SIN Suppl AL2023 02 01

You might also like

Download as pdf or txt
Download as pdf or txt
You are on page 1of 9

Deterministic photon source

interfaced with a programmable


silicon-nitride integrated circuit:
supplemental document
This document provides supplementary information for ”A bright single-photon source towards
on-chip functionalities”. [Main]

1. SOLID-STATE SINGLE-PHOTON SOURCE PLATFORM


A. Material and Device
The GaAs wafer used to fabricate the SPS device consists of a substrate with DBR, a sacrificial
layer, and a photonic membrane (MB) of p-i-n with InAs QDs in the middle shown in Fig. S1.
The wafer was grown using the molecular beam epitaxy (MBE) method on a (001)-oriented GaAs
substrate. Firstly, the buffer layer of intrinsic GaAs and supper lattice AlAs/GaAs was grown
on the top of the GaAs substrate at 620◦ . The DBR composed of 15 pairs of 79 nm AlAs/ 66 nm
GaAs was introduced and the SL of Al0.75 Ga0.25 As with a thickness of 1.15 µm followed with the
same growth temperature. The MB is 176.5 nm thick and the growth temperature was lowered
to 600◦ . The MB started with 7.5 nm GaAs, 8.5 nm Si-doped GaAs with doping density of
2×1018 cm−3 and tunnel barrier layer of 41 nm GaAs followed. QDs were formed by depositing
1.5 monolayers of InAs. Afterward, 0.3 nm AlAs, and 2 nm GaAs capping layer was grown
consecutively and flushed using annealing step [1]. The MB was completed with 7.2 nm GaAs,
47 nm barrier layer of Al0.3 Ga0.7 , 5 nm medium C-doping density ( 2 × 1018 cm−3 GaAs, 25 nm
high C-doping density (1 × 1019 cm−3 ) GaAs sequentially. Modulation density of InAs QDs grown

Fig. S1. The layout of the material growth and Ohmic contacts utilized to make the SPS de-
vices.

using Stranski-Krastanov mode has been exploited recently [2]. Here, the InAs adopted a 15 nm
gradient grown GaAs buffer, resulting in a QDs density modulation shown in Fig.S2 (a), curved
stripes distribute along the X direction. Where the intensity is an integration of the full spectrum
of QDs emission full at each position. Fig.S2(b) Displays the intensity distribution in the middle
of the wafer with respect to the emission wavelength along the X direction. Curved stripes with
respect to the wavelength are observed with a comparable big dip originating from an interplay of
the dielectric stack (DBR, sacrificial layer, and photonic membrane) [3]. In addition, the wetting
layer luminescence is absent in Fig.S2(b) attributed to the 0.3 nm AlAs layer [4]. The wafer-scaled
PL map is recorded as a reference to cleave a sample from the wafer accordingly.

B. SPS effiency
The QDs-based solid-state single photon source (SPS) efficiency is mainly limited by the collection
rather than the generation efficiency due to optical components loss and out-coupling of the
(a) ×107 (b) ×105
2.0 2.5

60 60 2.0

Intensity (cts/s)

Intensity (cts/s)
Y [mm] 1.5

X [mm]
40 40 1.5
1.0
1.0
20 20
0.5
0.5
0 0
0 25 50 0.0 900 950
0.0
X [mm] Wavelength [nm]

Fig. S2. Photoluminescence (PL) map of the full 3-inch wafer recorded at 80 K, where QDs are
excited with a 516 nm laser. (a) The intensity distribution of QDs emission full spectrum en-
semble for the whole wafer spatially labeled with respect to X and Y direction. (b) Spectral map
the direction of X from the center of the wafer( Y=35 mm, the black dash line in (a)). Which
shows QDs emission ranges from 880 nm (excited state luminescence) up to 990 nm (red part of
the inhomogeneously broadened QD spectra).

emitted photons. In order to boost the collection efficiency, we improved the material growth
and nano-device design. The thickness of the SL determines the fundamental mode confinement
of the light as well as the efficiency of the shallow etching grating (SEG) couplers [5]. Therefore
optimized SL thickness enables the back-reflected downward light constructively interfere with
the upward light. However, the poor reflectivity of the GaAs undermines the SEG capability.
An additional layer of DBR was implemented below the SL shown in Fig. S1 which possesses
much higher reflectivity and enables maximal reflection of the downwards-scattered light. The
QD couples to a planar waveguide and emits single photons in dual directions. A photonic
crystal mirror is placed instead on one side enabling to reflect the photons back to the collection
direction only, as shown in Fig. S1 (a). The SEG efficiency is 90% accounted for the DBR, which is

