Slides for Capri School: Surface Codes

Andrew N Cleland Capri April 24-28 2017

Lecture 1: Introduction
 Qubits “all” of
 Operators quantum
 Measurement mechanics
 Quantum circuits
 Superconducting qubits
 Error models
 References on general QM & QC
E D Commins, Quantum Mechanics: An Experimentalist's Approach (Cambridge, 2014)
ISBN 978‐1107063990

J J Sakurai and J Napolitano, Modern Quantum Mechanics (2nd ed) (Addison‐Wesley,

2011) ISBN 978‐0‐8053‐8291‐4

R Shankar, Principles of Quantum Mechanics (2nd ed) (Springer, 1994) ISBN 978‐1‐4757‐
0576‐8

M Nielsen and I Chuang, Quantum Computation and Quantum Information (Cambridge,

2011) ISBN‐13: 978‐1107002173

J Preskill, [Link] Lecture notes for

Quantum Information, Caltech Physics 229

A.G. Fowler et al, Phys Rev A 86, 032324 (2012)

 Qubits
| 〉 Bloch | 〉 | 〉
Single qubits: | 〉 sphere:
| 〉
“Z” basis
| 〉

“X” basis: | 〉
 Qubits
| 〉 | 〉
Two qubits: | 〉 | 〉
Product state:
No entanglement

qubit 1 “Z” basis

qubit 2

In general cannot be written in

product form (entangled states)
 Bell Basis

| 〉 | 〉
Two qubits: | 〉 | 〉

Bell states provide entangled state basis:

Indices relate to standard quantum circuit used to

generate Bell states: will discus more later
 Qubits

Four qubits: 2 16 independent coefficients

qubits:

100 qubits:
2 1.3 10 coefficients
More than world’s current digital storage capacity...
 Operators
Operator : Dual space “bras”
Adjoint operator :
Unitary operator:
Hermitian operator:
Given a set of basis vectors for the Hilbert space :

Representation of
⋆
Representation of
Representation of = c.c. transpose of ’s representation
 Hermitian Operators
equals : ⋆

Eigenvector-eigenvalue equation:

Eigenvector Eigenvalues are all real

Eigenvectors provide orthonormal basis;

representation in eigenvector basis is diagonal
 Operator products & commutation
Operator commutator:
If operators “commute”
Anti-commutator:
If operators “anti-commute”
Note means operate first with then with
Representation of is product of matrix representations
of and
 Single qubit operators
(all in the , | 〉 basis)

Any single qubit operator:

rotation about axis

Hadamard halfway between and ̂

⁄
“ ⁄8 gate”
 Single qubit operators

Bit-flip
operator:

Phase-flip
operator:

Combined
bit & phase
flip:

Basis
change
 Two qubit operators
CNOT gate: if control qubit in | 〉, flip (NOT) target qubit
| 〉 | 〉 | 〉
| 〉 | 〉 | 〉
Truth | 〉 | 〉 | 〉
table: | 〉 | 〉 | 〉
| 〉 | 〉 | 〉

CZ gate: if control qubit in | 〉, apply to target qubit

| 〉 | 〉 | 〉
| 〉 | 〉 | 〉
Truth | 〉 | 〉 | 〉
table: | 〉 | 〉 | 〉
| 〉 | 〉 | 〉
 Measurement
Any physical measurement is represented by a Hermitian operator
1. The measurement outcome is an eigenvalue of the operator
2. The state resulting from measurement is the corresponding
eigenvector
3. The probability of an outcome is the squared amplitude of
that eigenvector in the original state decomposition
.
Z measurement: +1 eigenvalue: Probability
Outcome | 〉 1
-1 eigenvalue: Probability
1
Outcome | 〉
X measurement: +1 eigenvalue: Probability ⁄2
1
Outcome | 〉
-1 eigenvalue: Probability ⁄2
1
Outcome | 〉
 Quantum circuits

time

Two qubit circuit

 Quantum circuits

time

Two qubit circuit

 Quantum circuits
CNOT: Can result in
entangled states

Bell state generation circuit:

1
| 〉 | 〉
2
Inputs determine output:
1
| 〉 | 〉 ⊗| 〉
2 1
| 〉 | 〉
2
Physical implementation: Transmon qubits
∝Φ

energy
C 1
2

1⁄ magnetic flux Φ
• Microwave frequency circuits: ⁄2 ~ few GHz range
• Ground state operation: ≪ ⇒ 100 mK
• Straightforward packaging, cabling, control & measurement
electronics
• Need energy level anharmonicity to enable quantum control
 Transmon qubits
Al

energy
E Lucero

Al
| 〉
AlOx
Josephson junction: Inductance LJ
A very nonlinear
phase
changes with each
inductor photon in circuit
 Josephson equations:
sin
2 2 cos
 Circuit Hamiltonian: Inductance
cos ,
2
Conjugate variables
 Transmon qubits
Al

energy
Al
| 〉
AlOx

phase
• Qubit levels | 〉 and
• Qubit frequency ~ 4 7 GHz
• Anharmonic at single photon level: 0.95 0.97
 Transmon qubits
Al

energy
Al
Φ | 〉
AlOx
Capacitor is critical
to qubit performance phase
• Qubit levels | 〉 and , gates: Microwaves at
• Qubit frequency ~ 4 7 GHz gate: Tuning
• Anharmonic at single photon level: 0.95 0.97
• Can tune & by changing magnetic flux through qubit
• Qubit lifetime probably limited by loss in capacitor
• Qubit probably limited by flux noise
 CZ adiabatic gate 1 0 0 0
⇒ 0 1 0 0
Direct 0 0 1 0
Other states left unchanged 0 0 0 1
capacitive
coupling Avoided crossing | 〉:

A B
Frequency
| 〉
Adiabatic trajectory
40 ns total duration!
| 〉 ⇒
XY Z XY Z
Qubit A tuning
 Pritzker Nanofabrication Facility
 Recharge‐based nanofabrication facility
 In Eckhardt Research Center on UChicago
campus (southside Chicago)
 Full professional staff
 Open to all users (academic, industrial...)
 Sub-10 nm wafer-scale ebeam
lithography and 150 mm I-line UV
stepper
 Chlorine & fluorine based etching
 Ebeam and sputter deposition of
metals & insulators; ALD metal &
insulators; wet processing
 SEM, AFM, profilometry, ellipsometry
 Dicing up to 150 mm wafers
 Automatic wire bonder
 UCSB transmon: “xmon” qubit
Josephson
Measurement: junctions
• Dispersive in flux loop
resonator readout
• Measure change in Coupling to
phase of few- other qubits
photon excitation in
readout resonator
• Projective & Z control
accurate
Z rotations:
• Flux tuning varies Capacitance
C
• Changes qubit
frequency Josephson
X and Y rotations: junctions
• Microwaves at (tunable
qubit frequency inductance L)

Barends et al. (PRL, 2013)

Qubit Infrastructure

Dilution refrigerator
~10 mK
 Five qubit linear array

readout resonators
direct (one per qubit)
capacitive
coupling
between two control lines
qubits per qubit

XY XY

Z XY Z XY Z XY Z Z

Barends et al. (Nature, 2014)

 Experimental status (UCSB)
Gate Fidelity ( 0.03)
 Complete set of single qubit gates
X 99.92
needed to execute an arbitrary
Y 99.92
quantum algorithm
X/2 99.93
 All single qubit Clifford gates Y/2 99.95
operate with fidelity > 99.9%
-X 99.92
(randomized benchmarking)
-Y 99.91
 Controlled Z gate (equivalent to -X/2 99.93
CNOT): -Y/2 99.95
40 ns execution time
H 99.91
99.5% fidelity
I 99.95
 State preparation and
S (Z/2 99.92
measurement fidelity ~ 90%
CZ 99.5
Barends et al. (Nature, 2014)
 Programming a 5 qubit GHZ state
| 〉
CZ
Y/2

| 〉
CZ
-Y/2 Y/2

| 〉
CZ |Ψ〉
-Y/2 Y/2

| 〉 Y/2
CZ
-Y/2

| 〉 -Y/2 Y/2

2 Bell 3 GHZ 4 GHZ 5 GHZ

0.995 0.960 0.863 0.817

Barends et al. (Nature, 2014)
Nine qubit linear array

Scaling up to larger systems

 Google
 IBM
 Lincoln Labs/MIT

Kelly et al. (Nature, 2015)

