2-level systems (qubits), n = 2N states |1i; superposition cat-states |ψi = |0i + eiφ|1i are situated on 101..01 (N qubits) the equator. |2. Initializationi of the qubit register: e.g. setting it to 000..00 |3. Toolsi for manipulation: 1- and 2-qubit gates, e.g. can be characterised by a unit vector on the Bloch sphere: Hadamard (H) gates to flip the spin to the equator, U 0 = ( 0 + 1 )/2, and Controlled-NOT (CNOT ) The state vector can be represented as a unitary vector H | i | i | i gates to create entangled states, UCNOT UH 00 >= on the Bloch sphere, and general unitary (rotation) oper- ( 00 + 11 )/2 (Bell state) | ations make it possible to reach every point on the Bloch 4.| Read-outi | i of single qubits ψ = a 0 + beiφ 1 a, b sphere. The qubit is therefore an analogue object with (spin projection; phase φ of qubit| i lost)| i | i → a continuum of possible states. Only in the case of spin 5. Long decoherence times: > 104 2-qubit gate opera- 1/2 systems do we have a true two-level system. In the tions needed for error correction to maintain coherence general case, the qubit is represented by the lowest levels ”forever”. of a multi-level system, which means that the length of 6. Transport qubits and to transfer entanglement be- the state vector may not be conserved due to transitions tween different coherent systems (quantum-quantum in- to other levels. The first condition will therefore be to terfaces). operate the qubit so that it stays on the Bloch sphere 7. Create classical-quantum interfaces for control, read- (fidelity). Competing with normal operation, noise from out and information storage. the environment may cause fluctuation of both qubit am- plitude and phase, leading to relaxation and decoherence. It is a delicate matter to isolate the qubit from a perturb- B. Qubits and entanglement ing environment, and desirable operation and unwanted perturbation (noise) easily go hand in hand. It is a major A qubit is a two-level quantum system caracterized by issue to design qubit control and read-out such that the the state vector necessary communication lines can be blocked when not in use. θ θ ψ = cos 0 + sin eiφ 1 (3.1) The state of N independent qubits can be represented | i 2| i 2 | i as a product state, Expressing 0 and 1 in terms of the eigenvectors of the | i | i ψ = ψ1 ψ2 .... ψN = ψ1ψ2....ψN (3.4) Pauli matrix σz, | i | i| i | i | i
1 0 involving any one of all of the configurations 00...0 >, 0 = , 1 = . (3.2) 00...1 >, ...., 11...1 >. A general state of an| N-qubit | i 0 | i 1 |memory register| (i.e. a many-body system) can then this can be described as a rotation from the north pole be written as a time-dependent superposition of many- of the 0 state, particle configurations | i θ θ ψ(t) = c1(t) 0...00 + c2(t) 0...01 (3.5) 1 0 cos 2 sin 2 1 ψ = iφ θ − θ (3.3) | i | i | i | i 0 e sin cos 0 + c3(t) 0...10 + .... + cn(t) 1...11 2 2 | i | i 5 where the amplitudes ci(t) are complex, providing phase describing the time evolution of the entire N-particle information. This state represents a time-dependent su- state in the interval [t,t0]. If the total Hamiltonian com- perposition of 2N N-body configurations which in gen- mutes with itself at different times, the time ordering can eral cannot be written as a product of one-qubit states be omitted, and then represents an entangled (quantum correlated) t i ˆ ′ ′ h¯ H(t )dt many-body state. t0 U(t,t0)= e− . (3.17) In the case of two qubits, the maximally entangled R states are the so-called Bell states, This describes the time-evolution controlled by a ho- mogeneous time-dependent potential or electromagnetic ψ = ( 00 + 11 )/2 (3.6) field, e.g. dc or ac pulses with finite rise times but having | i | i | i ψ = ( 00 11 )/2 (3.7) no space-dependence. If the Hamiltonian is constant in | i | i − | i ψ = ( 01 + 10 )/2 (3.8) the interval [t0,t], then the evolution operator takes the | i | i | i simple form ψ = ( 01 10 )/2 (3.9) i | i | i − | i Hˆ (t t0) U(t,t )= e− h¯ − , (3.18) where the last one is the singlet state. In the case 0 of three qubits, the corresponding maximally entan- describing the time-evolution controlled by square dc gled (”cat”) states are the Greenberger-Horne-Zeilinger pulses. (GHZ) states73 The time-development will depend on how many terms are switched on in the Hamiltonian during this time in- ψ = ( 000 111 )/√8 (3.10) terval. In the ideal case, usually not realizable, all terms | i | i ± | i are switched off except for those selected for the specific Another interesting entangled three-qubit state appears computational step. A single qubit gate operation then in the teleportation process, involves turning on a particular term in the Hamiltonian for a specific qubit, while a two-qubit gate involves turn- ψ = [ 00 (a 0 + b 1 )+ 01 (a 0 b 1 ) (3.11) | i | i | i | i | i | i− | i ing on an interaction term between two specific qubits. + 10 (b 0 + a 1 )+ 11 (b 0 a 1 )]/√8 In principle one can perform an N-qubit gate operation | i | i | i | i | i− | i by turning on interactions for all N qubits. In practical cases, many terms in the Hamiltonian are turned on all C. Operations and gates the time, leading to a ”background” time development that has to be taken into account. Quantum computation basically means allowing the The basic model for a two-level qubit is the spin-1/2 in N-body state to develop in a fully coherent fashion a magnetic field. A system of interacting qubits can then through unitary transformations acting on all N qubits. be modelled by a collection of interacting spins, described The time evolution of the many-body system of N two- by the Heisenberg Hamiltonian level subsystems can be described by the Schr¨odinger 1 equation for the N-level state vector ψ(t) , Hˆ (t)= hi(t) Si + Jij (t) Si Sj (3.19) | i 2 X X i¯h∂ ψ = Hˆ ψ . (3.12) controlled by a time-dependent external magnetic field t| i | i hi(t) and by a time-dependent spin-spin coupling Jij (t). in terms of the time-evolution operator characterizing by Expressing the Hamiltonian in Cartesian components, ˆ the time-dependent many-body Hamiltonian H(t) of the hi(t) = (hx,hy,hz), Si(t) = (Sx,Sy,Sz) and introducing system determined by the external control operations and 1 the Pauli σ-matrices, (Sx,Sy,Sz) = 2 ¯h(σx, σy, σz) we the perturbing noise from the environment, obtain a general N-qubit Hamiltonian with general qubit- qubit coupling: ψ(t) = U(t,t0) ψ(t0) . (3.13) | i | i 1 Hˆ = (ǫiσzi + Re∆iσxi + Im∆iσyi) (3.20) The solution of Schr¨odinger equation for U(t,t0) −2 i X ˆ 1 i¯h∂tU(t,t0)= HU(t,t0) (3.14) + λ (t) σ σ 2 ν,ij νi νj ij;ν may be written as X + (fi(t)σzi + gxi(t)σxi + gyi(t)σyi) i t i U(t,t )=1 Hˆ (t′)U(t′,t )dt′ , (3.15) X 0 − ¯h 0 Zt0 We have here introduced time-independent components of the external field defining qubit energy level splittings and finally, in terms of the time-ordering operator T, as ǫi, Re∆i and Im∆i along the z and x,y axes, as well i t ′ ′ as time-dependent components f (t) and g (t) explicitly Hˆ (t )dt i i h¯ t U(t,t0)= T e− 0 , (3.16) describing qubit operation and readout signals and noise. R 6
Inserted into Eq.(3.17), this Hamiltonian determines to the operation fields. This also opens up the system the time evolution of the many-qubit state. In the ideal to a noisy environment, which puts great demands on case one can turn on and off each individual term of the signal-to-noise ratios. Hamiltonian, including the two-body interaction, giving The readout operation is a particularly critical step. complete control of the evolution of the state. The ultimate purpose is to perform a ”single-shot” quan- It has been shown that any unitary transformation can tum non-demolition (QND) measurement, determining be achieved through a quantum network of sequential the state ( 0 or 1 ) of the qubit in a single measure- application of one- and two-qubit gates. Moreover, the ment and then| i leaving| i the qubit in that very state. Dur- size of the coherent workspace of the multi-qubit memory ing the measurement time, the back-action noise from the can be varied (in principle) by switching on and off qubit- ”meter” will cause relaxation and mixing, changing the qubit interactions. qubit state. The ”meter” must then be sensitive enough In the common case of NMR applied to to detect the qubit state on a time scale shorter than the 74,75,76 molecules , one has no control of the fixed, induced relaxation time T1∗ in the presence of the dissipa- direct spin-spin coupling. With an external magnetic tive back-action of the ”meter” itself. Under these con- field one can control the qubit Zeeman level splittings ditions it is possible to detect the qubit projection in a (no control of individual qubits). Individual qubits can single measurement (single shot read-out). Performing a be addressed by external RF-fields since the qubits have single-shot projective measurement in the qubit eigenba- different resonance frequencies (due to different chemical sis then provides a QND measurement (note the ”trivial” environments in a molecule). Two-qubit coupling can fact that the phase is irreversibly lost in the measurement be induced by simultaneous resonant excitation of two process under any circumstances). qubits. The readout/measurement processes described above In the case of engineered solid state JJ-circuits, indi- can be related to the Stern-Gerlach (SG) experiment80. vidual qubits can be addressed by local gates, control- A SG spin filter acts as a beam splitter for flying qubits, ling the local electric or magnetic field. Extensive single- creating separate paths for spin-up and spin-down atoms, qubit operation has recently been demonstrated by Collin preparing for the measurement by making it possible in et al.23. Regarding two-qubit coupling, the field is just principle to distinguish spatially between the two states starting up, and different options are only beginning to of the qubit. The measurement is then performed by be tested. The most straightforward approaches involve particle counters: a click in, say, the spin-up path col- fixed capacitive35,36,40 or inductive37,38 coupling; such lapses the atom to the spin-up state in a single shot. systems could be operated in NMR style, by detuning There is essentially no decohering back-action from the specific qubit pairs into resonance. Moreover, there are detector until the atom is detected. However, after de- various solutions for controlling the direct physical qubit- tection of the spin-up atom the state is thoroughly de- qubit coupling strength, as will be described in Section stroyed: this qubit has not only decohered and relaxed, IX. but is removed from the system. However, if it was entan- gled with other qubits, these are left in a specific eigen- state. For example, if the qubit was part of the Bell pair D. Readout and state preparation ψ = ( 00 + 11 )/2, detection of spin up ( 0 ) selects |00i and| leavesi | thei other qubit in state 0 . If instead| i the |qubiti was part of the Bell pair ψ = ( |01i + 10 )/2, de- External perturbations, described by fi(t) and gi(t) in | i | i | i Eq.(3.20) can influence the two-level system in typically tection of spin up ( 0 ) selects 01 and leaves the other qubit in the spin-down| i state 1| .i Furthermore, for the two ways: (i) shifting the individual energy levels, which | i 3-qubit GHZ state ψ = ( 000 111 )/√8, detection may change the transition energy and the phase of the | i | i ± | i qubit; and (ii) inducing transitions between the levels, of the first qubit in the spin up ( 0 ) state leaves the remaining 2-qubit system in the 00| istate. changing the level populations. These effects arise both | i from desirable control operations and from unwanted Finally, in the case of the three-qubit entangled state noise (see42,77,78,79 for a discussion of superconducting in the teleportation process, Eq.(3.11), detection of the circuits, and Grangier et al.80 for a discussion of state first qubit in the spin up state 0 , leaves the remaining | i preparation and quantum non-demolition (QND) mea- 2-qubit system in the entangled state [ 0 (a 0 + b 1 )+ | i | i | i surements). 