Table S1. Comparison of the improved component’s efficiency of current single-sided PCW
SPS with previous two-sided PCW [6].
Component two-sided PCW [6] single-sided PCW
SEG (50±1)% (90 ± 1)%
SEG to Fiber (59±1)% (64 ± 1)%
Directionality 50% 100%
Etalon 87% 67%
Total efficiency 6.8% 21.5%

comparable to the simulation value of over 92% [6], corresponding to efficiency improvement by
a factor of 1.8. In principle, the directionality of the single-sided PCw (which can not be measured
directly) is supposed to be less than 1 simply because we can not fabricate perfect photonic
crystals to date. We can confidently estimate directionality approaching 1 due to our optimized
fabrication process minimizing the fabrication imprecision. Thereby, the total efficiency is ready
to enhance by a factor of 3.6. However, the etalon used for phonon-side band filtering during the
period for the current SPS permitted an efficiency of 67%. Consequently, the total efficiency of
the current SPS only raches 21.5%, improved by a factor of over 3 compared to the previous SPS
[6]. Commercialized Etalon filters with 92% efficiency are generalized. In addition, the coupling
efficiency of SEG to fiber is ready to improve. The mode mismatch of the SEG and the collection

2
fiber lens only allows 64% of the photons to couple into the fiber. A pair of cylindrical lenses
replace the present Gaussian lenses, a coupling efficiency of 95% is achievable. Accordingly, our
SPS total efficiency is ready to reach 42 %, which is comparable to the latest reported out-plane
QDs-based SPSs [7, 8].

C. Nano-fabrication

Fig. S3. (a) Scanning electron microscope (SEM) image of the photonic crystal waveguide
where one side is terminated by a row of holes shown in the zoomed-in SEM image. (b) The
p-i-n diode behavior of applying an external DC bias at the cryogenic temperature of 1.6 K.
The green area is the bias required for the population of the neutral exciton. The corresponding
currents are in the order of 1-10 muA.

The fabrication to realize our devices is a soft-mask-based technique [9]. The procedure is
composed of ohmic contact deposition, nano-structures patterning, and membrane release [6].
Further optimization of the recipe to ensure a high yield. The metal gate quality is evaluated by
the I-V curve shown in Fig. S3 [10] where we observe equipment-resolution scaled low current
at reversed bias. The current exponentially increases while the applied bias crosses the ’turn on’
point( 0.8 V). The bias range demanded to populate neutral excitons is the green area, illustrating
a corresponding current of a few µA. The contact yield, which is a key element for scaling up the
devices, is estimated to be over 90% statistically [11].

(a) (b)
1.0 1.0
Transmission [arb.]

Transmission [arb.]

f=318.233
0.5 0.5

0.0 0.0 f=311.289


317.5 320.0 322.5 325.0 315 320 325
Frequency (THz) Frequency [THz]

Fig. S4. The transmission profile of the PCW on which both sides are terminated by SEG grat-
ings. (a) The transmission spectra of the 11 PCW devices were designed with identical lattice
constant a=240 nm and radius r=64 nm. (b) The transmission spectrum of the PCW with a=248
nm, and r=66 nm, is identical to the single-sided PCW. Where the black dash line denotes the
PCW band-edge ( 311.289 THz) and the QD emission frequency (318.233 THz) respectively.

The fabrication accuracy of the photonic crystals’ hole diameter, measured on more than
10 devices, is < 6 nm. Accordingly, the PCW band-edge frequency fluctuation is observed in
Fig. S4 (a), where 11 devices were characterized. The fluctuation is up to 3 nm caused by the
inhomogeneous resist thickness. The fabrication imprecision of GaAs devices on such a scale is
quite low. Thus, we can refer to the two-sided PCW transmission spectrum and spectrally search

3
QDs for single-sided PCW whose emission frequencies are beyond the band-edge as well as close
to the transmission peak shown in Fig. S4 (b).