 Error Models
One accepted and relatively straightforward way to introduce
error‐producing events in quantum computation is to use
Kraus operators. We will not cover that formalism here.
Consider the following less formal approach:

environment | | slightly
less than 1 | | small,
1

| | slightly less than 1

| | small,
1
 Error Models
Projectors onto | 〉 and | 〉:
1 0 1 0 0 1
0 0 2 0 1 2

We can write: ⇒
| 〉
⇒
| 〉
Then with | 〉 we can write
⇒
| 〉
 Error Models
Copying from previous slide, we can then write
⇒ | 〉

| 〉
2 2

2 2

| 〉
2 2
1 | 〉
We can treat the environment interaction as generating discrete
X, Y or Z errors!
 Summary of Monday’s lecture
• Superconducting qubits have single qubit fidelities ~99.95% (for
gate times ~50 ns)
• Qubit idle ~ same fidelity – due to interaction with environment
• To do useful computing need fidelities of order

• How to get there?

• Build perfection from imperfection – stabilizers & surface
codes to create logical qubits
• Models predict physical qubits with 99.99% fidelity can form
logical qubits with very high levels of fidelity, assuming:
 No strong long-range correlated errors
 Large circuits can be built maintaining fidelities
 Large entangled states don’t need any new physics
 ...
Slides for Capri School: Surface Codes

Andrew N Cleland Capri April 24-28 2017

Lecture 2: The Surface Code

 Codewords & error syndromes
 Surface code architecture
 Stabilizer operations
 Quiescent state
 Single qubit errors
 Logical qubits, logical operators
 Summary of Monday’s lecture
• Superconducting qubits have single qubit fidelities ~99.95% (for
gate times ~50 ns)
• Qubit idle ~ same fidelity – due to interaction with environment
• To do useful computing need fidelities of order

• How to get there?

• Build perfection from imperfection – stabilizers & surface
codes to create logical qubits
• Models predict physical qubits with 99.99% fidelity can form
logical qubits with very high levels of fidelity, assuming:
 No strong long-range correlated errors
 Large circuits can be built maintaining fidelities
 Large entangled states don’t need any new physics
 ...
 Literature
 S. B. Bravyi and A. Y. Kitaev, arXiv:quant‐ph/9811052
 E. Dennis, A. Y. Kitaev, A. Landahl, and J. Preskill, J. Math. Phys. 43, 4452 (2002)
 A. Y. Kitaev, in Proc. 3rd Intl. Conf. Quant. Comm. Comp. Meas., ed. O. Hirota, A. S.
Holevo, C. M. Caves (Plenum Press, New York, 1997)
 A. Y. Kitaev, Russian Math Surveys 52, 1191 (1997)
 A. Y. Kitaev, Ann. Phys. 303, 2 (2003)
 M. H. Freedman and D. A. Meyer, Found. Comp. Math. 1, 325 (2001)
 C. Wang, J. Harrington, and J. Preskill, Ann. Phys. 303, 31 (2003)
 R. Raussendorf, J. Harrington, and K. Goyal, Ann. Phys. 321, 2242 (2006)
 R. Raussendorf and J. Harrington, Phys. Rev. Lett. 98, 190504 (2007)
 R. Raussendorf, J. Harrington, and K. Goyal, New J. Phys. 9, 199 (2007)
 A.G. Fowler, A.M. Stephens, P. Groszkowski, Phys. Rev. A 80, 052312 (2009)
 A.G. Fowler, D.S. Wang, L.C.L. Hollenberg, Quant. Inf. Comp. 11, 8 (2011)
 D.S. Wang, A.G. Fowler, L.C.L. Hollenberg, Phys. Rev. A 83, 020302 (2011)
 A.G. Fowler, A.C. Whiteside, L.C.L. Hollenberg, Phys. Rev. Lett. 108, 180501 (2012)
 A.G. Fowler, M. Mariantoni, J.M. Martinis, A.N. Cleland, Phys Rev A 86, 032324
(2012)
 Error Syndromes & Codewords: Bit-flips
Simple example: Bit‐flip detection codewords

Simple quantum circuit enables encoding

Qubits 1 & 3 can out‐

vote qubit 2
Error example: X bit‐flip error on qubit 2 ( error):

How do we detect & fix this error (or a single or error)?