1 (a 0 b 1 )]/2. Moreover, detection of also the second qubit| i | ini− the| i spin-up state 0 will leave the third qubit To control a qubit register, the important thing is to | i control the decoherence during qubit operation and read- in the state (a 0 + b 1 )/√2. In this way, measurement | i | i out, and in the memory state. In the qubit memory state, can be used for state preparation. the qubit must be isolated from the environment. Op- Applying this discussion to the measurement and read- eration and read-out devices should be decoupled from out of JJ-qubit circuits, an obvious difference is that the the qubit and, ideally, not cause any dephasing or relax- JJ-qubits are not flying particles. The detection can ation. The resulting intrinsic qubit life times should be there not be turned on simply by the qubit flying into long compared to the duration of the calculation. In the the detector. Instead, the detector is part of the JJ-qubit operation and readout state, the qubit must be connected circuitry, and must be turned on to discriminate between 7 the two qubit states. These are not spatially separated and project onto the basis states and therefore sensitive to level mixing by detector noise with frequency around the qubit transition energy. The Hˆ m m ψ = E ψ (4.4) | ih | i | i sensitivity of the detector determines the time scale T m m X of the measurement, and the detector-on back-action de- obtaining the ususal matrix equation termines the time scale of qubit relaxation T1∗. Clearly T T ∗ is needed for single-shot discrimination of 0 m ≪ 1 | i k Hˆ m a = Ea (4.5) and 1 . This requires a detector signal-to-noise (S/N) h | | i m k | i m ratio 1, in which case one will have the possibility also X to turn≫ off the detector and leave the qubit in the deter- where ak = k ψ , and mined eigenstate, now relaxing on the much longer time h | i scale T1 of the ”isolated” qubit. This would then be the ultimate QND measurement. ǫ ∆ Hqp = k σz m k σx m (4.6) Note that in the above discussion, the qubit dephas- −2h | | i− 2 h | | i ing time (”isolated” qubit) does not enter because it is giving the Hamiltonian matrix assumed to be much longer than the measurement time, T T . This condition is obviously essential for uti- m ≪ φ lizing the measurement process for state preparation and 1 ǫ ∆ Hˆ = (4.7) error correction. −2 ∆ ǫ − The Schr¨odinger equation is then given by IV. DYNAMICS OF TWO-LEVEL SYSTEMS 1 ǫ +2E ∆ a (Hˆ E) ψ = 1 =0 − | i −2 ∆ ǫ +2E a2 − To perform computational tasks one must be able to (4.8) put a qubit in an arbitrary state. This is usually done The eigenvalues are determined by in two steps. The first step, initialization, consists of relaxation of the initial qubit state to the equilibrium 1 det(Hˆ E)= E2 (ǫ2 + ∆2) = 0 (4.9) state due to interaction with environment. At low tem- − − 4 perature, this state is close to the ground state. During the next step, time dependent dc- or rf-pulses are ap- with the result plied to the controlling gates: electrostatic gate in the 1 E = ǫ2 + ∆2 (4.10) case of charge qubits, bias flux in the case of flux qubits, 1,2 ±2 and bias current in the case of the JJ qubit. Formally, p the pulses enter as time-dependent contributions to the The eigenvectors are given by: Hamiltonian and the state evolves under the action of the time-evolution operator. To study the dynamics of a sin- ∆ gle qubit two-level system we therefore first describe the a2 = a1 (4.11) two-level state, and then the evolution of this state under − ǫ +2E the influence of the control pulses (”perturbations”). After normalisation
A. The two-level state ∆ 2 1 ǫ a1 =1/ 1+ = 1 (4.12) s ǫ +2E √2 ± 2E r The general 1-qubit Hamiltonian has the form | |
1 2 1 ǫ Hˆ = (ǫ σz + ∆ σx) (4.1) a2 = 1 a = 1 (4.13) −2 ± − 1 ±√2 ∓ 2E q r | | The qubit eigenstates are to be found from the stationary We finally simplify the notation by fixing the signs of Schr¨odinger equation the amplitudes,
Hˆ ψ = E ψ (4.2) 1 ǫ 1 ǫ | i | i a1 = 1+ ; a2 = 1 ; (4.14) √2 2E √2 − 2E To solve the S-equation we expand the 1-qubit state in a r | | r | | complete basis, e.g. the basis states of the σz operator, and explicitly writing down all the energy eigenstates,
E1 = a1 0 + a2 1 (4.15) ψ = ak k = c0 0 + c1 1 (4.3) | i | i | i | i | i | i | i E2 = a2 0 a1 1 (4.16) Xk | i | i− | i 8 where giving a 3-vector representation for the Hamiltonian,
1 2 2 1 2 2 E = ǫ + ∆ ; E =+ ǫ + ∆ (4.17) H = (Hx, Hy, Hz). (4.25) 1 −2 1 1 2 2 1 1 q q shown in Fig. 4. The sign of E2 has been chosen to give the familiar expression for| thei superposition at the degeneracy point ǫ = 0 where a = a =1/√2, 0 1 | 1| | 2| H 1 E1 = ( 0 + 1 ) (4.18) | i √2 | i | i 1 E2 = ( 0 1 ) (4.19) | i √2 | i − | i ρ
B. The state evolution on the Bloch sphere
To study the time-evolution of a general state, a conve- nient way is to expand in the basis of energy eigenstates,
iE1t iE2t ψ(t) = c E e− + c E e− (4.20) | i 1| 1i 2| 2i If we know the coefficients at t = 0, then we know the time evolution. On the Bloch sphere this time evolution 1 is represented by rotation of the Bloch vector with con- stant angular speed (E1 E2)/¯h around the direction defined by the energy eigenbasis.− Indeed, by introduc- FIG. 4: The Bloch sphere: the Bloch vector ρ represents the ing parameterization, c = cos θ , c = sin θ eiφ , we see states of the two-level system (same as in Fig. 3). The poles 1 ′ 2 ′ ′ of the Bloch sphere correspond to the energy eigenstates; the that according to Eq. (4.20) the polar angle remains vector H represents the two-level Hamiltonian. constant, θ′ = const, while the azimuthal angle grows, φ′(t)φ′(0) + (E1 E2)t/¯h. The primed angles here refer to a new coordinate− system on the Bloch sphere related to The time evolution of the density matrix is given by the energy eigenbasis, which is obtained by rotation from the Liouville equation, the earlier introduced computational basis, Eq. (3.1). i¯h∂ ρˆ = [H,ˆ ρˆ]. (4.26) The dynamics on the Bloch sphere is conveniently de- t scribed in terms of the density matrix for a pure quantum 81 The vector form of the Liouville equation is readily de- state , rived by inserting Eqs.(4.24),(4.25) and using the com- ρˆ = ψ ψ . (4.21) mutation relations among the Pauli matrices, | ih | 1 This is a 2 2 Hermitian matrix whose diagonal ele- ∂ ρ = [ H ρ ]. (4.27) × t ¯h ments ρ1 and ρ2 define occupation probabilities of the × basis states, hence satisfying the normalization condition This equation coincides with the Bloch equation for a ρ1 +ρ2 = 1, while the off-diagonal elements give informa- magnetic moment evolving in a magnetic field, the role tion about the phase. The density matrix can be mapped of the magnetic moment being played by the Bloch vector on a real 3-vector by means of the standard expansion in ρ which rotates around the effective ”magnetic field” H terms of σ-matrices, associated with the Hamiltonian of the qubit (plus any 1 driving fields) (Fig. 4). ρˆ = (1 + ρ σ + ρ σ + ρ σ ). (4.22) 2 x x y y z z Direct calculation of the density matrix Eq.(4.21) using C. dc-pulses, sudden switching and free precession Eq.(3.1) and comparing with Eq.(4.22) shows that the vector ρ = (ρx,ρy,ρz) coincides with the Bloch vector, To control the dynamics of the qubit system, one method is to apply dc (square) pulses which suddenly ρ = (sin θ cos φ, sin θ sin φ, cos θ) (4.23) change the Hamiltonian and, consequently, the time- introduced in Fig. 3 and also shown in Fig. 4. In the evolution operator. Sudden pulse switching means that same σ-matrix basis, the general two-level Hamiltonian the time-dependent Hamiltonian is changed so fast on takes the form the time scale of the evolution of the state vector that the state vector can be treated as time-independent - Hˆ = (Hxσx + Hyσy + Hzσz). (4.24) frozen - during the switching time interval. This implies 9 that the system is excited by a Fourier spectrum with an upper cut-off given by the inverse of the switching time. H In the specific scheme of sudden switching of different terms in the Hamiltonian using dc-pulses, the initial state ρ is frozen during the switching event, and begins to evolve in time under the influence of the new H ρ 0 = ψ(0) = c E + c E (4.28) | i | i 1| 1i 2| 2i To find the coefficients we project onto the charge basis, a) b) k=0,1 H H 0 0 = c 0 E + c 0 E (4.29) h | i 1h | 1i 2h | 2i 1 0 = c 1 E + c 1 E (4.30) h | i 1h | 1i 2h | 2i and use the explict results for the energy eigenstates to calculate the matrix elements, obtaining ρ ρ
1= c1a1 + c2a2 (4.31)
0= c1a2 c2a1 (4.32) c) d) − As a result, FIG. 5: Qubit operations with dc-pulses: the vector H repre- sents the qubit Hamiltonian, and the vector ρ represents the 0 = a E + a E (4.33) | i 1| 1i 2| 2i qubit state. a) The qubit is initialized to the ground state; b) the Hamiltonian vector H is suddenly rotated towards x- This stationary state then develops in time governed by axis, and the qubit state vector ρ starts to precess around the constant Hamiltonian as H; c) when qubit vector reaches the south pole of the Bloch sphere, the Hamiltonian vector H is switched back to the iE1t iE2t ψ(t) = a1e− E1 + a2e− E2 (4.34) initial position; the vector ρ remains at the south pole, indi- | i | i | i cating complete inversion of the level population (π-pulse); d) Inserting the energy eigenstates we finally obtain the time if the Hamiltonian vector H is switched back when the qubit evolution in the charge basis, vector reaches the equator of the Bloch sphere (π/2-pulse), then the ρ vector remains precessing at the equator, repre- senting equal-weighted superposition of the qubit states (cat ψ(t) = iφ 2 iE1t 2 iE1t iE1t | iEi1t states) |ψi = |0i + e |1i; this operation is the basis for the 0 a e− + a e + 1 a a (e− e ) | i 1 2 | i 1 2 − Hadamard gate. (4.35)
The probability amplitudes of finding the system in one Let us consider, for example, the diagonal qubit Hamil- of the two charge states is then tonian, Hˆ = (ǫ/2)σz, and apply a pulse δǫ during a
2 iE1t 2 iE1t time τ. This operation will shift phases of the qubit 0 ψ(t) = a e− + a e (4.36) h | i 1 2 eigenstates by, δǫτ/2¯h. If the applied pulse is such iE1t iE1t ± 1 ψ(t) = a1a2(e− e ) (4.37) that ǫ is switched off, and instead, the σx component, h | i − ∆, is switched on, Fig 5, then the state vector will ro- If the system is driven to the degeneracy point ǫ1 = 0, tate around the x-axis, and after the time ∆τ/2¯h = π where a1 = a2 =1/√2, then (π-pulse) the ground state, + , will flip and become, | | | | + . This manipulation| correspondsi to the quan- 0 ψ(t) = cos E t (4.38) | i → |−i h | i 1 tum NOT operation. Furthermore, if the pulse duration 1 ψ(t) = sin E1t (4.39) is twice smaller (π/2-pulse), then the ground state vector h | i will approach the equator of the Bloch sphere and pre- In particular, the probability of finding the system in cess along it after the end of the operation. Such state is state 1 (level 2) oscillates like an equal-weighted superposition of the basis states (cat | i state). p (t)= 1 ψ(t) = sin2 E t 2 |h | i| 1 1 = [1 cos(E2 E1)t] (4.40) 2 − − D. Adiabatic switching with the frequency of the interlevel distance. On the Bloch sphere, this describes free precession around the Adiabatic switching represents the opposite limit to X-axis. sudden switching, namely that the state develops so fast 10 on the time scale of the Hamiltonian that this can be Thus, the dynamics of a driven qubit is characterized by regarded as ”frozen” , i.e. the time-dependence of the a linear combination of the two wave functions, Hamiltonian becomes parametric. This implies that en- ergy is conserved and no transitions are induced - the (1) 1 iλxt/2¯h iE1t/h¯ iE2t/h¯ ψ = e− e− E1 + e− E2 , system stays in the same energy level (although the state | i √2 | i | i changes). (2) 1 iλxt/2¯h iE1t/h¯ iE2t/h¯ ψ = e e− E1 e− E2 . | i √2 | i− | i E. Harmonic perturbation and Rabi oscillation (4.48) Let us assume that the qubit was initially in the ground A particularly interesting and practically important state, E1 , and that the perturbation was switched on case concerns harmonic perturbation with small ampli- instantly.| i Then the wave function of the driven qubit tude λ and resonant frequencyhω ¯ = E2 E1. Let us will take the form, consider the situation when the harmonic− perturbation is added to the z-component of the Hamiltonian corre- λxt iE1t/h¯ λxt iE2t/h¯ ψ = cos e− E + i sin e− E . sponding to a modulation of the qubit bias with a mi- | i 2¯h | 1i 2¯h | 2i crowave field. (4.49) In the eigenbasis of the non-perturbed qubit, E1 , Correspondingly, the probabilities of the level occupa- | i E2 , the Hamiltonian will take the form, tions will oscillate in time, | i ˆ H = E1σz + cos ωt (λz σz + λx σx) , (4.41) λ t 1 λ t P = cos2 x = 1+cos x , 1 2¯h 2 ¯h ǫ ∆ λ = λ , λ = λ . (4.42) 2 λxt 1 λxt z x P2 = sin = 1 cos , (4.50) E2 E2 2¯h 2 − ¯h The first perturbative term determines small periodic os- with small frequency Ω = λ /¯h ω, Rabi oscillations, cillations of the qubit energy splitting, while the second R x ≪ term will induce interlevel transitions. Despite the am- illustrated in Fig. 6. plitude of the perturbation being small, λ/E2 1, the system will be driven far away from the initial≪ state be- cause of the resonance. Indeed, let us consider the wave 1 function of the driven qubit on the form