D. Neutral exciton characterization

SNR
Intensity [cts]
(a) (b)
2500 5000 7500 10000
100000 200000

Frequnecy [THz]
Frequnecy [THz]

318.25
318.25

318.23
318.23
1.24 1.26 1.28 1.30
1.24 1.26 1.28 1.30
(c) (d) Voltage [V]
Voltage [V]
×105 105 γ=(1.551±0.002) /ns
data
Intensity [cts]

Intensity [cts]
5 5.88 GHz fit
103

0
1.26 1.27 1.28
Voltage [V] 0.0 2.5 5.0 7.5 10.0
Time [ns]

Fig. S5. (a) Resonantly pump the neutral exciton with a weak continuous-wave tunable diode
laser. Two bright lines are charge plateaus arising from the fine structure splitting of the neu-
tral exciton. (b) The signal-to-noise ratio (SNR) is recorded while carrying out charge plateau
measurements in (a). (c) The spectrum of the black dash line cut area is plotted and the fine
structure splitting is measured around 5.88 GHz. (d) Lifetime measurement performed on the
right dipole in plot (c). The background signal is measured (light green curve).

The single-sided PCW used to generate single photons is shown in Fig. S1 (b), where the QD
candidates are targeted in the slow-light region [6] shaded by purple. The slow-light region
width is extended to 15 µm, which not only increases the probability of finding emitters, but it
ensures the ideal emitter’s position is far away from the interface to the fast-light region. The QD
defined for the generation of single photons is spatially very close to the center of PCW along
the propagation direction. The distance between the excitation spot and the collection SEG is
around 60 µm seen in Fig.S3 (a). Clear charge plateaus of the neutral exciton are seen in Fig.S5 (a).
Correspondingly, the SNR is measured as high as 104 S5 (b). Which mainly attribute to optimized
excitation beam spot and polarization minimizing the laser leaking into the waveguide mode,
as well as the proper distance between excitation and collection spot reducing the scattering
light into the collection. Moderate fine structure splitting is measured on this neutral exciton
(Fig. S5 (c)), where the left dipole is coupled better to the PCW than the right one. We optimized
the pulse length using a 4f grating system. We managed to populate the left and right dipoles
with different polarizations and measured the lifetime separately. Only single exponential decay
curves are observed and we do not see two dipoles decay beating. This suggests that the fine
structure splitting is big enough and one pulse can not populate both dipoles simultaneously.
The right dipole shows a slightly faster decay rate of 1.53 ns−1 simply because the right dipole
is spectrally closer to the band edge. The bumps in the background signal curve arise from the
inference on the setup. The fitting model is a single exponential function with the deconvolution
of the Instrument response function (IRF) of the setup.

E. Purity
The purity of the SPS is evaluated by g2 (0). Conventionally, a stream of single photons is sent to
the Hanbury-Brown and Twiss (HBT) setup, and photon counts of the two outports are correlated,

4
×103

Normalized Coincidence cts


(a) (b)
data

Coincidence cts
1.02 𝜏=5.3𝜇s
fit
2
1.00

0 0.98
−40 −20 0 20 40 0 20 40
Time [𝜇s] Time [𝜇s]

Fig. S6. QD emission blinking characterization by g2 correlation in long time scale. (a) A peak
appears in the center of the correlation histogram arising from the multi-photon emission con-
tribution. (b) Integrate coincidence counts under each pulse area and normalize them to long-
time counts. Single exponentially fit gives the blinking time is a few µs and magnitude is 2%.

see Fig.2 (c) in the main text, where only shows the anti-bunched central peak and its nearby
peaks. g2 (0) is the normalization of the central area to the long-time area after the fit. On-chip
g2 (0) is measured in a similar way using the SiN circuit, see Fig.S8 (a) up-panel. In1 is interfaced
with the SPS and activates the PhIA ensuring a 50:50 beamsplitter. The correlation histogram is
recorded as shown in Fig.2(d) in the main text, where g2 (0) is the normalization of the central
area to the average of all the side peak areas with subtraction of the background. The uncertainty
is estimated from the standard deviation of the g2 (0) obtained from the Monte-Carlo simulation.
The g2 (0) measured from free-space HBT setup is 0.8 % at π pulse excitation. while g2 (0) of
0.4 % is obtained from the on-chip measurement, which is much lower. As we see in Fig.2 (e) in
the main text, the background line is much higher than 0 because the fiber used cannot shield
room light completely during the measurement. Calculation of the g2 (0) with subtraction of the
background gives the low bound value.
The g2 (0) value of this SPS is so low simply because of the negligible laser leakage and weak
blinking. The signal-to-noise ratio can reach 1000 at π pulse excitation. This suggests the leaked
laser contribution to g2 (0) is less than 0.1%. Moreover, high-quality metal gates maximally
quenched the charge fluctuation thereby reducing multi-photon emission see Fig.S6. Where a
very tiny bunching peak is observed in a long-time g2 correlation histogram. This suggests the
QD exciton state blinks due to the charge environment. The blinking amplitude is very low
around 2% in a time scale of a few µs which is much longer than the lifetime of QD. Thus, high
purity of single photons is generated.