We build an error syndrome circuit
 Error Syndromes & Error Codes

Ancilla qubit

Double controlled Z:
If in | 〉 then apply to
qubits 1 and 2

1
| 〉 ⊗| 〉 1
2 | 〉
2
1
| 〉
2
 Error Syndromes & Error Codes

Measure the
ancilla along :
If 1:
Result is | 〉
1 and 1
| 〉
2 If 1:
Error Result is | 〉
| 〉
syndrome and 1
2 2
Error Syndromes & Error Codes

No error: No syndrome, 1
error: signals error with 1
error: and signal errors with 1
error: signals errors with 1
 Error Syndromes & Error Codes

 This 3 qubit codeword+2 ancilla error syndrome circuit can

only detect & identify single bit-flips
 There are (famous) 5 qubit codeword+4 ancilla and 7 qubit
codeword+6 ancilla syndromes that can detect and identify
single , and errors on any codeword qubit
 Surface code cycle & eigenstates
Measure Z

Measure X

Data qubit
Measure qubit

 Data qubits store computational state

 Measure qubit stabilize data qubits
 Measure Z stabilizes product of data qubits
 Measure X stabilizes product of data qubits
Surface code cycle & eigenstates
Measure Z

Measure X

 Data qubit states stabilized by measure

qubits
 Single bit-flip (phase-flip) detected by
change in Z (X) measurement outcome
 Stabilizer errors don’t repeat in time
Surface code logical operators

Stabilizer with value 1

 Surface code logical operators
Logical X:

Logical Z:
Same properties as

 and anti-commute,
as desired

Array of data & measure qubits forms

a logical qubit (81 physical qubits)  All required properties
 Errors & misidentification

 Measure Z qubits 1 and 3

change measurement sign
 Error!

 Could be due to errors on

data qubits 2 and 3

 Could be due to errors on

data qubits 1 and 4 and 5
• Probability of any single qubit error is (assumed small!)
• Probability of two qubit error is This is more likely!
• Probability of three qubit error is ≪
Natural conclusion is two-qubit error
If actually a 3-qubit error, this assumption generates a logical error!
 Errors & misidentification

 Chain with distance 5

 Most common mis-
identifications occur for
1 ⁄2 same-cycle qubit
errors being identified as
1 ⁄2 same-cycle errors
 Here a 3-qubit error erroneously
identified as a 2-qubit error (note
these are complementary)
 This causes a logical error

⁄ Bigger ⇒ smaller
 This occurs at a rate ∝
logical error rate
 Errors & misidentification
Simple statistical mode
10
Logical X error rate

10
∝

∝ ∝ ∝
10

10 10 10
Per step error rate
 Errors & misidentification
10
10
10
Logical X error rate

10

10

10

0.57%
Per step error rate
 Summary of today’s lecture
A array of data & measure qubits forms a logical qubit
 Sequence of and stabilizer measurements detects ,
or errors (maximum one per qubit per cycle)
 Spatial pattern of changed stabilizer outcomes uniquely
identifies location & type of error (if not too many!)
 Stabilizer measurement errors identified by non‐repeatability
in time
⁄
 Physical qubit error rate gives logical error rate ∝ (if
below threshold ~0.57%)
A 5 array takes 0.05% to ~99.998% fidelity
A 9 array takes 0.05% to ~99.999995% fidelity
Slides for Capri School: Surface Codes

Andrew N Cleland Capri April 24-28 2017

Lecture 3: Logical Qubits

 Forming logical qubits
 Initialization, measurement, errors
 Logical qubit operations
 Moving & braiding qubits
 Hadamard, S and T gates
 Estimates for sizes
 Surface code logical operators
Logical X:

Logical Z:
Same properties as

 and anti-commute,
as desired

Array of data & measure qubits forms

a logical qubit (81 physical qubits)  All required properties
 Summary of Tuesday’s lecture
A array of data & measure qubits forms a logical qubit
 Sequence of and stabilizer measurements detects ,
or errors (maximum one per qubit per cycle)
 Spatial pattern of changed stabilizer outcomes uniquely
identifies location & type of error (if not too many!)
 Stabilizer measurement errors identified by non‐repeatability
in time
⁄
 Physical qubit error rate gives logical error rate ∝ (if
0.57%)
A 5 array takes 0.05% to ~99.998% fidelity
A 9 array takes 0.05% to ~99.999995% fidelity
 This is a memory qubit only – no logic capability!
 Logical qubits: Z-cut qubit

Edge logical Z‐cut qubit: Double Z‐cut qubit:

fig 9 Turn off one Z stabilizer Turn off two Z stabilizers fig 10
 Logical qubits: Double Z,X-cut qubits

fig 11
 Initializing a Logical Qubit
Initializing an X-cut qubit in an X eigenstate