iE1t/h¯ iE2t/h¯ ψ = a(t)e− E + b(t)e− E . (4.43) | i | 1i | 2i Substituting this ansatz into the Schr¨odinger equation, i¯h ψ˙ = Hˆ (t) ψ , (4.44) | i | i we get the following equations for the coefficients, 0 i(E1 E2)t/h¯ t i¯ha˙ = λx cos ωte − b, T i(E2 E1)t/h¯ R i¯hb˙ = λx cos ωte − a. (4.45) (Here we have neglected a small diagonal perturbation, FIG. 6: Rabi oscillation of populations of lower level (full line) and upper level (dashed line) at exact resonance (zero λz.) detuning). TR = 2π/ΩR is the period of Rabi oscillations. Let us now focus on the slow evolution of the coeffi- cients on the time scale of qubit precession, and average Eqs.(4.45) over the period of the precession. This approx- imation is known in the theory of two-level systems as the ”rotating wave approximation (RWA)”. Then, taking F. Decoherence of qubit systems into account the resonance condition, we get the simple equations, Descriptions of the qubit dynamics in terms of the den- sity matrix and Liouville equation are more general than λx λx i¯ha˙ = b, i¯hb˙ a, (4.46) descriptions in terms of the wave function, allowing the 2 2 effects of dissipation to be included. The density ma- whose solutions read, trix defined in Section IV B for a pure quantum state 2 (1) (1) iλxt/2¯h possesses the projector operator property,ρ ˆ =ρ ˆ. This a (t) = b (t)= e− , assumption can be lifted, and then the density matrix a(2)(t) = b(2)(t)= eiλxt/2¯h. (4.47) describes a statistical mixture of pure states, say energy − 11 eigenstates, processes, relaxation and dephasing, are referred to as decoherence. ρˆ = ρ E E . (4.51) Coupling to the environment leads, in the simplest i| iih i| i case, to the following modification of the Liouville X equation83,84, The density matrix acquires off-diagonal elements when the basis rotates away from the energy eigenbasis. Such 1 (0) a mixed state cannot be represented by the vector on the ∂tρz = (ρz ρz ), (4.55) −T1 − Bloch sphere; however, its evolution is still described by the Louville equation (4.26), i 1 ˆ ∂tρ12 = Eρ12 ρ12 . (4.56) i¯h∂tρˆ = [H, ρˆ]. (4.52) ¯h − T2 The density matrix in Eq. (4.51) is the stationary solu- This equation is known as the Bloch-Redfield equa- tion of the Liouville equation. The evolution of an arbi- tion. The first equation describes relaxation of the (0) trary density matrix, off-diagonal in the energy eigenba- level population to the equilibrium form, ρz = sis, is given by the equations, (1/2) tanh(E/2kT ), T being the relaxation time. The − 1 i(E1 E2)t/h¯ second equation describes disappearance of the off- ρ1,ρ2 =const, ρ12 e − . (4.53) ∝ diagonal matrix element during characteristic time T2, Dissipation is included in the density matrix descrip- dephasing. tion by extending the qubit Hamiltonian and including The relaxation time is determined by the spectral den- interaction with an environment. The environment for sity of the environmental fluctuations at the qubit fre- macroscopic superconducting qubits basically consists of quency, various dissipative elements in external circuits which 1 λ2 provide bias, control, and measurement of the qubit. The = ⊥ Sφ(ω = E). (4.57) ”off-chip” parts of these circuits are usually kept at room T1 2 temperature and produce significant noise. Examples are The particular form of the spectral density depends on the fluctuations in the current source producing magnetic the properties of the environment, which are frequently field to bias flux qubits and, similarly, fluctuations of the expressed via the impedance (response function) of the voltage source to bias gate of the charge qubits. Elec- environment. The most common environment consists tromagnetic radiation from the qubit during operation is of a pure resistance, in this case, Sφ(ω) ω, at low another dissipative mechanisms. There are also intrinsic frequencies. ∝ microscopic mechanisms of decoherence, such as fluctuat- The dephasing time consists of two parts, ing trapped charges in the substrate of the charge qubits, and fluctuating trapped magnetic flux in the flux qubits, 1 1 1 = + . (4.58) believed to produce dangerous 1/f noise. Another intrin- T2 2T1 Tφ sic mechanism is possibly the losses in the tunnel junction dielectric layer. Various kinds of environment are com- The first part is generated by the relaxation process, monly modelled with an infinite set of linear oscillators while the second part results from the pure dephasing due in thermal equilibrium (thermal bath), linearly coupled to the longitudinal coupling to the environment. This to the qubit (Caldeira-Leggett model2,3). The extended pure dephasing part is proportional to the spectral den- qubit-plus-environment Hamiltonian has the form in the sity of the fluctuation at zero frequency. qubit energy eigenbasis82, 1 λ2 = z S (ω = 0). (4.59) 1 T 2 φ Hˆ = Eσz + (λiz σz + λi σ )Xi φ −2 ⊥ ⊥ i X There is already a vast recent literature on de- Pˆ2 mω2X2 coherence and noise in superconducting circuits, i + i i , (4.54) 2m 2 qubits and detectors, and how to engineer the i ! X qubits and environment to minimize decoherence and 20,42,86,87,88,89,90,91,92,93,94,95,96,97,98,99,100,101,102,103,104,105,106 where E = E1 E2. The physical effects of the two relaxation coupling terms in− Eq. (4.54) are quite different. The Many of these issues will be at the focus of this article. ”transverse” coupling term proportional to λ induces interlevel transitions and eventually leads to⊥ the relax- ation. The ”longitudinal” coupling term proportional to V. CLASSICAL SUPERCONDUCTING CIRCUITS λz commutes with the qubit Hamiltonian and thus does not induce interlevel transitions. However, it randomly changes the level spacing, which eventually leads to the In this section we describe a number of elementary su- loss of phase coherence, dephasing. The effect of both perconducting circuits with tunnel Josephson junctions, 12
U(φ) current through the Josephson element has the form85,
IJ = Ic sin φ, (5.3)
where Ic is the critical Josephson current, i.e. I the maximum non-dissipative current that may flow J CR through the junction. The microscopic theory of superconductivity111,112,113 gives the following equation for the Josephson current, φ π∆ ∆ Ic = tanh , (5.4) FIG. 7: Current-biased Josephson junction (JJ) (left), equiv- 2eRN 2T alent circuit (center), and effective (washboard-like) potential where ∆ is the superconducting order parameter, and T (right). The superconducting leads are indicated with dark color, and the tunnel junction with light color. is the temperature. Using these relations and expressing voltage through the superconducting phase, we can write down Kirchhoff’s rule for the circuit, which are used as building blocks in qubit applications. ¯h ¯h These basic circuits are: single current biased Joseph- Cφ¨ + φ˙ + Ic sin φ = Ie, (5.5) 2e 2eR son junction; single Josephson junction (JJ) included in a superconducting loop (rf SQUID); two Josephson junc- where Ie is the bias current. This equation describes the tions included in a superconducting loop (dc SQUID); dynamics of the phase, and it has the form of a damped and an ultra-small superconducting island connected to non-linear oscillator. The role of the non-linear induc- a massive superconducting electrode via tunnel Joseph- tance is here played by the Josephson element. son junction (Single Cooper pair Box, SCB). The dissipation determines the qubit lifetime, and therefore circuits suitable for qubit applications must have extremely small dissipation. Let us assume zero A. Current biased Josephson junction level of the dissipation, dropping the resistive term in Eq. (5.5). Then the circuit dynamic equations, using the The simplest superconducting circuit, shown in Fig. 7, mechanical analogy, can be presented in the Lagrangian consists of a tunnel junction with superconducting elec- form, and, equivalently, in the Hamiltonian form. The trodes, a tunnel Josephson junction, connected to a cur- circuit Lagrangian consists of the difference between the rent source. An equivalent electrical circuit, which repre- kinetic and potential energies, the electrostatic energy of sents the junction consists of the three lumped elements the junction capacitors playing the role of kinetic energy, connected in parallel: the junction capacitance C, the while the energy of the Josephson current plays the role junction resistance R, which generally differs from the of potential energy. normal junction resistance RN and strongly depends on The kinetic energy corresponding to the first term in temperature and applied voltage, and the Josephson ele- the Kirchhoff equation (5.5) reads, ment associated with the tunneling through the junction. ¯h 2 Cφ˙2 K(φ˙)= . (5.6) The current-voltage relations for the junction ca- 2e 2 pacitance and resistance have standard forms, IC = This energy is equal to the electrostatic energy of the C (dV/dt), and IR = V/R. To write down a similar rela- 2 tion for the Josephson element, it is necessary to intro- junction capacitor, CV /2. It is convenient to introduce duce the superconducting phase difference φ(t) across the the charging energy of the junction capacitor charged junction, often simply referred to as the superconducting with one electron pair (Cooper pair), phase, which is related to the voltage drop across the (2e)2 junction, E = , (5.7) C 2C 2e φ(t)= V dt + φ, (5.1) in which case Eq. (5.6) takes the form ¯h Z ¯h2φ˙2 where φ is the time-independent part of the phase dif- K(φ˙)= . (5.8) ference. The phase difference can be also related to a 4EC magnetic flux, The potential energy corresponds to the last two terms in Eq. (5.5), and consists of the energy of the Josephson 2e Φ φ = Φ=2π , (5.2) current, and the magnetic energy of the bias current, ¯h Φ0 ¯h U(φ)= E (1 cos φ) I φ, (5.9) where Φ0 = h/2e is the magnetic flux quantum. The J − − 2e e 13 where EJ =h/ ¯ 2e Ic is the Josephson energy. This po- L tential energy has a form of a washboard (see Fig. 7). In the absence of bias current this potential corresponds to Φ a pendulum with the frequency of small-amplitude oscil- C lations given by J
2eIc R ωJ = . (5.10) r ¯hC This frequency is known as the plasma frequency of the FIG. 8: Superconducting quantum interference device - Josephson junction. When current bias is applied, the SQUID (left) consists of a superconducting loop (dark) in- pendulum potential becomes tilted, its minima becom- terrupted by a tunnel junction (light); magnetic flux Φ is sent through the loop. Right: equivalent circuit. ing more shallow, and finally disappearing when the bias current becomes equal to the critical current, Ie = IC . At this point, the plasma oscillations become unstable, U(φ) which physically corresponds to switching to the dissipa- tive regime and the voltage state. Now we are ready to write down the Lagrangian for the circuit, which is the difference between the kinetic and potential energies. Combining Eqs. (5.8) and (5.9), we get,
2 ˙2 ˙ ¯h φ ¯h L(φ, φ)= EJ (1 cos φ)+ Ieφ. (5.11) φ 4EC − − 2e 0
It is straightforward to check that the Kirchhoff equation FIG. 9: SQUID potential: the full (dark) curve corresponds (5.5) coincides with the dynamic equation following from to integer bias flux (in units of flux quanta), while the dashed the Lagrangian (5.11), using (light) curve corresponds to half-integer bias flux.