F. Indistinguishability
Conventionally, the indistinguishability is characterized by the Hong-Ou-Mandel setup. See
Fig.2 (d) in the main text, the correlation histograms of co-polarization and cross-polarization are
recorded separately. The raw indistinguishability is ( Across − Aco )/Across . Across and Aco is the
ratio of the central area to the long-time area with the respect to the cross and co-polarization
histogram after the fit. To get the intrinsic value of the indistinguishability, the raw value is
calibrated with the HOM setup imperfection and the g2 (0) contributed from the laser leakage
using Eq. S1. HOM setup imperfection includes the imbalanced beamsplitter (R ̸= T) and
interference offset measured by classical light fringes contrast ϵ.

( R2 + T 2 1
Vintrinsic = Vraw (S1)
2RT (1 − ϵ)2

The HOM visibility value in the main text is corrected only with the factor of the HOM setup
imperfection. Since the SNR was approximately 1000, the laser leakage can be considered
negligible and its correction was not taken into account.
On-chip indistinguishability is characterized by an MZI circuit shown in Fig.S8(a) up-panel,
where two streams of single photons feed through 1n1 and In2 simultaneously. The phase shift is
performed using the MZI_A phase shifter applying external bias. The out1 and out2 coincidence

5
event is recorded see Fig.2 (f), The coincident counts respond cosineally with a phase shift, and
model the curve using the cosine function. The indistinguishability is calculated by V=(Imax -
2Imin )/Imax [12]. Where Imax is the maximum intensity obtained from the fit while Imin is the
minimum intensity.

2. SIN PLATFORM
A. Setup losses
The pulse repetition rate of the excitation laser is 72.6 MHz. The photon count rate after going
through all of the components of the setup was up to 40 kHz for a good input-output configuration
on the chip and down to 10 kHz for a bad one as measured on our SNSPDs. We should therefore
account for 32 to 36 db of losses through our setup, which we try to describe in this section.
Due to technical difficulties with the etalon, which we could not replace at the moment of the
experiment the source efficiency was 5.5%, which is less than the previously measured value of
21.5%, accounting for an initial 12.6 db of losses.
The transmission through the two mode demultiplexer was measured to be 25% with an
optimized polarization. However, due to polarization drifts of the single photons over the course
of many minutes, the transmission is estimated to be on average 20% for an hour-long experiment.
This accounts for 7 db of losses.
The insertion losses on the chip are hard to estimate since the light goes through an interferom-
eter with many different outputs where the intensity at each output depends on the phase offsets
of the interferometer0000000000000.0
We measured the transmission through a test structure close to the interferometer and found
the losses to be of 8.5 db, suggesting insertion losses of 4.25 db per facet. However these losses
are expected to be significantly higher for the couplings to the interferometer, as it is harder to
align all of the outputs of the fiber array simultaneously than to align only two of them, and the
propagation distance is much larger. An upper bound for the coupling losses was estimated to be
of 7 db by looking at the input-output transmission of the worst channel in the interferometer
after phase optimization. The coupling losses in the chip can therefore account for 8.5 to 14 db of
the total losses.
Our SNSPDs have efficiencies around 80% as measured by direct characterization, and the fiber
to fiber losses among all of the components of the setup can be estimated to be of 1-2 db in total,
accounting for extra 2-3 db of losses. Summing all these losses together give a total loss of with a
lower bound of 30 db and an upper bound of 37 db, which are in agreement with the count rates
measured at the SNSPDs. An overview of this analysis is summarized in table S2

Table S2. Losses found through the different setups of the experiment
Component Expected loss
Source setup 12.6 db
Two-mode demultiplexer 7 db
SiN chip 8.5 to 14 db
Fibers 1 to 2 db
SNSPDs 1 db
Total expected losses 30 to 36 db
Total measured losses 32 to 36 db

B. Circuit and analysis of infidelities


In this section, we analyse in more details the integrated circuits implemented for the bosonic
suppression laws and post-selected Bell-state experiments, investigating noise models and com-
paring them to the theoretical expectations to describe the origin of the remaining experimental
discrepancies.
The circuit is schematized in Fig. S8 (a), where Phi_A, Phi_B, MZI_A and MZI_B indicate
programmable phase shifters. The DFT unitary utilized for implementing the suppression laws is

