a. c. hughes, j. f. goodwin, m. a. sepiol, j. e. tarlton ... · single-qubit memory error metrology...
TRANSCRIPT
Single-qubit memory error metrologyat short timescales
Ion Trap Quantum Computing Group - Department of Physics, University of Oxford
[1] D. T. C. Allcock et. al., App. Phys. Lett. 102, 044103 (2012) [2] T. P. Harty et. al., Phys. Rev. Lett. 113, 20501 (2014) [3] T. P. Harty et. al., Phys. Rev. Lett. 117, 140501 (2016) [4] E. Magesan et. al., Phys. Rev. Lett. 109, 080505 (2012)
[email protected] [email protected] Department of Physics, Clarendon Laboratory,Parks Road, Oxford, OX1 3PU, United Kingdom
www.physics.ox.ac.uk/users/iontrap
- Surface (chip) rf trap featuring integrated microwave circuitry [1]
- Gates driven by near-field microwaves
- Heating rate 1.4(3) quanta/ms, axial secular
frequency 2π × 500 kHz
"Atomic clock" qubit
A. C. Hughes, J. F. Goodwin, M. A. Sepiol, J. E. Tarlton, C. J. Ballance, D. P. Nadlinger, T. P. Harty, A. M. Steane, D. M. Lucas
397 Doppler Ca oven
866
397 σ+
854PI, 850σ+
393 σ+
RFDC
DCgate μ.w.
RF
DCstate prep. μ.w.
groundμ.w. + DCDC controlRF + μ.w. + DC
850 π
B0
4S1/2
3D3/2
3D5/2
4P1/2
4P3/2
854nm
850nm
866nm
397nm
393nm
mF=4
mF=3
mF=-3
mF=-4
mF=1
F=0
3.2 GHz
4S1/2, F=4
4S1/2, F=3
m
B0 = 146 G
ψ C1 C3C2τ τ Cm+1
example of Cliffordgate decomposition
C1 C3C2 Cm+1
C1 C3C2τ τ
Yπ/2
ΦπΦπ 3Φ2πXπ/2
Cm+1
C4 m τ+
Φπ3Φ2πYπ/2
Xπ/2 Yπ/2Xπ/2
Φπ~ ~ ~
BB1 sequencedecompositionsΦ=Φ+ 180°~
ψ
ψ x8
x8
SRS Rb frequency reference
Rohde &Schwarzsignal
generator
DDSfrequencyoctupler
Trap Measuring short-timescale decoherence
- "Memory error" over a time delay is often measured using Ramsey experiments to extract a T2* time
- We measure T2* ~ 1 minute
π2
π2
delay
0° φ°φ + 180°
stateprep.
- At fault-tolerant threshold (~10-2 – 10-4), εmemory ~ εSPAM so measurement is inefficient
- Information about computationally relevant
timescales is usually extrapolated from longer times
- Instead, we use randomised benchmarking [4], [5]
to directly measure very small errors (~10-6)
Randomised benchmarking (RBM) results
- Ramsey experiments struggle when εmemory ~ εSPAM (10-3)
- RBM allows probing of much smaller errors
- We measure εmemory< 10-4 up to ~100 ms – at least 3 orders of magnitude
longer than gate or measurement times
- Extrapolating from data at T2*-like timescales overestimates
memory errors at short timescales by an order of magnitude
Detuning the field- Shuttling ions between regions of larger traps could improve scalability [6]
- During shuttling, an ion is likely to
travel through areas of less well- controlled magnetic field strength
- We consider memory errors incurred
by detuning the field from the clock field without adjusting the microwave frequency
- Qubit remains robust to memory
errors over typical shuttling times at large detunings
- εmemory < 10-3 maintained for 20 ms
even at 25 mG detuning
Dynamical decoupling- Including a simple CPMG dynamical decoupling sequence of repeated X gates during the delay periods, at a rate of
1 per 100 ms, reduces εmemory to < 10-3 for longer than 1 s
- Memory performance at computationally relevant timescales is limited by microwave local oscillator
noise (see below)
Sources of decoherenceRBM sequences and BB1 gates- RBM involves m periods of dephasing τ instead of a single one, and
interleaving the delays with gates randomly sampled from the single-qubit Clifford group
- Memory error is enhanced and adds incoherently
- Error per delay can be calculated by comparing the fidelity of interleaved
sequences vs sequences without the interleaved delays
- Each Clifford is composed of +-Xπ/2, +-Yπ/2 rotations
- Error per Clifford is very
sensitive to drifts in microwave power - could limit accuracy
of the RBM experiment
- Decomposing each π/2 rotation
into a BB1 composite pulse [7] makes them more robust to these pulse area errors
- Memory data is well-approximated by a fit to a combination of white and 1/f frequency noise
- CPMG sequence suppresses the 1/f component
(possible candidate: varying AC Zeeman shifts due to movement of the ion in the trapping field)
- Noise on each component of microwave chain
measured by duplicating chain and making any noise upstream of the component in question common mode (at dashed lines)
We achieve a coherence time of 1 minute in a 43Ca+ qubit, but the relevant timescale for fault-tolerant quantum computing is the time for which the memory error remains < 10-2 – 10-4. Information regarding these initial stages of decoherence is usually provided by extrapolation of models only verified at long timescales. Here we use randomised benchmarking techniques to directly measure errors at the 10-6 level and to characterise the underlying noise spectrum. We find εmemory < 10-4 up to ~ 100 ms (far longer than predicted from the T2* time). Using a simple CPMG dynamical decoupling sequence, we observe εmemory < 10-3 for storage times greater than 1 s. At this level, we conclude that our memory performance is primarily limited by microwave local oscillator noise.
10-3 10-2 10-1 100 101
Delay τ (s)
10-7
10-6
10-5
10-4
10-3
10-2
10-1
SPA
M-c
orre
cted
mem
ory
erro
r
Clock field RBM dataFit to white and pink frequency noiseB field 10 mG detunedB field 25 mG detunedB field 50 mG detunedClock field with dynamical decouplingInferred error due to oscillator noise
- Direct measurement of noise on Rb reference oscillator agrees closely with white noise component of fit to RBM data
SPAM Single-qubit gate Two-qubit gate
Error ~10-3 ~10-6 ~10-3
ref [2], [3]
10-3 10-2 10-1 100 101 102
Delay τ (s)
10-7
10-6
10-5
10-4
10-3
10-2
10-1
SPA
M-c
orr
ecte
d m
emor
y er
ror
RBM dataFit to white and 1/f frequency noiseRamsey dataExponential fit to Ramsey delay > 1 s
- 43Ca + hyperfine ground state clock qubit at 146 G
- Initialisation: optical pumping and microwave transfer pulses ~ 3.2 GHz
- Readout: "shelve" → 3D5/2 via 4P3/2, + fluorescence detection
[5] P. J. J. O'Malley et. al., Phys. Rev. Applied 3, 044009 (2015) [6] D. Kielpinski et. al., Nature 417, 709711 (2002)[7] S. Wimperis, J. Mag. Res. 109, 221231 (1994)