d ∂L ∂L =0. (5.12) dt ∂φ˙ − ∂φ where Φe is the external magnetic flux threading the SQUID loop. The Kirchhoff rule for this circuit takes It is important to emphasize, that the resistance of the form the junction can only be neglected for low temperatures, ¯h ¨ ¯h ˙ ¯h and also only for slow time evolution of the phase; both Cφ + φ + Ic sin φ + (φ φe)=0. (5.14) 2e 2eR 2eL − the temperature and the characteristic frequency must be small compared to the magnitude of the energy gap in the While neglecting the Josephson tunneling (Ic = 0), this superconductor: T, ¯hω ∆. The physical reason behind equation describes a damped linear oscillator of a con- this constraint concerns≪ the amount of generated quasi- ventional LC-circuit. The resonant frequency is then, particle excitations in the system: if the constraint is 1 fulfilled, the amount of equilibrium and non-equilibrium ω = , (5.15) LC √ excitations will be exponentially small. Otherwise, the LC gap in the spectrum will not play any significant role, and the (weak) damping is γ =1/RC. dissipation becomes large, and the advantage of the su- In the absence of dissipation, it is straightforward to perconducting state compared to the normal conducting write down the Lagrangian of the rf-SQUID, state will be lost. 2 2 2 ¯h φ˙ (φ φe) L(φ, φ˙)= EJ (1 cos φ) EL − . (5.16) 4EC − − − 2 B. rf-SQUID The last term in this equation corresponds to the energy of the persistent current circulating in the loop, The rf-SQUID is the next important superconducting circuit. It consists of a tunnel Josephson junction in- Φ2 E = 0 . (5.17) serted in a superconducting loop, as illustrated in Fig. 8. L 4π2L This circuit realizes magnetic flux bias for the Josephson junction113. To describe this circuit, we introduce the The potential energy U(φ) corresponding to the last two current associated with the inductance L of the leads, terms in Eq. (5.16) is schetched in Fig. 9. For bias flux equal to integer number of flux quanta, ¯h 2e or φe2πn, the potential energy of the SQUID has one ab- I = (φ φ ), φ = Φ (5.13) L 2eL − e e ¯h e solute minimum at φ = φe. For half integer flux quanta 14
Cg
φ φ 1 Φ 2 I Vg
FIG. 11: Single Cooper pair box (SCB): a small supercon- FIG. 10: dc SQUID consists of two tunnel junctions included ducting island connected to a bulk superconductor via a tun- in a superconducting loop; arrows indicate the direction of nel junction; the island potential is controlled by the gate the positive Josephson current. voltage Vg. the potential energy has two degenerate minima, which and correspond to the two persistent current states circulat- I I φ ing in the SQUID loop in the opposite directions. This tan α = c1 − c2 tan e . (5.22) configuration of the potential energy provides the basis Ic1 + Ic2 2 for constructing a persistent-current flux qubit (PCQ). For a symmetric SQUID with Ic1 = Ic2, giving α = 0, Eq. (5.20) reduces to the form C. dc SQUID 2Ic cos(φe/2) sin φ = Ie. (5.23) − We now consider consider the circuit shown in Fig. 10 The potential energy generated by Eq. (5.20) has the consisting of two Josephson junctions coupled in paral- form, lel to a current source. The new physical feature here, compared to a single current-biased junction, is the de- φe ¯h U(φ)=2EJ cos (1 cos φ ) Ieφ , (5.24) pendence of the effective Josephson energy of the double 2 − − − 2e − junction on the magnetic flux threading the SQUID loop. which indeed is similar to the potential energy of a sin- Let us evaluate the effective Josephson energy. Now gle current biased junction, Eq. (5.9) and Fig. 9, but the circuit has two dynamical variables, superconducting with flux-controlled critical current. This property of phases, φ1,2 across the two Josephson junctions. Defin- the SQUID is used in qubit applications for controlling ing phases as shown in the figure, and applying consider- the Josephson coupling, and also for measuring the qubit ations from Sections V A and V B we find for the static flux. Josephson currents, The kinetic energy of the SQUID can readily be writ- ten down noticing that it is associated with the charging Ic1 sin φ1 Ic2 sin φ2 = Ie, (5.18) energy of the two junction capacitances connected in par- − allel, where Ie is the biasing current. Let us further assume small inductance of the SQUID loop and neglect the mag- ¯h 2 φ˙2 K(φ˙)= (C + C ) . (5.25) netic energy of circulating currents. Then the total volt- 2e 1 2 2 age drop over the two junctions is zero, V1 + V2 = 0, and therefore, φ1 + φ2 = φe, where φe is the biasing phase Thus the Lagrangian for the SQUID has a form similar 2 related to biasing magnetic flux. Introducing new vari- to Eq. (5.11) where EC = (2e) /2(C1 + C2), ables, 2 2 ¯h φ˙ φe ¯h φ1 φ2 L(φ, φ˙)= 2EJ cos (1 cos φ)+ Ieφ. φ = ± , (5.19) 4EC − 2 − 2e ± 2 (5.26) and taking into account that 2φ+ = φe, we rewrite equa- tion (5.18) on the form, D. Single Cooper Pair Box (SCB) Ic(φe) sin(φ + α)= Ie, (5.20) − There is a particularly important Josephson junction where circuit consisting of a small superconducting island con- nected via a Josephson tunnel junction to a large super- 2 2 Ic(φe)= Ic1 + Ic2 + Ic1Ic2 cos φe, (5.21) conducting reservoir (see Fig. 11). q 15
The effect of the gate electrode is only essential for the kinetic term, and repeating previous analysis we arrive at the following equation for the kinetic energy (assuming I for simplicity identical junctions), 2 CΣ ¯h Cg K(φ˙ )= φ˙ Vg , − 2 2e − − C Σ CΣ =2C + Cg. (5.29) Combining this kinetic energy with the potential en- ergy derived in previous subsection, we arrive at the La- Vg grangian of the SCB (cf. Eq. (5.28)) where both the charging energy and the Josephson energy can be con- trolled, FIG. 12: Single Cooper pair transistor (SCT): SCB with loop- shape bulk electrodes; charge fluctuations on the island pro- 2 CΣ ¯h ˙ Cg duces current fluctuation in the loop. L = φ Vg 2 2e − − C Σ φe +2EJ cos cos φ . (5.30) 2 − The island is capacitively coupled to another massive electrode, which may act as an electrostatic gate. The voltage source Vg controls the gate potential. In the nor- mal state, such a circuit is named a Single Electron Box VI. QUANTUM SUPERCONDUCTING (SEB)114 for the following reason: if the junction resis- CIRCUITS tance exceeds the quantum resistance Rq 26 kΩ, and the temperature is small compared to the≈ charging en- One may look upon the Kirchhoff rules, as well as the ergy of the island, the system is in a Coulomb blockade circuit Lagrangians, as the equations describing the dy- regime1,115 where the electrons can only be transferred namics of electromagnetic field in the presence of the to the island one by one, the number of electrons on electric current. Generally, this electromagnetic field is a the island being controlled by the gate voltage. In the quantum object, and therefore there must be a quantum superconducting state, the same circuit is called a Sin- generalization of the equations in the previous section. gle Cooper pair Box (SCB)116,117; for a review see the At first glance, the idea of quantization of an equation de- book118. An experimental SCB device is shown in Fig. scribing a macroscopic circuit containing a huge amount 2. In this section we consider a classical Lagrangian for of electrons may seem absurd. To convince ourselves that the circuit. Since the structure now has two capacitances, the idea is reasonable, it is useful to recall an early ar- one from the tunnel junction, C, and another one from gument in favor of the quantization of electron dynamics the gate, Cg, the electrostatic term in the Hamiltonian in atoms, and to apply it to the simplest circuit, an rf- must be reconsidered. SQUID: When the current oscillations are excited in the Let us first evaluate the electrostatic energy of the SQUID, it works as an antenna radiating electromagnetic SCB. It has the form, waves. Since EM waves are quantized, the same should apply to the antenna dynamics. CV 2 C (V V )2 To quantize the circuit equation, we follow the conven- + g g − , (5.27) 2 2 tional way of canonical quantization: first we introduce the Hamiltonian and then change the classical momen- where V is the voltage over the tunnel junction. Then the tum to the momentum operator. The Hamiltonian is Lagrangian can be written (omitting the constant term), related to the Lagrangian as
2 ˙ C ¯h C H(p, φ)= pφ L, (6.1) L(φ, φ˙)= Σ φ˙ g V E (1 cos φ), (5.28) − 2 2e − C g − J − Σ where p is the canonical momentum conjugated to coor- dinate φ, where CΣ = C + Cg. The interferometer effect of two Josephson junctions ∂L p = . (6.2) connected in parallel (Fig. 12) can be used to control the ∂φ˙ Josephson energy of the single Cooper pair box. This setup can be viewed as a flux-biased dc SQUID where For the simplest case of a single junction, Eq. (5.6), the Josephson junctions have very small capacitances and are momentum reads, placed very close to each other so that the island confined 2 between them has large charging energy. The gate elec- ¯h p = Cφ.˙ (6.3) trode is connected to the island to control the charge. 2e 16
The so defined momentum has a simple interpretation: The commutation relation between the phase operator it is proportional to the charge q = CV on the junction and the pair number operator has a particularly simple capacitor, p = (¯h/2e)q, or the number n of electronic form, pairs on the junction capacitor, [φ, nˆ]= i. (6.13) p =¯hn. (6.4) The meaning of the quantization procedure is the fol- The Hamiltonian for the current-biased junction has lowing: the phase and charge dynamical variables can the form (omitting a constant), not be exactly determined by means of physical measure- ments; they are fundamentally random variables with the ¯h probability of realization of certain values given by the H = E n2 E cos φ I φ. (6.5) C − J − 2e e modulus square of the wave function of a particular state, Similarly, the Hamiltonian for the SQUID circuit has the φ = ψ∗(φ) φ ψ(φ) dφ, q = ψ∗(φ)ˆq ψ(φ) dφ. form h i h i Z Z (6.14) 2 (φ φe) The time evolution of the wave function is given by the H(n, φ)= E n2 E cos φ + E − . (6.6) C − J L 2 Schr¨odinger equation, The dc SQUID considered in the previous Section V C ∂ψ(φ, t) i¯h = Hψˆ (φ, t) (6.15) has two degrees of freedom. The Hamiltonian can be ∂t written by generalizing Eqs. (6.5), (6.6) for the phases φ . In the symmetric case we have, where H¯ is the circuit quantum Hamiltonian. The ex- ± plicit form of the quantum Hamiltonian for the circuits 2 2 H = EC n+ + EC n 2EJ cos φ+ cos φ considered above, is the following: − − − 2 (2φ+ φe) ¯h + EL − + Ieφ . (6.7) rf-SQUID: 2 2e − 2 2 (φ φe) For the SCB, Eq. (5.28), the conjugated momentum has Hˆ = EC nˆ EJ cos φ + EL − ; (6.16) the form, − 2 Current biased JJ: ¯hCΣ ¯h ˙ Cg p = φ Vg , (6.8) ¯h 2e 2e − CΣ Hˆ = E nˆ2 E cos φ + I φ; (6.17) C − J 2e e and the Hamiltonian reads, dc-SQUID: 2 H = EC (n ng) EJ cos φ, (6.9) ˆ 2 2 − − H = EC nˆ+ + EC nˆ 2EJ cos φ+ cos φ − − − 2 2 where now EC = (2e) /2CΣ, and n = p/¯h has the mean- (2φ+ φe) ¯h + EL − + Ieφ . (6.18) ing of the number of electron pairs (Cooper pairs) on the 2 2e − island electrode. ng = CgVg/2e is the charge on the gate capacitor (in units of− Cooper pairs (2e)), which can and finally the single Cooper pair box (SCB): be tuned by the gate potential, and which therefore plays Hˆ = E (ˆn n )2 E cos φ, (6.19) the role of external controlling parameter. C − g − J The quantum Hamiltonian results from Eq. (6.2) by For junctions connecting macroscopically large elec- substituting the classical momentum p for the differential trodes, the charge on the junction capacitor is a con- operator, tinuous variable. This implies that no specific boundary ∂ conditions on the wave function are imposed. The sit- pˆ = i¯h . (6.10) uation is different for the SCB: in this case one of the − ∂φ electrodes, the island, is supposed to be small enough Similarly, one can define the charge operator, to show pronounced charging effects. If tunneling is for- bidden, electrons are trapped on the island, and their ∂ number is always integer (the charge quantization con- qˆ = 2ei , (6.11) − ∂φ dition). However, there is a difference between the ener- gies of even and odd numbers of electrons on the island: and the operator of the pair number, while an electron pair belongs to the superconducting condensate and has the additional energy EC , a single ∂ electron forms an excitation and thus its energy consists nˆ = i . (6.12) − ∂φ of the charging energy, EC /2 plus the excitation energy, 17
∆ (parity effect)116. To prevent the appearance of indi- resonance width, γ = 1/RC compared to the resonance vidual electrons on the island and to provide the SCB frequency, γ ω . It is intuitively clear that the quan- ≪ LC regime, the condition ∆ EC /2 must be fulfilled. Thus tization effect will be destroyed when the level broaden- when the tunneling is switched≫ on, only Josephson tun- ing exceeds the level spacing. For quantum behavior of neling is allowed since it transfers Cooper pairs and the the circuit, narrow resonances, γ ωLC, are therefore number of electrons on the island must change pairwise, essential. However, even in this case≪ it is hard to ob- n = integer. In order to provide such a constraint, peri- serve the quantum dynamics in linear circuits such as odic− boundary conditions on the SCB wave function are LC-resonators123 because the expectation values of the imposed, linear oscillator follow the classical time evolution. Thus the presence of non-linear circuit elements is essential. ψ(φ)= ψ(φ +2π). (6.20) The linear oscillator provides the simplest example of This implies that arbitrary state of the SCB is a super- a quantum energy level spectrum: it only consists of dis- position of the charge states with integer amount of the crete levels with equal distance between the levels. The Cooper pairs, level spectrum of the Josephson junction associated with a pendulum potential is more complicated. Firstly, be- inφ ψ(φ)= ane . (6.21) cause of the non-linearity (non-parabolic potential wells), n the energy spectrum is non-equidistant (anharmonic), X The uncertainty of the dynamical variables are not im- the high-energy levels being closer to each other than the portant as long as the relative mean deviations of dy- low-energy levels. Moreover, for energies larger than the namical variables are small, i.e. the amplitudes of the amplitude of the potential (top of the barrier), EJ , the quantum fluctuations are small. In this case, the particle spectrum is continuous. Secondly, one has to take into behaves as a classical particle. It is known from quan- account the possibility of a particle tunneling between tum mechanics, that this corresponds to large mass of neighboring potential wells: this will produce broadening the particle, in our case, to large junction capacitance. of the energy levels into energy bands. The level broad- Thus we conclude that the quantum effects in the circuit ening is determined by the overlap of the wave function dynamics are essential when the junction capacitances tails under the potential barriers, and it must be small are sufficiently small. The qualitative criterion is that for levels lying very close to the bottom of the potential the charging energy must be larger than, or comparable wells. Such a situation may only exist if the level spac- ing, given by the plasma frequency of the tunnel junction to, the Josephson energy of the junction, EC EJ . Typical Josephson energies of the tunnel junctions∼ in is much smaller than the Josephson energy,hω ¯ J EJ , i.e. when E E . This almost classical regime≪ with qubit circuits are of the order of a few degrees Kelvin C ≪ J or less. Bearing in mind that the insulating layers of Josephson tunneling dominating over charging effects, is the junctions have the thickness of few atomic distances, called the phase regime, because phase fluctuations are and modeling the junction as a planar capacitor, the esti- small and the superconducting phase is well defined. In the opposite case, E E , the lowest energy level lies mated junction area should be smaller than a few square C ≫ J micrometers to observe the circuit quantum dynamics. well above the potential barrier, and this situation corre- One of most important consequences of the quantum sponds to wide energy bands separated by small energy dynamics is quantization of the energy of the circuit. Let gaps. In this case, the wave function far from the gap us consider, for example, the LC circuit. In the classical edges can be well approximated with a plane wave, case, the amplitude of the plasma oscillations has con- iqφ tinuous values. In the quantum case the amplitude of ψ (φ) = exp . (6.23) q 2e oscillation can only have certain discrete values defined through the energy spectrum of the oscillator. The linear This wave function corresponds to an eigenstate of the oscillator is well studied in the quantum mechanics, and charge operator with well defined value of the charge, q. its energy spectrum is very well known, Such a regime with small charge fluctuations is called the charge regime. En =¯hωLC(n +1/2), n =0, 1, 2, ... (6.22) One may ask, why is quantum dynamics never ob- served in ordinary electrical circuits? After all, in high- VII. BASIC QUBITS frequency applications, frequencies up to THz are avail- able, which corresponds to a distance between the quan- The quantum superconducting circuits considered tized oscillator levels of order 10K, which can be observed above contain a large number of energy levels, while for at sufficiently low temperature. For an illuminative dis- qubit operation only two levels are required. Moreover, cussion on this issue see the paper by Martinis, Devoret these two qubit levels must be well decoupled from the and Clarke123. According to Ref. 2 it is the dissipation other levels in the sense that transitions between qubit that kills quantum fluctuations: as known from classi- levels and the environment must be much less probable cal mechanics, the dissipation (normal resistance) broad- than the transitions between the qubit levels itself. Typi- ens the resonance, and good resonators must have small cally that means that the qubit should involve a low-lying 18 pair of levels, well separated from the spectrum of higher levels, and not being close to resonance with any other transitions.