6
(a) (b)
Single phase, D1+D3 Single phase, D1+D4
Double phase, D1+D3 Double phase, D1+D4

1.0

Norm. Counts
0.5
Measured transmi�vity
Op�mal transmi�vity
Experiment wavelength 0.0
0 20 40 60 80 100 120 140

Fig. S7. (a) Measured transmitivity of the directional couplers in the chip for different wave-
lengths. Ideally, the transmitivity should be 50%, but for our experiment, it is about 62%. (b)
The plot of the quantum interference fringes is used to characterize the post-selected Bell state.
The single-phase data are for scanning Phi_A only and the double-phase data are for simultane-
ous scans of Phi_A and Phi_B. The D number represents the output number.

obtained with MZI_A=MZI_B=π/2. The post-selected entanglement is instead obtained after the
passive circuit part highlighted in S8 (a), while the programmable phases implement different
single-qubit projections. In particular, a Pauli Z projection is implemented setting MZI_i=0 on the
associated i-th qubit, while Pauli X (Y) is obtained setting MZI_i=π/2 and Phi_i=0 (π/2).
For the post-selected Bell-state experiment, the fidelity of the state tomography achieved was
of 91%. This fidelity was achieved by taking measurements in different base pairs through scans
of scans of the on-chip thermal phaseshifters, labeled Phi_A and Phi_B in. Fig. S8 (a). Fig. S7
(b) shows the raw data of 4 of such scans. It can be seen that scanning only one phase changes
the output distribution with half the frequency of scanning both phases, which is expected from
changing one measurement basis versus changing both of them simultaneously. The fitted curve
for every scan is subsequently fed to a higher dimensional fit to get the final fidelity of the state.
To understand why the fidelity is non-ideal we made simulations of the experiment where it was
possible to include the errors we expected to have. The simulations were made in Python by
calculating the permanents of the corresponding unitary matrices for the photonic circuit. For
this error analysis only the results for the X1 X2 , Y1 Y2 and Z1 Z2 measurements bases will be used.
The state tomography takes into account all of the values of the density matrix, but a different
measure of the fidelity of the generated Bell state can be achieved by comparing only the X1 X2 ,
Y1 Y2 and Z1 Z2 bases, which makes simulation easier. The fidelity of the experiment compared
to the perfect theoretical case is then 94%. One source of error in the experiment is that due to a
fabrication error the on-chip directional couplers had an imperfect split ratio. We measured this
ratio to be consistently around 62/38 instead of 50/50 for the structures we were working on, as
can be seen in figure Fig. S7 (a). Building this mistake into the simulation yielded a fidelity of
96.7%. The second error to be treated is photonic distinguishability, which gave rise to a HOM
visibility of 94.53 ± 1.70% (Fig. 2c). This error can be simulated by treating the distinguishability
of the photons as a variable beam splitter which probabilistically sends away some of the photons
to separate interferometers, such that they cannot interfere. Adding both error sources into the
simulation gives the output distribution plotted with red bars in Fig. S8. (b), which should
be compared to the experimental results on the same figure. The fidelity between experiment
and simulation is of 99.4% suggesting that these two errors are enough to explain most of the
discrepancies between theory and experiment. A source of error not taken into account in the
simulation is the finite g(2) (0) of the photons, which by definition is the source of the very
small discrepancies for the measurements in the Z1 Z2 basis, but its presence has a negligible
contribution to the fidelity between experiment and simulation.
For the bosonic suppression laws experiment a similar analysis can be performed in order
to understand the discrepancies between theory and experiment. The right plot in Fig. S8 (c)
shows the expected distribution in the case with no errors, the plot in the center shows the
simulation result when photonic distinguishability and non-ideal splitting ratio of the directional-
couplers is taken into account, and the plot on the left shows the experimental data. In this
case, the simulation results and the experimental results are significantly different, meaning
that other errors are playing a role in the result. One posibility is the difference in the coupling

7
(a) (b)
X-basis Y-basis Z-basis
Phi_A MZI_A Out 1 0.5

Experiment Simulation
In 1 0.35

Probability
Out 2 0.2
In 2
0
Out 3
In 3 0.2
0.35
Phi_B MZI_B Out 4
In 4 0.5
|00 |01 |10 |11 |00 |01 |10 |11 |00 |01 |10 |11
Entanglement generation State analysis
State
(c)
Theoretical probability Simulation result Experimental result 0.5

(1,2) (1,2) (1,2)


0.4
Input configuration

(1,3) (1,3) (1,3)