Logical X value
is that of X
stabilizer prior
to forming cut

To measure:
Simply turn
measure X
qubit back on
and report
value
fig 13
 Logical qubits: Initialize & Measure
Initializing X-cut along Z Measuring X-cut along Z

fig 14
 Logical qubits: Moving a qubit

Z Z Z Z Z Z

fig 17
 Logical qubits: Moving a qubit

Extend to by
multiplying by X values
of intermediate data
qubits

Extend to by
multiplying by
intermediate Z
stabilizer values

Convert to by
multiplying by final Z
stabilizer values
fig 18
 Logical qubits: Empty braid

Final
Sign determined by
enclosed stabilizer
values …

fig 19
 Logical qubits: Empty braid

Final
Sign determined by
intermediate Z
stabilizers in move

fig 20
 CNOT: Operator picture
CNOT gate: if control qubit in | 〉, flip (NOT) target qubit
| 〉 | ′〉 1 0 0 0
 Want to validate
0 1 0 0
an operation as 1 0 0 0 1
a CNOT 0 0 0 1 0
… 4 basis states
42=16 parameters
 Verifying the operation performs a CNOT can be done by finding how
pre‐operation basis states are mapped to post‐operation basis states
 Alternative is to check that all operators are transformed correctly
(Heisenberg picture)
 All one‐qubit operators are linear combinations of , , ,
 All two‐qubit operators are linear combinations of ⊗ , ⊗ , ⊗
, ⊗ … 4 16 combinations)
 Check ⊗ ⊗ for all combinations ,
 CNOT: Operator picture
However only four operator combinations are independent:
⊗ ⇒ ⊗ ⊗ ⇒ ⊗
⊗ ⇒ ⊗ ⊗ ⇒ ⊗

For example ⊗ ⊗ ⊗
⊗ ⊗
⊗ ⊗
⊗ ⊗

Hence only need to check four operator transformations to test

a proposed CNOT operation...
 Braiding

Final step in braid:

⊗ ⊗
(first CNOT identity)

fig 22
 Braiding

Final step in braid:

⊗ ⊗
(second CNOT identity)

fig 21
 Braiding and

⊗ :
Final step:
Turn on all Z stabilizers
Restores but generates
⊗ ⇒ ⊗

⊗ :
Easy to show that
transformation does
not change operators
 Braid operation is equivalent to CNOT
 Surface code protection maintained
fig 23
 Extending to all qubit cuts
 All operations involve Z‐cut control qubit and X‐cut target qubit
 Quantum circuits extend to Z‐Z and X‐X with 2 ancillas

Z control, Z target X control, X target

Z-cut X-cut X-cut

target in target in target out

fig 24
 Logical Hadamard

fig 26
 Logical Hadamard

on all data Shift by ½

qubits unit cell

fig 27&28
 Logical S and T gates

State is
recycled!

Result is
probabilistic

State is not
recycled 
⁄
fig 29
 “Short” qubit

 Can inject arbitrary state

| 〉 | 〉 ⁄ 2
 Amplitudes & phase
limited by control
electronics
 Cannot expect better
than ~ precision
 How to use in logical
qubit circuit?
| 〉 | 〉 ⁄ 2
fig 31
 Magic “Y” state distillation

 Seven
approximate
| 〉 states
used in seven
gates
 Output| 〉 is
purified | 〉
state
 Used as input
for next round
of purification

fig 32
 Magic “A” state distillation
 Fifteen approximate
| 〉 states used in 15
gates
 Output | 〉 is purified
| 〉 state
 Used as input for next
round of purification
 Four rounds requires
154 =50,625 initial | 〉
states to generate one
purified state
fig 33  State not recycled!
 Surface Code
 Assume error rate 1/10th threshold (99.95% fidelity)
Logical memory qubit from array of physical qubits
 1,000 smaller error rate: ~600 physical qubits
 1,000,000 smaller error rate: ~2,000 physical qubits
 1,000,000,000 smaller rate: ~4,500 physical qubits
Circuit to demonstrate topological CNOT:
 With 1,000 smaller error rate: ~1,800 physical qubits
Prime factoring with Shor’s algorithm:
 Factor a 15 bit number (105): ~40,000,000 qubits
 Factor a 2000 bit number (10600): ~1,000,000,000 qubits
~99% of factoring computer is “A‐state factories”
Fowler et al. (PRA, 2012)