A. Single Josephson Junction (JJ) qubit
The simplest qubit realization is a current biased JJ with large Josephson energy compared to the charging energy. In the classical regime, the particle representing the phase either rests at the bottom of one of the wells of the ”washboard” potential (Fig. 7), or oscillates within the well. Due to the periodic motion, the average voltage FIG. 13: Quantized energy levels in the potential of a current biased Josephson Junction across the junction is zero, φ˙ = 0. Strongly excited states, where the particle may escape from the well, correspond 3/2 to the dissipative regime with non-zero average voltage where Umax =2√2(Φ0/2π)(1 Ie/Ic) is the height of across the junction, φ˙ = 0. the potential barrier at given− bias current. In the quantum regime6 described by the Hamiltonian (6.5), B. Charge qubits 2 ¯h Hˆ = EC nˆ EJ cos φ Ieφ, (7.1) − − 2e 1. Single Cooper pair Box - SCB particle confinement, rigorously speaking, is impossi- ble because of macroscopic quantum tunneling (MQT) An elementary charge qubit can be made with the SCB operating in the charge regime, E E . Neglecting through the potential barrier, see Fig. 13. However, C ≫ J the probability of MQT is small and the tunneling may the Josephson coupling implies the complete isolation of be neglected if the particle energy is close to the bot- the island of the SCB, with a specific number of Cooper pairs trapped on the island. Correspondingly, the eigen- tom of the local potential well, i.e. when E EJ . To find the conditions for such a regime, it is≪ convenient functions, to approximate the potential with a parabolic function, 2 2 EC (ˆn ng) n = En n , (7.4) U(φ) (1/2)EJ cos φ0 (φ φ0) , where φ0 corresponds to − | i | i ≈ − the potential minimum, EJ sin φ0 = (¯h/2e)Ie. Then the correspond to the charge states n =0, 1, 2..., with the en- lowest energy levels, Ek =hω ¯ p(k +1/2) are determined 2 ergy spectrum En = EC (n ng) , as shown in Fig. 14. by the plasma frequency, ω = 21/4ω (1 I /I )1/4. It − p J − e c The ground state energy oscillates with the gate voltage, then follows that the levels are close to the bottom of the and the number of Cooper pairs in the ground state in- potential if E E , i.e. when the Josephson junction C ≪ J creases. There are, however, specific values of the gate is in the phase regime, and moreover, if the bias current voltage, e.g. ng = 1/2 where the charge states 0 and is not too close to the critical value, Ie < Ic. 1 become degenerate. Switching on a small Josephson| i It is essential for qubit operation that the spectrum in |couplingi will then lift the degeneracy, forming a tight the well is not equidistant. Then the two lowest energy two-level system. levels, k =0, 1 can be employed for the qubit operation. The qubit Hamiltonian is derived by projecting the Truncating the full Hilbert space of the junction to the full Hamiltonian (6.19) on the two charge states, 0 , 1 , subspace spanned by these two states, 0 and 1 , we leading to | i | i may write the qubit Hamiltonian on the| form,i | i ˆ 1 1 HSCB = (ǫ σz + ∆ σx), (7.5) H = ǫσ , (7.2) −2 q −2 z where ǫ = E (1 2n ), and ∆ = E . The qubit level C − g J where ǫ = E1 E0. energies are then given by the equation The interlevel− distance is controlled by the bias cur- rent. When bias current approaches the critical cur- 1 2 2 2 E1,2 = E (1 2ng) + E , (7.6) rent, level broadening due to MQT starts to play a role, ∓2 C − J E E + iΓ /2. The MQT rate for the lowest level is q k k k the interlevel distance being controlled by the gate volt- given→ by82 age. At the degeneracy point, ng = 1/2, the diag- onal part of the qubit Hamiltonian vanishes, the lev- 52ω U 7.2U Γ = p max exp max , (7.3) els being separated by the Josephson energy, EJ , and MQT 2π ¯hω − ¯hω s p p the qubit eigenstates corresponding to the cat states, 19
current is distributed between the two junctions. The EC answer to this question is apparently equivalent to eval- uating the persistent current circulating in the SQUID loop. This current was neglected so far because the as- sociated induced flux φ˜ = 2φ φ was assumed to be + − e frozen, φ˜ = 0, in the limit of infinitely small SQUID in- ductance, L = 0. Let us now lift this assumption and allow fluctuation of the induced flux; the Hamiltonian 0 n will then take the form (6.7), in which a gate potential is g included in the charging term of the SCB, and the term n = −10 1 containing external current is dropped, ˆ 2 2 HSCT = EC (ˆn ng) + EC nˆ+ − − 2 (2φ+ φe) 2EJ cos φ+ cos φ + EL − . (7.7) − − 2 FIG. 14: Single Cooper pair box (SCB): charging energy (up- (ˆn = i∂/∂φ ). For small but non-zero inductance, per panel), and charge on the island (lower panel) vs gate + − + potential. Washed-out onsets of the charge steps indicate the amplitude of the induced phase is small, φ˜ =2φ+ − quantum fluctuations of the charge on the island. φe 1, and the cosine term containing φ+ can be ex- panded,≪ yielding the equation
HˆSCT = HˆSCB(φ )+ Hˆosc(φ˜)+ Hˆint . (7.8) E −
HˆSCB(φ ) is the SCB Hamiltonian (6.19) with the flux − ∆/2 dependent Josephson energy, EJ (φe)=2EJ cos(φe/2). Hˆosc(φ˜) describes the linear oscillator associated with the
variable φ˜, ∋ φ˜2 0 Hˆ (φ˜)=4E n˜ˆ2 + E , (7.9) osc C L 2 and the interaction term reads,
φe Hˆint = EJ sin cos(φ ) φ˜ . (7.10) 2 − FIG. 15: Energy spectrum of the SCB (solid lines): it results Thus, the circuit consists of the non-linear oscillator of from hybridization of the charge states (dashed lines). the SCB linearly coupled to the linear oscillator of the SQUID loop. This coupling gives the possibility to mea- sure the charge state of the SCB by measuring the per- E , E = 0 1 . For theses states, the average 1 2 sistent currents and the induced flux. charge| i | oni the| islandi ∓ | isi zero, while it changes to 2e far Truncating Eq. (7.8) we finally arrive at the Hamil- from the degeneracy point, where the qubit eigenstates∓ tonian which is formally equivalent to the spin-oscillator approach pure charge states. Hamiltonian, The SCB was first experimentally realized by Lafarge 117 et al. , observing the Coulomb staircase with steps of 1 13 Hˆ = (ǫσ + ∆(φ )σ )+ λφσ˜ + H . (7.11) 2e and the superposition of the charge states, see also . SCT −2 z e x x osc Realization of the first charge qubit by manipulation of the SCB and observation of Rabi oscillations was done In this equation, ∆(φe) = 2EJ cos(φe/2), and λ = 1,119,120 by Nakamura et al. , and further investigated the- EJ sin(φe/2). oretically by Choi et al.121.
C. Flux qubit 2. Single Cooper pair Transistor - SCT 1. Quantum rf-SQUID In the SCB, charge fluctuations on the island gener- ate fluctuating current between the island and large elec- An elementary flux qubit can be constructed from an trode. In the two-junction setup discused at the end of rf-SQUID operating in the phase regime, E E . Let J ≫ C Section V D, an interesting question concerns how the us consider the Hamiltonian (6.6) at φe = π, i.e. at half 20
E subspace spanned by these two levels. The starting point of the truncation procedure is to approximate the double well potential with Ul and Ur, as shown in Fig. 17, to confine the particle to the left or to the right well, respec- tively. The corresponding ground state wave functions l and r satisfy the stationary Schr¨odinger equation, | i
∆ | i 0 φ Hˆ l = E l , Hˆ r = E r . (7.14) l| i l| i r| i r| i
FIG. 16: Double-well potential of the rf-SQUID with de- The averaged induced flux for these states, φl and φr generate quantum levels in the wells (black). Macroscopic have opposite signs, manifesting opposite directions of quantum tunneling (MQT) through the potential barrier in- the circulating persistent currents. Let us allow the bias troduces a level splitting ∆, and the lowest level pair forms a flux to deviate slightly from the half integer value, φe = qubit. π + f, so that the ground state energies are not equal but still close to each other, E E . The tunneling will l ≈ r (φ) hybridize the levels, and we can approximate the true U eigenfunction, E , | i Hˆ E = E E , (7.15) Ur Ul | i | i with a superposition,
E = a l + b r . (7.16) | i | i | i 0 φ The qubit Hamiltonian is given by the matrix elements of the full Hamiltonian, Eq. (7.15), with respect to the FIG. 17: Flux qubit: truncation of the junction Hamiltonian; states l and r , dashed lines indicate potentials of the left and right wells with | i | i ground energy levels. H = E + l U U l , ll l h | − l| i H = E + r U U r , rr r h | − r| i integer bias magnetic flux. The potential, U(φ), shown Hrl = El r l + r U Ul l . (7.17) h | i h | − | i in Fig. 16 has two identical wells with equal energy levels when MQT between the wells is neglected (phase regime, In the diagonal matrix elements, the second terms are ωJ EJ ). These levels are connected with current fluc- small because the wave functions are exponentially small tuations≪ within each well around averaged values cor- in the region where the deviation of the approximated responding to clockwise and counterclockwise persistent potential from the true one is appreciable. The off di- currents circulating in the loop (the flux states). Let agonal matrix element is exponentially small because of us consider the lowest, doubly degenerate, energy level. small overlap of the ground state wave functions in the When the tunneling is switched on, the levels split, and left and right wells, and also here the main contribution a tight two-level system is formed with the level spacing comes from the first term. Since the wave functions can determined by the MQT rate, which is much smaller than be chosen real, the truncated Hamiltonian is symmet- ric, H = H . Then introducing ǫ = E E , and the level spacing in the well. In the case that the tunnel- lr rl r − l ing barrier is much smaller than the Josephson energy, ∆/2 = Hrl, we arrive at the Hamiltonian of the flux the potential in Eq. (5.9) can be approximated, qubit,
2 1 (φ φe) ˆ U(φ) = E (1 cos φ)+ E − H = (ǫσz + ∆σx). (7.18) J − L 2 −2 ˜2 φ 1+ ε 4 The Hamiltonian of the flux qubit is formally equiva- EL ε fφ˜ + φ˜ , (7.12) ≈ − 2 − 24 ! lent to that of the charge qubit, Eq. (7.5), but the phys- ical meaning of the terms is rather different. The flux where φ˜ = φ π, f = φe π, and where qubit Hamiltonian is written in the flux basis, i.e. the − − basis of the states with certain averaged induced flux, φl E ε = J 1 1. (7.13) and φr (rather than the charge basis of the charge qubit). EL − ≪ The energy spectrum of the flux qubit is obviously the same as that of the charge qubit, determines the height of the tunnel barrier. The qubit Hamiltonian is derived by projecting the 1 E = ǫ2 + ∆2, (7.19) whole Hilbert space of the full Hamiltonian (6.6) on the 1,2 ∓2 p 21
Φ
FIG. 18: Persistent current flux qubit (PCQ) with 3 junctions. The side junctions are identical, while the central junction has smaller area. as shown in Fig. 15. However, for the flux qubit the dashed lines indicate persistent current states in the ab- sence of macroscopic tunneling, and the current degener- acy point (ǫ = 0) corresponds to a half-integer bias flux. The energy levels are controlled by the bias magnetic flux (instead of the gate voltage for the charge qubit). At the flux degeneracy point, f = 0 (φe = π), the level spacing is determined by the small amplitude of FIG. 19: Potential energy landscape U(φ1,φ2) = U(φ+,φ−) tunneling through macroscopic potential barrier, and the of the Delft qubit as a function of the two independent phase wave functions correspond to the cat states, which are variables φ1 and φ2, or equivalently, φ+ (horizontal axis) and equally weighted superpositions of the flux states. Far φ− (vertical axis). Black represents the bottom of the poten- tial wells, and white the top of the potential barriers. The from the degeneracy point, the qubit states are almost qubit double-well potential is determined by the potential pure flux states. landscape centered around the origin in the horizontal direc- The possibility to achieve quantum coherence of tion, and typically has the shape shown in Fig. 16. Courtesy macroscopic current states in an rf-SQUID with a small of C.H. van der Wal. capacitance Josephson junction was first pointed out in 1984 by Leggett122. However, successful experimental observation of the effect was achieved only in 2000 by Friedman et al.8. To explain the idea, let us consider the potential energy. The three phases are not independent and satisfy the relation φ1 +φ2 +φ3 = φe. Let us suppose that the qubit 2. 3-junction SQUID - persistent current qubit (PCQ) is biased at half integer flux quantum, φe = π. Then introducing new variables, φ = (φ1 φ2)/2, we have ± ± The main drawback of the flux qubit with a single Josephson junction (rf-SQUID) described above concerns U(φ+, φ )= EJ [2 cos φ cos φ+ (1/2+ ε)cos2φ+]. the large inductance of the qubit loop, the energy of − − − − (7.21) which must be comparable to the Josephson energy to The potential landscape is shown in Fig. 19, and the form the required double-well potential profile. This im- qubit potential consists of the double well structure near plies large size of the qubit loop, which makes the qubit the points (φ+, φ ) = (0, 0). An approximate form of vulnerable to dephasing by magnetic fluctuations of the the potential energy− is given by environment. One way to overcome this difficulty was pointed out by Mooij et al.6, replacing the large loop in- φ4 ductance by the Josephson inductance of an additional U(φ , 0) E 2εφ2 + + . (7.22) + ≈ J − + 4 tunnel junction, as shown in Fig. 18, The design employs three tunnel junctions connected in series in a supercon- ducting loop. The inductive energy of the loop is chosen to be much smaller than the Josephson energy of the Each well in this structure corresponds to clock- and junctions. The two junctions are supposed to be identi- counterclockwise currents circulating in the loop. The cal while the third junction is supposed to have smaller amplitude of the structure is given by the parameter area, and therefore smaller Josephson and larger charging ǫEJ , and for ǫ 1 the tunneling between these wells energy. The Hamiltonian has the form, dominates. Thus≪ this qubit is qualitatively similar to the single-junction qubit described above, but the quan- Hˆ = E [ˆn2 +ˆn2 +ˆn2/(1/2+ ε)] C 1 2 3 − titative parameters are different and can be significantly EJ [cos φ1 + cos φ2 + (1/2+ ε)cos φ3]. (7.20) optimized. 22