(1,4) (1,4) (1,4) 0.3

(2,3) (2,3) (2,3) 0.2

(2,4) (2,4) (2,4)


0.1
(3,4) (3,4) (3,4)

(1,2) (1,3) (1,4) (2,3) (2,4) (3,4) (1,2) (1,3) (1,4) (2,3) (2,4) (3,4) (1,2) (1,3) (1,4) (2,3) (2,4) (3,4) 0.0
Output configuration Output configuration Output configuration

Fig. S8. (a) On chip interferometer design. (b) Correlation data for Pauli X1 X2 , Y1 Y2 , and Z1 Z2
qubit measurements on the generated Bell state ψ+ . Top histograms are measured expecta-
tion values, while bottom values are values obtained from simulation. (c) Bosonic suppression
laws experiment input-output distributions for the perfect theoretical case, simulation with
imperfections taken into account, and experimental results.

efficiency between different modes of the chip, which apart from being intrinsically different due
to fabrication imperfections, can also vary between two different runs of the experiment due to
lack of precision in the alignment. This error was probably not relevant in the post-selected Bell
state experiment as the setup did not have to be realigned within one run of the experiment.

8
REFERENCES
1. A. Ludwig, J. H. Prechtel, A. V. Kuhlmann, J. Houel, S. R. Valentin, R. J. Warburton, and A. D.
Wieck, “Ultra-low charge and spin noise in self-assembled quantum dots,” J. Cryst. Growth
477, 193–196 (2017).
2. N. Bart, C. Dangel, P. Zajac, N. Spitzer, J. Ritzmann, M. Schmidt, H. Babin, R. Schott,
S. Valentin, S. Scholz et al., “Wafer-scale epitaxial modulation of quantum dot density,” Nat.
communications 13, 1–7 (2022).
3. H. G. Babin, J. Ritzmann, N. Bart, M. Schmidt, T. Kruck, L. Zhai, M. C. Löbl, G. N. Nguyen,
C. Spinnler, L. Ranasinghe et al., “Charge tunable gaas quantum dots in a photonic nip diode,”
Nanomaterials 11, 2703 (2021).
4. M. C. Löbl, S. Scholz, I. Söllner, J. Ritzmann, T. Denneulin, A. Kovács, B. E. Kardynał, A. D.
Wieck, A. Ludwig, and R. J. Warburton, “Excitons in ingaas quantum dots without electron
wetting layer states,” Commun. Phys. 2, 1–7 (2019).
5. X. Zhou, I. Kulkova, T. Lund-Hansen, S. L. Hansen, P. Lodahl, and L. Midolo, “High-efficiency
shallow-etched grating on gaas membranes for quantum photonic applications,” Appl. Phys.
Lett. 113, 251103 (2018).
6. R. Uppu, F. T. Pedersen, Y. Wang, C. T. Olesen, C. Papon, X. Zhou, L. Midolo, S. Scholz, A. D.
Wieck, A. Ludwig et al., “Scalable integrated single-photon source,” Sci. advances 6, eabc8268
(2020).
7. N. Tomm, A. Javadi, N. O. Antoniadis, D. Najer, M. C. Löbl, A. R. Korsch, R. Schott, S. R.
Valentin, A. D. Wieck, A. Ludwig et al., “A bright and fast source of coherent single photons,”
Nat. Nanotechnol. 16, 399–403 (2021).
8. H. Wang, Y.-M. He, T.-H. Chung, H. Hu, Y. Yu, S. Chen, X. Ding, M.-C. Chen, J. Qin, X. Yang
et al., “Towards optimal single-photon sources from polarized microcavities,” Nat. Photonics
13, 770–775 (2019).
9. L. Midolo, T. Pregnolato, G. Kiršanskė, and S. Stobbe, “Soft-mask fabrication of gallium
arsenide nanomembranes for integrated quantum photonics,” Nanotechnology 26, 484002
(2015).
10. S. M. Sze, Y. Li, and K. K. Ng, Physics of semiconductor devices (John wiley & sons, 2021).
11. C. Papon, Y. Wang, R. Uppu, S. Scholz, A. D. Wieck, A. Ludwig, P. Lodahl, and L. Midolo,
“Independent operation of two waveguide-integrated single-photon sources,” arXiv preprint
arXiv:2210.09826 (2022).
12. J. C. Adcock, C. Vigliar, R. Santagati, J. W. Silverstone, and M. G. Thompson, “Programmable
four-photon graph states on a silicon chip,” Nat. communications 10, 3528 (2019).

You might also like