D. Potential qubits ∆ 1
The superconducting qubits that have been discussed in previous sections exploit the fundamental quantum un- certainty between electric charge and magnetic flux. This E 0 R ∆ uncertainty appears already in the dynamics of a single 2 Josephson junction (JJ), which is the basis for elementary JJ qubits. There are however other possibilities. One of them is to delocalize quantum information in a JJ net- work by choosing global quantum states of the network ∆ −1 as a computational basis. Recently, some rather com- 00 1φ 2π2 plicated JJ networks have been discussed, which have the unusual property of degenerate ground state, which might be employed for efficient qubit protection against FIG. 20: Energy spectrum of microscopic bound Andreev lev- decoherence124,125. els; the level splitting is determined by the conact reflectivity. An alternative possibility of further miniaturization of superconducting qubits could be to replace the stan- net Josephson current is negligibly small. dard Josephson junction by a quantum point contact In QPCs, the Josephson effect is associated with mi- (QPC), using the microscopic conducting modes in the croscopic Andreev levels, localized in the junction area, JJ QPC, the bound Andreev states, as a computational which transport Cooper pairs from one junction electrode basis, allowing control of intrinsic decoherence inside the 127,128 105,126 to the other . As shown in Fig. 20, the Andreev junction . levels, two levels per conducting mode, lie within the su- To explain the physics of this type of qubit, let us perconducting gap and have the phase-dependent energy consider an rf SQUID (see Fig. 8) with a junction that spectrum, has such a small cross section that the quantization of electronic modes in the (transverse) direction perpendic- E = ∆ cos2(φ/2)+ R sin2(φ/2), (7.23) ular to the current flow becomes pronounced. In such a a ± junction, quantum point contact (QPC), the Josephson q current is carried by a number of independent conduct- (here ∆ is the superconducting order parameter in the junction electrodes). For very small reflectivity, R 1, ing electronic modes, each of which can be considered an ≪ elementary microscopic Josephson junction characterized and phase close to π (half integer flux bias) the An- by its own transparency. The number of modes is propor- dreev two-level system is well isolated from the contin- tional to the ratio of the junction cross section and the uum states. The expectation value for the Josephson area of the atomic cell (determined by the Fermi wave- current carried by the level is determined by the Andreev length) of the junction material. In atomic-sized QPCs level spectrum, with only a few conducting modes, the Josephson current 2e dEa can be appreciable if the conducting modes are transpar- Ia = , (7.24) ¯h dφ ent (open modes). If the junction is fully transparent e (reflectivity R = 0) then current is a well defined quan- and it has different sign for the upper and lower level. tity. This will correspond to a persistent current with Since the state of the Andreev two-level system is deter- certain direction circulating in the qubit loop. On the mined by the phase difference and related to the Joseph- other hand, for a finite reflectivity (R = 0), the electronic son current, the state can be manipulated by driving 6 back scattering will induce hybridization of the persistent magnetic flux through the SQUID loop, and read out current states giving rise to strong quantum fluctuation by measuring circulating persistent current129,130. of the current. This microscopic physics underlines recent proposal for Such a quantum regime is distinctly different from the Andreev level qubit105,126. The qubit is similar to the macroscopic quantum coherent regime of the flux qubit macroscopic flux qubits with respect to how it is manip- described in Section VII, where the quantum hybridiza- ulated and measured, but the great difference is that the tion of the persistent current states is provided by charge quantum information is stored in the microscopic quan- fluctuations on the junction capacitor. Clearly, charging tum states. This difference is reflected in the more com- effects will not play any essential role in quantum point plex form of the qubit Hamiltonian, which consists of contacts, and the leading role belongs to the microscopic the two-level Hamiltonian of the Andreev levels strongly mechanism of electron back scattering. This mechanism coupled to the quantum oscillator describing phase fluc- is only pronounced in quantum point contacts: in classi- tuations, cal (large area) contacts, such as junctions of macroscopic qubits with areas of several square micrometers, the cur- iσx√R φ/2 φ φ Hˆ = ∆ e− cos σ + √R sin σ +Hˆ [φ], rent is carried by a large number (> 104) of statistically 2 z 2 y osc independent conducting modes, and fluctuation of the (7.25) 23
2 Hˆosc[φ]= EC nˆ +(EL/2)(φ φe) . Comparing this equa- tion with e.g. the SCT Hamiltonian− (7.8), we find that the truncated Hamiltonian of the SCB is replaced here by the Andreev level Hamiltonian.
VIII. QUBIT READOUT AND MEASUREMENT OF QUANTUM INFORMATION
In this section we present a number of proposed, and realized, schemes for measuring quantum states of vari- ous superconducting qubits.
A. Readout: why, when and how? FIG. 21: Measurement of the phase qubit. Long-living levels form the qubit, while the dashed line indicates a leaky level As already mentioned in Section III D, the ultimate with large energy. objective of a qubit readout device is to distinguish the eigenstates of a qubit in a single measurement ”without destroying the qubit”, a so called ”single-shot” quantum eigenstate. Then the value is predetermined and the non-demolition (QND) projective measurement. This qubit left in the eigenstate (Stern-Gerlach-style). objective is essential for several reasons: state prepara- On the other hand, to extract the desired final result it tion for computation, readout for error correction during may be necessary to create an ensemble of calculations to the calculation, and readout of results at the end of the be able to perform a complete measurement to determine the expectation values of variables of interest, performing calculation. Strictly speaking, the QND property is only 131 needed if the qubit must be left in an eigenstate after quantum state tomography . the readout. In a broader sense, readout of a specific qubit must of course not destroy any other qubits in the system. B. Direct qubit measurement It must be carefully noted that one cannot ”read out the state of a qubit” in a single measurement - this is pro- Direct destructive measurement of the qubit can be il- hibited by quantum mechanics. It takes repeated mea- lustrated with the example of a single JJ (phase) qubit, surements on a large number of replicas of the quantum Section VII A. After the manipulation has been per- state to characterize the state of the qubit (Eq. (3.1)) - formed (e.g. Rabi oscillation), the qubit is left in a su- ”quantum tomography”131. perposition of the upper and lower energy states. To The measurement connects the qubit with the open determine the probability of the upper state, one slowly system of the detector, which collapses the combined sys- increases the bias current until it reaches such a value tem of qubit and measurement device to one of its com- that the upper energy level equals Or gets close to) the mon eigenstates. If the coupling between the qubit and top of the potential barrier, see Fig. 21. Then the junc- the detector is weak, the eigenstates are approximately tion, being at the upper energy level, will switch from the those of the qubit. In general however, one must con- Josephson branch to the dissipative branch, and this can sider the eigenstates of the total qubit-detector system be detected by measuring the finite average voltage ap- and manipulate gate voltages and fluxes such that the pearing across the junction (voltage state). If the qubit readout measurement is performed in a convenient en- is in the lower energy state the qubit will remain on the ergy eigenbasis (see e.g.42,79). Josephson branch and a finite voltage will not be de- Even under ideal conditions, a single-shot measure- tected (zero-voltage state). An alternative method to ment can only determine the population of an eigenstate activate switching30 is to apply an rf signal with reso- if the system is prepared in an eigenstate: then the an- nant frequency (instead of tilting the junction potential) swer will always be either ”0” or ”1”. If an ideal single- in order to excite the upper energy level and to induce the shot measurement is used to read out a qubit superpo- switching event, see Fig. 21 (also illustrating a standard sition state, e.g. during Rabi oscillation, then again the readout method in atomic physics). answer can only be ”0” or ”1”. To determine the qubit It is obvious that, in this example, the qubit upper 2 2 population (i.e. the a1 and a2 probabilities) requires energy state is always destroyed by the measurement. repetition of the measurement| | | to| obtain the expectation Single-shot measurement is possible provided the MQT value. During the intermediate stages of quantum com- rate for the lower energy level is sufficiently small to putation one must therefore not perform a measurement prevent the junction switching during the measurement on a qubit unless one knows, because of the design and time. It is also essential to keep a sufficiently small rate of timing of the algorithm, that this qubit is in an energy interlevel transitions induced by fluctuations of the bias 24
P
t << t ms t < t ms t = t ms SCB
0 t I(n 0 ) t I(n 1 ) m SET x x FIG. 23: Probability distributions P of counted electrons as functions of time after the turning on the measurement beam of electrons. Courtesy of G. Johansson, Chalmers. Vg capacitively coupled to a charge qubit has also been ex- tensively discussed in literature132,139,140,141,142,143,144. FIG. 22: Single Electron Transistor (SET) capacitively cou- The induced charge on the SET gate depends on the pled to an SCB. state of the qubit, affecting the SET working point and determining the conductivity and the average cur- rent. The development of the probability distributions current and by the current ramping. of counted electrons with time is shown in Fig. 23. A similar kind of direct destructive measurement was As the number of counted electrons grows, the distribu- performed by Nakamura et al.1 to detect the state of the tions separate and become distinguishable, the distance charge qubit. The qubit operation was performed at the between the peaks developing as N and the width √N. Detailed investigations144∼show that the two charge degeneracy point, ug = 1, where the level splitting ∼ is minimal. An applied gate voltage then shifted the SCB electron-number probability distributions correlate with working point (Fig. 15), inducing a large level splitting the probability of finding the qubit in either of two energy of the pure charge states 0 and 1 (the measurement levels. The long-time development depends on the inten- preparation stage). In this| processi | thei upper 1 charge sity and frequency distribution of the back-action noise state went above the threshold for Cooper pair| i decay, from the electron current. With very weak detector back action, the qubit can relax to 0 during the natural re- creating two quasi-particles which immediately tunnel | i out via the probe junction into the leads. These quasi- laxation time T1. With very strong back-action noise at particles were measured as a contribution to the classical the qubit frequency, the qubit may become saturated in charge current by repeating the experiment many times. a 50/50 mixed state. Obviously, this type of measurement is also destructive. D. Measurement via coupled oscillator C. Measurement of charge qubit with SET Another method of qubit read out that has attracted much attention concerns the measurement of the proper- Non-destructive measurement of the charge qubit has ties of a linear or non-linear oscillator coupled to a qubit. been implemented by connecting the qubit capacitively 69 This method is employed for the measurement of induced to a SET electrometer . The idea of this method is to magnetic flux and persistent current in the loop of flux use a qubit island as an additional SET gate (Fig. 22), qubits and charge-phase qubits, as well as for charge mea- controlling the dc current through the SET depending surement on charge qubits. With this method, the qubit on the state of the qubit. When the measurement is to affects the characteristics of the coupled oscillator, e.g. be performed, a driving voltage is applied to the SET, changes the shape of the oscillator potential, after which and the dc current is measured. Another version of the the oscillator can be probed to detect the changes. There measurement procedure is to apply rf bias to the SET 69,133,134,135 are two versions of the method: resonant spectroscopy of (rf-SET ) in Fig. 22, and to measure the dis- a linear tank circuit/cavity, and threshold detection us- sipative or inductive response. In both cases the trans- ing biased JJ or SQUID magnetometer. missivity will show two distinct values correlated with The first method uses the fact that the resonance fre- the two states of the qubit. Yet another version has re- 136 quency of a linear oscillator weakly coupled to the qubit cently been developed by the NEC group to perform undergoes a shift depending on the qubit state. The single-shot readout: the Cooper pair on the SCB island effect is most easily explained by considering the SCT then tunnels out onto a trap island (instead of the leads) Hamiltonian, Eq. (11.1), used as a gate to control the current through the SET. The physics of the SET-based readout has been exten- 1 42,137,138 Hˆ = (ǫσ + ∆(φ )σ )+ λ(φ )φσ˜ sively studied theoretically (see and references SCT −2 z e x e x therein). A similar idea of controlling the transmission 1 +4E n˜ˆ2 + E φ˜2. (8.1) of a quantum point contact (QPC) (instead of an SET) C 2 L 25
Let us proceed to the qubit energy basis, in which casee Vout (t) u(t) the qubit Hamiltonian takes the form (E/2)σ , E = − z √ǫ2 + ∆2. The interaction term in the qubit eigenbasis will consist of two parts, the longitudinal part, λzφσ˜ z, λz = (∆/E)λ, and the transverse part, λxφσ˜ x, λx = (ǫ/E)λ. In the limit of weak coupling the transverse part SCB of interaction is the most essential. In the absence of interaction (φe = 0), the energy spectrum of the qubit + oscillator system is E 1 En = +¯hω(n + ), (8.2) ∓ ∓ 2 2 where ¯hω = √8E E is the plasma frequency of the C L g oscillator. The effect of weak coupling is enhanced in V the vicinity of the resonance, when the oscillator plasma frequency is close to the qubit level spacing,hω ¯ E. Let ≈ us assume, however, that the coupling energy is smaller FIG. 24: SCT qubit coupled to a readout oscillator. The than the deviation from the resonance, λ ¯hω E . x ≪ | − | qubit is operated by input pulses u(t). The readout oscillator Then the spectrum of the interacting system in the lowest is controlled and driven by ac microwave pulses Vg(t). The perturbative order will acquire a shift, output signal will be ac voltage pulses Vout(t), the amplitude 2 or phase of which may discriminate between the qubit ”0” λx¯hω δEn = (n + 1) . (8.3) and ”1” states. ± ± E (¯hω E) L − This shift is proportional to the first power of the oscilla- tor quantum number n, which implies that the oscillator frequency acquires a shift (the frequency of the qubit is also shifted145,146,147,148,149). Since the sign of the oscil- lator frequency shift is different for the different qubit states, it is possible to distinguish the state of the qubit SCT Ib by probing this frequency shift. In the case of the SCT, the LC oscillator is a generic part of the circuit. It is equally possible to use an addi- Vg tional LC oscillator inductively coupled to a qubit. This type of device has been described by Zorin71 for SCT readout, and recently implemented for flux qubits by Il’ichev et al.27,39. FIG. 25: SCT qubit coupled to a JJ readout quantum oscil- Figure 24 illustrates another case, namely a charge lator. The JJ oscillator is controlled by dc/ac current pulses qubit capacitively coupled to an oscillator, again pro- Ib(t) adding to the circulating currents in the loop due to the viding energy resolution for discriminating the two qubit SCT qubit. The output will be dc/ac voltage pulses Vout(t) levels150. Analysis of this circuit is similar to the one dis- discriminating between the qubit ”0” and ”1” states. cussed below in the context of qubit coupling via oscilla- tors, Section IX. The resulting Hamiltonian is similar to Eq. (9.20), namely, inserted in the qubit loop, as shown in Fig. 25. When the measurement of the qubit state is to be per- ˆ ˆ ˆ H = HSCB + λσyφ + Hosc. (8.4) formed, a bias current is sent through the additional junc- In comparison with the case of the SCT, Eq. (8.4) has tion. This current is then added to the qubit-state depen- a different form of the coupling term, which does not dent persistent current circulating in the qubit loop. If change during rotation to the qubit eigenbasis. There- the qubit and readout currents flow in the same direction, fore the coupling constant λ directly enters Eq. (8.3). the critical current of the readout JJ is exceeded, which Recently, this type of read out has been implemented for induces the junction switching to the resistive branch, a charge qubit by capacitively coupling the SCB of the sending out a voltage pulse. This effect is used to distin- qubit to a superconducting strip resonator151,152,153. guish the qubit states. The method has been extensively used experimentally by Vion et al.21,22,23. To describe the circuit, we add the Lagrangian of a E. Threshold detection biased JJ, Eq. (5.11),
2 ˙2 To illustrate the threshold-detection method, let us ¯h φ m ¯h L = m + EJ cos φ + Ieφ, (8.5) consider an SCT qubit with a third Josephson junction 4EC 2e 26 to the SCT Lagrangian (generalized Eq. (5.30), cf. Eqs. U (6.7), (7.7)),
2 2 2 ¯h ˙ 2eCg ¯h ˙2 LSCT = φ Vg + φ+ 4E − − ¯hC 4E C Σ C 1 2 2EJ cos φ+ cos φ ELφ˜ (8.6) − − − 2 (here we have neglected a small contribution of the gate capacitance to the qubit charging energy). The phase quantization condition will now read: 2φ+ + φ = φe + φ˜. The measurement junction will be assumed in the phase m m θ regime, EJ EC , and moreover, the inductive energy will be the largest≫ energy in the circuit, E Em. The L ≫ J latter implies that the induced phase is negligibly small FIG. 26: Josephson potential energy of the measurement and can be dropped from the phase quantization condi- junction during the measurement: for the ”0” qubit eigen- tion. We also assume that φe = 0, thus 2φ+ + φ = 0. state there is a well (full line) confining a level, while for the Then, after having omitted the variable φ+, the kinetic ”1” qubit state there is no well (dashed line). energy term of the qubit can be combined with the much larger kinetic energy of the measurement junction lead- ing to insignificant renormalization of the measurement while for the excited state it does not. This implies that junction capacitance, Cm + C/4 Cm. As a result, the when the junction is in the ground state, no voltage will total Hamiltonian of the circuit will→ take the form, be generated. However, if the junction is in the excited state, it will switch to the resistive branch, generating a
2 φ voltage pulse that can be detected. Hˆ = EC (ˆn ng) 2EJ cos cos φ − − − 2 − With the discussed setup the direction of the persis- tent current is measured. It is also possible to arrange m 2 m ¯h +EC nˆ EJ cos φ Ieφ. (8.7) the measurement of the flux by using a dc SQUID as a − − 2e threshold detector. Such a setup is suitable for the mea- Since the measurement junction is supposed to be almost surement of flux qubits. Let us consider, for example, the classical, its phase is fairly close to the minimum of the three junction flux qubit from Section VII C inductively junction potential. During qubit operation, the bias cur- coupled to a dc SQUID. Then, under certain assump- rent is zero; hence the phase of the measurement junction tions, the Hamiltonian of the system can be reduced to is zero. When the measurement is made, the current is the following form: ramped to a large value close to the critical current of 1 m ˆ the measurement junction, I = (2e/¯h)E δI, tilting H = (ǫσz + ∆σx) e J −2 the junction potential and shifting the minimum− towards s s ¯h π/2. Introducing a new variable φ = π + θ, we expand +EC nˆ2 (EJ + λσz)cos φ δI φ, (8.10) the potential with respect to small θ 1 and, truncating − − 2e ≪ s the qubit part, we obtain where EJ is an effective (bias flux dependent) Josephson energy of the SQUID, Eq. (5.24), and λ is an effective ǫ ∆ θ coupling constant proportional to the mutual inductance Hˆ = σz 1 σx −2 − 2 − 2 of the qubit and the SQUID loops. θ3 ¯h +Em nˆ2 Em + δI θ, (8.8) C J 6 2e − IX. PHYSICAL COUPLING SCHEMES FOR TWO QUBITS where ∆ = 2√EJ . The ramping is supposed to be adi- abatic so that the phase remains at the minimum point. Let us analyze the behavior of the potential minimum A. General principles by omitting a small kinetic term and diagonalizing the Hamiltonian (8.8). The corresponding eigenenergies de- A generic scheme for coupling qubits is based on pend on θ, the physical interaction of linear and non-linear os- cillators constituting a superconducting circuit. The 3 2 E m θ ¯h ∆ Hamiltonians for the SCB, rf-SQUID, and plain JJ con- E (θ)= EJ + δI θ, (8.9) ± ∓ 2 − 6 2e ± 4E tain quadratic terms representing the kinetic and poten- tial energies plus the non-linear Josephson energy term, as shown in Fig. 26. Then within the interval of the bias which is quadratic when expanded to lowest order, currents, δI (2e/¯h)(∆2/4E), the potential energy corresponding| |≤− to the ground state has a local minimum, Hˆ = E (ˆn n )2 + E (1 cos φ) (9.1) C − g J − 27
SCB SCB
C3 Vg Vg FIG. 27: Fixed inductive (flux) coupling of elementary flux qubit. The loops can be separate, or have a common leg like in the figure. FIG. 28: Fixed capacitive coupling of charge qubits 2 2 (φ φe) Hˆ = EC nˆ + EJ (1 cos φ)+ EL − (9.2) − 2 for each qubit. The last matrix element is exponentially small, while the first two ones are approximately equal to 2 ¯h the minimum points of the potential energy, φl and φr, Hˆ = EC nˆ + EJ (1 cos φ)+ Ieφ. (9.3) − 2e respectively. This implies that the truncated interaction In a multi-qubit system the induced gate charge in the basically has the zz-form, SCB, or the flux through the SQUID loop, or the phase Hˆint = λσz1σz2, in the Josephson energy, will be a sum of contributions 2 from several (in principle, all) qubits. The energy of the 1 ¯h 1 λ = (L− )12 (φl φr)1(φl φr)2. (9.7) system can therefore not be described as the sum of two 8 2e − − independent qubits because of the quadratic dependence, and the cross terms represent interaction energies of dif- ferent kinds: capacitive, inductive and phase/current. C. Capacitive coupling of charge qubits Moreover, using JJ circuits as non-linear coupling ele- ments we have the advantage that the direct physical One of the simplest coupling schemes is the capacitive coupling strength may be controlled, e.g tuning the in- coupling of charge qubits. Such a coupling is realized by ductance via current biased JJs, or tuning the capaci- connecting the islands of two SCBs via a small capacitor, tance by a voltage biased SCB. as illustrated in Fig. 28. This will introduce an addi- tional term in the Lagrangian of the two non-interacting SCBs, Eq. (5.27), namely the charging energy of the B. Inductive coupling of flux qubits capacitor C3, C V 2 A common way of coupling flux qubits is the inductive δL = 3 3 . (9.8) coupling: magnetic flux induced by one qubit threads 2 the loop of another qubit, changing the effective external The voltage drop V3 over the capacitor is expressed via flux. This effect is taken into account by introducing the the phase differences across the qubit junctions, inductance matrix L , which connects flux in the i-th ik ¯h loop with the current circulating in the k-th loop, V = (φ˙ φ˙ ), (9.9) 3 2e 1 − 2 Φi = LikIk. (9.4) and thus the kinetic part of the Lagrangian (5.28) will Xk take the form The off-diagonal element of this matrix, L12, is the mu- 1 ¯h 2 tual inductance which is responsible for the interaction. K(φ˙1, φ˙2)= Cikφ˙iφ˙k By using the inductance matrix, the magnetic part of the 2 2e Xi,k potential energy in Eq. (6.16) can be generalized to the 2 ¯h case of two coupled qubits, C V φ˙ , (9.10) −2e gi gi i 2 i 1 ¯h 1 X (L− )ik(φi φei)(φk φek). (9.5) 2 2e − − where the capacitance matrix elements are Cii = CΣi + ik X C3, and C12 = C3. Then proceeding to the circuit quan- Then following the truncation procedure explained in tum Hamiltonian as described in Section VI, we find the Section VII C, we calculate the matrix elements, interaction term, 2 1 l φ˜ f l , r φ˜ f r , l φ˜ f r , (9.6) Hˆ =2e (C− ) nˆ nˆ . (9.11) h | − | i h | − | i h | − | i int 12 1 2 28
SCB SCB C3 JJ1 JJ2
JJ Vg Vg
FIG. 30: Capacitive coupling of single JJ qubits
FIG. 29: Fixed phase coupling of charge qubits
This interaction term is diagonal in the charge basis, and SCB SCB therefore leads to the zz-interaction after truncation,
2 e 1 Hˆ = λσ σ , λ = (C− ) . (9.12) int z1 z2 2 12 L C The qubit Hamiltonians are given by Eq. (7.5) with charging energies renormalized by the coupling capaci- Vg Vg tor.
FIG. 31: Two charge qubits coupled to a common LC- D. JJ phase coupling of charge qubits oscillator.
Instead of the capacitor, the charge qubits can be con- nected via a Josephson junction154, as illustrated in Fig. turns to σy, and the qubit-qubit interaction takes the 29, yy-form, In this case, the Josephson energy of the coupling junc- tion EJ3 cos(φ1 φ2) must be added to the Lagrangian 2 − ˆ 2 ¯h ωp1ωp2 1 in addition to the charging energy. This interaction term Hint = λσy1σy2, λ =2e (C− )12. (9.15) s EC1EC2 is apparently off-diagonal in the charge basis and, after truncation, gives rise to xx- and yy-couplings, E Hˆ = λ(σ σ + σ σ ), λ = J3 , (9.13) F. Coupling via oscillators int x1 x2 y1 y2 4 or equivalently, Besides the direct coupling schemes described above, several schemes of coupling qubits via auxiliary oscilla- Hˆint =2λ(σ+,1σ ,2 + σ ,1σ+,2). (9.14) 42 − − tors have been considered . Such schemes provide more flexibility, e.g. to control qubit interaction, to couple two remote qubits, and to connect several qubits. Moreover, E. Capacitive coupling of single JJs in many advanced qubits, the qubit variables are generi- cally connected to the outside world via an oscillator (e.g. Capacitive coupling of JJ qubits, illustrated in Fig. 30 the Delft and Saclay qubits). To explain the principles is described in a way similar to the charge qubit, in terms of such a coupling, we consider the coupling scheme for of the Lagrangian Eqs. (9.8), (9.9), and the resulting charge qubits suggested by Shnirman et al.14. interaction Hamiltonian has the form given in Eq. (9.11).
Generally, in the qubit eigenbasis, 0 and 1 , all ma- trix elements of the interaction Hamiltonian| i are| i non-zero. 1. Coupling of charge (SCB, SCT) qubits However, if we adopt a parabolic approximation for the Josephson potential, then the diagonal matrix elements In this circuit the island of each SCB is connected to turn to zero, n00 = n11 = 0, while the off-diagonal matrix ground via a common LC-oscillator, as illustrated in Fig. 1/4 elements remain finite, n01 = n10 = i(EJ /EC ) . 31. The kinetic energy (5.27) of a single qubit should Then, after truncation, the charge− number− operatorn ˆ now be modified taking into account the additional phase 29 difference φ across the oscillator, JJ 2 ˙ ˙ 1 ¯h ˙2 ˙ ˙ 2 K(φ ,i; φ)= 2Cφ ,i + Cg(Vgi φ φ ,i) . SCT − 2 2e − − − − SCT h (9.16)i The cross term in this equation can be made to vanish by a change of qubit variable,
Cg φ ,i = φi aφ, a = . (9.17) − − CΣ Vg Vg The kinetic energy will then split into two independent parts, the kinetic energy of the qubit in Eq. (5.28), and an additional quadratic term, FIG. 32: Charge (charge-phase) qubits coupled via a com- mon Josephson junction providing phase coupling of the two 1 ¯h 2 CC circuits g φ˙2, (9.18) 2 2e C Σ which should be combined with the kinetic energy of the the displacement does not change the oscillator ground oscillator, leading to renormalization of oscillator capac- state energy, which then drops out after the averaging, we finally arrive at the Hamiltonian of the direct effective itance. Expanding the Josephson energy, after the change of qubit coupling, variable, gives λ1λ2 Hˆint = σy1σy2. (9.23) E cos(φ aφ) E cos φ E aφ sin φ . (9.19) − EL Ji i − ≈ Ji i − Ji i provided the amplitude of the oscillations of φ is small. for the oscillator-coupled charge qubits in Fig. 31. The last term in this equation describes the linear cou- pling of the qubit to the LC-oscillator. Collecting all the terms in the Lagrangian and perform- 2. Current coupling of SCT qubits ing quantization and truncation procedures, we arrive at the following Hamiltonian of the qubits coupled to the Charge qubits based on SCTs can be coupled by con- oscillator (this is similar to Eq. (8.4) for the SCT), necting loops of neighboring qubits by a large Josephson junction in the common link155,156,157,158,159,160,161, as il- Hˆ = (HˆSCB,i + λiσyiφ)+ Hˆosc, (9.20) lustrated in Fig. 32, i=1,2 The idea is similar to the previous one: to couple qubit X variables to a new variable, the phase of the coupling ˆ (i) where HSCB is given by Eq. (7.5), and Josephson junction, then to arrange the phase regime for the junction with large plasma frequency (E Ccoupl ≪ EJiCg EJcoupl), and then to average out the additional phase. λi = , (9.21) CΣ Technically, the circuit is described using the SCT Hamil- tonian, Eqs. (7.7), (7.8), for each qubit, is the coupling strength, and ˆ 2 2 ˆ 2 2 HSCT = EC (ˆn ng) + EC nˆ+ Hosc = ECoscnˆ + ELφ /2, (9.22) − − 2 (2φ+ φe) 2EJ cos φ+ cos φ + EL − , (9.24) is the oscillator Hamiltonian where the term in Eq. (9.18) − − 2 has been included. The physics of the qubit coupling in this scheme is and adding the Hamiltonian of the coupling junction, the following: quantum fluctuation of the charge of one 2 qubit produces a displacement of the oscillator, which Hˆc = EC,cnˆ EJ,c cos φc. (9.25) c − perturbs the other qubit. If the plasma frequency of the LC oscillator is much larger than the frequencies of all The phase φc across the coupling junction must be added qubits, then virtual excitation of the oscillator will pro- to the flux quantization condition in each qubit loop; ˜ duce a direct effective qubit-qubit coupling, the oscillator e.g., for the first qubit 2φ+,1 + φc = φe,1 + φ1 (for the staying in the ground state during all qubit operations. second qubit the sign of φc will be minus). Assuming To provide a small amplitude of the zero-point fluctu- small inductive energy, EL EJ,c, we may neglect φ˜; ations, the oscillator plasma frequency should be small then assuming the flux regime≪ for the coupling Joseph- compared to the inductive energy, or ECosc EL. Then son junction we adopt a parabolic approximation for the ≪ 2 the fast fluctuations can be averaged out. Noticing that junction potential, EJ,cφc /2. 30
With these approximations, the Hamiltonian of the first qubit plus coupling junction will a take form sim- ilar to Eq. (9.24) where EJ,c will substitute for EL, and φ will substitute for2φ φ . Finally assuming the am- c + − e plitude of the φc-oscillations to be small, we proceed as in the previous subsection, i.e. expand the cosine term obtaining linear coupling between the SCB and the oscil- lator, truncate the full Hamiltonian, and average out the oscillator. This will yield the following interaction term,
λ1λ2 φi Hˆint = σx1σx2, λi = EJ sin (9.26) EJ,c 2
This coupling scheme also applies to flux qubits: in this FIG. 33: Flux transformer with variable coupling controlled case, the coupling will have the same form as in Eq. (9.7), by a SQUID. but the strength will be determined by the Josephson energy of the coupling junction, cf. Eq. (9.26), rather than by the mutual inductance. on the controlling flux is given by the equation167,
1 Φ E πΦ − G. Variable coupling schemes 2 = 1+ J cos cx , (9.27) Φ E Φ 1 L 0 Computing with quantum gate networks basically as- sumes that one-and two-qubit gates can be turned on where EJ is the Josephson energy of the SQUID junction, and off at will. This can be achieved by tuning qubits and EL is the inductive energy of the transformer. with fixed, finite coupling in and out of resonance, in NMR-style computing162. Here we shall discuss an alternative way, namely to vary the strength of the physical coupling between 2. Variable Josephson coupling nearest-neighbour qubits, as discussed in a number of re- cent papers155,156,158,159,160,163,164,165,166. A variable Josephson coupling is obtained when a sin- gle Josephson junction is substituted by a symmetric dc SQUID whose effective Josephson energy 2EJ cos(φe/2) 1. Variable inductive coupling depends on the magnetic flux threading the SQUID loop (see the discussion in Section V C). This property is To achieve variable inductive coupling of flux qubits commonly used to control level spacing in both flux and one has to be able to the control the mutual inductance charge qubits introduced in Section V D, and it can also of the qubit loops. This can be done by different kinds be used to switch on and off qubit-qubit couplings. For of controllable switches (SQUIDS, transistors)163 in the example, the coupling of the charge-phase qubits via circuit. In a recent experiment, a variable flux trans- Josephson junction in Fig. 32 can be made variable former was implemented as a coupling element (see Fig. by substituting the single coupling junction with a dc 33) by controlling the transforming ratio167. The flux SQUID155,156. transformer is a superconducting loop strongly induc- The coupling scheme discussed in Section IX F 1 is tively coupled to the qubit loops, which are distant from made controllable by using a dc SQUID design for the each other so that the direct mutual qubit inductance is SCB as explained in Section V D). Indeed, since the cou- negligibly small. Because of the effect of quantization of pling strength depends on the Josephson energy of the magnetic flux in the transformer loop112, the local vari- qubit junction, Eq. (9.21), this solution provides vari- ation of magnetic flux Φ1 induced by one qubit will af- able coupling of the qubits. Similarly, the coupling of fect a local magnetic flux Φ2 in the vicinity of the other the SCTs considered in Section IXF2 can be made con- qubit creating effective qubit-qubit coupling. When a dc trollable by employing a dc SQUID as a coupling ele- SQUID is inserted in the transformer loop, as shown in ment. A disadvantage of this solution is that the qubit Fig. 33, it will shortcircuit the transformer loop, and the parameters will vary simultaneously with varying of the transformer ratio Φ2/Φ1 will change. The effect depends coupling strength. A more general drawback of the dc on the current flowing through the SQUID, and is propor- SQUID-based controllable coupling is the necessity to tional to the critical current of the SQUID. The latter is apply magnetic field locally, which might be difficult to controlled by applying a magnetic flux Φcx to the SQUID achieve without disturbing other elements of the circuit. loop, as explained in Section V C and shown in Fig. 33. This is however an experimental question, and what are Quantitatively, the dependence of the transformer ratio practical solutions in the long run remains to be seen. 31
Ib
SCT SCT
Vg Vg FIG. 35: Variable capacitance tuned by a voltage-controlled SCB.
FIG. 34: Coupled charge qubits with current-controlled phase coupling: the arrow indicates the direction of the controlling nˆ2, and it also depends on the gate voltages of the qubits bias current. and the coupling junction. In contrast to the previous scheme, here the coupling junction is not supposed to be in the phase regime; however, it is still supposed to be 3. Variable phase coupling fast, EJ EJi. Then the energy gap in the spectrum of the coupling≫ junction is much bigger than the qubit An alternative solution for varying the coupling energy, and the junction will stay in the ground state is based on the idea of controlling the properties during qubit operations. Then after truncation, and av- of the Josephson junction by applying external dc eraging out the coupling junction, the Hamiltonian of the 158,159,160 current , as illustrated in Fig. 34. circuit will take the form, Let us consider the coupling scheme of Section IX F 2: the coupling strength here depends on the plasma fre- Hˆ = HˆSCB,i + ǫ0 (σz1 + σz2) , (9.30) quency of the coupling Josephson junction, which in turn i depends on the form of the local minimum of the junction X potential energy. This form can be changed by tilting the where the qubit Hamiltonian is given by Eq. (7.5), and junction potential by applying external bias current (Fig. the function ǫ0 is the ground state energy of the coupling 34), as discussed in Section V A. The role of the exter- junction. The latter can be generally presented as a linear nal phase bias, φe, will now be played by the minimum combination of terms proportional to σz1σz2 and σz1 + point φ0 of the tilted potential determined by the applied σz2, bias current, EJ,c sin φ0 = (¯h/2e)Ie. Then the interaction term will read, ǫ0 (σz1 + σz2)= α + νσz1σz2 + β (σz1 + σz2) . (9.31)
2 2 with coefficients depending on the gate potentials. The EJ sin (φ0/2) Hˆint = λσx1σx2, λ = , (9.28) second term in this expression gives the zz-coupling (in E cos φ J,c 0 the charge basis), and the coupling constant ν may, ac- 165 and local magnetic field biasing is not required. cording to the analysis of Ref. , take on both positive and negative values depending on the coupling junction gate voltage. In particular it may turn to zero, implying 4. Variable capacitive coupling qubit decoupling.
Variable capacitive coupling of charge qubits based on H. Two qubits coupled via a resonator a quite different physical mechanism of interacting SCB charges has been proposed in Ref.165. The SCBs are then connected via the circuit presented in Fig. 35. In the previous discussion, the coupling oscillator plays The Hamiltonian of this circuit, including the charge a passive role, being enslaved by the qubit dynamics. qubits, has the form However, if the oscillator is tuned into resonance with a qubit, then the oscillator dynamics will become essential, Hˆ = Hˆ +E (ˆn q(ˆn +ˆn ))2 E cos φ, (9.29) leading to qubit-oscillator entanglement. In this case, SCB,i C − 1 2 − J i the approximation of direct qubit-qubit coupling is not X appropriate; instead, manipulations explicitly involving where EC and EJ Ec are the charging and Josephson the oscillator must be considered. energies of the coupling∼ junction, andn ˆ and φ are the Let us consider, as an example, operations with two charge and the phase of the coupling junction. The func- charge qubits capacitively coupled to the oscillator. As- tion q is a linear function of the qubit charges,n ˆ1, and suming the qubits to be biased at the degeneracy point 32 and proceeding to the qubit eigenbasis (phase basis in The manipulation should not necessarily be step-like, it this case), we write the Hamiltonian on the form (cf. Eq. is sufficient to pass the resonance rapidly enough to pro- (7.10), vide the Landau-Zener transition, i.e. the speed of the frequency ramping should be comparable to the qubit ∆i level splittings. Hˆ = σzi λiσx φi + Hˆosc[φ]. (9.32) − 2 − References to recent work on the entanglement of i=1,2 X qubits and oscillators will be given in Section IV C. Let us consider the following manipulation involving the variation of the oscillator frequency (cf. Ref164): at time t = 0, the oscillator frequency is off-resonance with both X. DYNAMICS OF MULTI-QUBIT SYSTEMS qubits, A. General N-qubit formulation ¯hω(0) < ∆1 < ∆2. (9.33) Then the frequency is rapidly ramped so that the oscil- A general N-qubit Hamiltonian with general qubit- lator becomes resonant with the first qubit, qubit coupling can be written on the form
¯hω(t ) = ∆ , (9.34) ǫi ∆i 1 1 1 Hˆ = ( σ + σ ) + λ σ σ − 2 zi 2 xi 2 ν,ij νi νj i i,j;ν the frequency remaining constant for a while. Then the X X frequency is ramped again and brought into resonance (10.1) with the second qubit, To solve the Schr¨odinger equation ¯hω(t2) = ∆2. (9.35) Hˆ ψ = E ψ (10.2) Finally, after a certain time it is ramped further so that | i | i the oscillator gets out of resonance with both qubits at we expand the N-qubit state in a complete basis, e.g. the N N the end, 2 basis states of the (σz) operator,
¯hω(t>t3) > ∆2. (9.36) ψ = a q | i q| i When passing through the resonance, the oscillator is = a0 0..00 + a1 0..01 + a2 0..10 + ...a(2N X1) 1..11 | i | i | i − | i hybridized with the corresponding qubit, and after pass- (10.3) ing the resonance, the oscillator and qubit have become entangled. For example, let us prepare our system at and project onto the basis states t = 0 in the excited state ψ(0) = 100 = 1 0 0 , | i | i| i| i where the first number denotes the state of the oscilla- Hˆ p p ψ = E ψ (10.4) | ih | i | i tor (first excited level), and the last numbers denote the p (ground) states of the first and second qubits, respec- X tively. After the first operation, the oscillator will be obtaining the ususal matrix equation entangled with the first qubit, q Hˆ p ap = Eaq (10.5) iα h | | i ψ(t