PHYSICAL REVIEW APPLIED 15, 054059 (2021)
Integrated Magnetometry Platform with Stackable Waveguide-Assisted Detection
Channels for Sensing Arrays
Michael Hoese ,
1,†
Michael K. Koch ,
1,2,†
Vibhav Bharadwaj ,
3
Johannes Lang,
1
John P. Hadden ,
4
Reina Yoshizaki ,
5
Argyro N. Giakoumaki,
3
Roberta Ramponi,
3
Fedor Jelezko,
1,2
Shane M. Eaton,
3
and Alexander Kubanek
1,2,*
1
Institute for Quantum Optics, Ulm University, Ulm D-89081, Germany
2
Center for Integrated Quantum Science and Technology (IQst), Ulm University, Ulm D-89081, Germany
3
Institute for Photonics and Nanotechnologies (IFN) - CNR, Piazza Leonardo da Vinci, 32, Milano 20133, Italy
4
School of Physics and Astronomy, Cardiff University, Cardiff CF24 3AA, United Kingdom
5
Department of Mechanical Engineering, School of Engineering, The University of Tokyo, Tokyo 113-8656, Japan
(Received 22 December 2020; revised 26 April 2021; accepted 10 May 2021; published 26 May 2021)
The negatively charged nitrogen vacancy (N-V
−
) center in diamond has shown great success in
nanoscale, high-sensitivity magnetometry. Efficient fluorescence detection is crucial for improving the
sensitivity. Furthermore, integrated devices enable practicable sensors. Here, we present an integrated
architecture which allows us to create N-V
−
centers a few nanometers below the diamond surface, and at
the same time covering the entire mode field of femtosecond-laser-written type-II waveguides. We exper-
imentally verify the coupling efficiency, showcase the detection of magnetic resonance signals through
the waveguides and perform proof-of-principle experiments in magnetic field and temperature sensing.
The sensing task can be operated via the waveguide without direct light illumination through the sam-
ple, which is important for magnetometry in biological systems that are sensitive to light. In the future,
our approach will enable the development of two-dimensional sensing arrays facilitating spatially and
temporally correlated magnetometry.
DOI: 10.1103/PhysRevApplied.15.054059
I. INTRODUCTION
Quantum sensing performed with negatively charged
nitrogen vacancy (N-V
−
) centers in diamond has success-
fully measured strain [1], temperature [2], and magnetic
fields [3–5] at the nanoscale and with high sensitivity.
N-V
−
-based magnetometry is widely applied to character-
ize biological samples such as cells [6] or complex materi-
als [7,8]. The combination with atomic-force microscopy
has revolutionized scanning-probe magnetometry [9,10]
with applications in material science and life sciences.
Advanced photonics could further improve the perfor-
mance in terms of losses, signal-to-noise, and operation
speed [11,12] and includes integration into optical fibers
[13,14], diamond nanopillar arrays [15,16], and integrated
photonics [17–19]. The latter enables, in addition, the real-
ization of compact devices. In this context, femtosecond
laser writing is an outstanding fabrication method that
does not rely on lithography steps and has, in partic-
ular, the capability to fabricate three-dimensional struc-
tures. Using diamond as the photonics host offers the
*
Corresponding author. alexander.kubanek@uni-ulm.de
†
These authors contributed equally to this work.
additional advantage that N-V
−
centers can be integrated
directly into the photonic device. Laser-written type-II
waveguides in diamond have recently been developed
[20,21] and show great potential to efficiently interface
N-V
−
centers. Thereby, a focused laser beam creates two
nearby lines of reduced refractive index, where the stressed
region between the two lines serves as a waveguide. The
controlled creation of defect centers has been demon-
strated with ion implantation [22] and laser writing [23–
25]. However, until now there has been no strategy that
enables deterministic postprocessing of each waveguide
in a three-dimensional platform, a problem that becomes
even more challenging when color centers are required
close to the diamond surface as is the case for N-V
−
-based
magnetometry.
Here, we present an integrated approach that is capable
of functionalizing each waveguide in a three-dimensional
architecture individually with N-V
−
centers as depicted in
Fig. 1(a). After laser writing the waveguides, we create
N-V
−
centers through shallow implantation of nitrogen
ions on the front facet of the diamond photonics platform.
The implantation depth of a few nanometers below the dia-
mond surface enables the combination of efficient photon
routing through diamond waveguides with sensing tasks on
2331-7019/21/15(5)/054059(11) 054059-1 © 2021 American Physical Society
MICHAEL HOESE et al. PHYS. REV. APPLIED 15, 054059 (2021)
x
y
z
2.85 2.90
MW frequency (GHz)
0.0
0.5
1.0
1.5
2.0
Magnetic field (mT)
(a)
(c)
(d)
(b)
V
FIG. 1. (a) Sketch of the waveguide-assisted sensor array. Shallow-implanted N-V
−
centers act as local magnetometers on the
diamond surface. The N-V
−
centers are optically accessed by means of laser-written type-II waveguides and low-numerical-aperture
(low-NA) optics. The architecture can be extended to sensor arrays. The sketch incorporates microscope images of the waveguides
in top and front view. (b) Sample fabrication process. First, the waveguide walls (black lines) are laser written from the top into the
diamond. Second, nitrogen ions are shallow implanted into the front facet to form N-V
−
centers after annealing. (c) Level scheme
and dynamics of the N-V
−
center as utilized to perform optically detected magnetic resonance (ODMR) spectroscopy. Dashed arrows
depict phonon transitions and bold arrows photon transitions, respectively. The arrow width reflects the interaction strength. (d) Model
for the splitting of the m
s
=±1 ground states with increasing magnetic field. The dashed lines indicate the magnetic field strengths
that we apply in the experiments to perform ODMR scans.
the diamond surface. Our architecture separates the optical
access to the N-V
−
centers from the object to be sensed by
excitation and detection through the waveguides, in a simi-
lar way to demonstrations based on total internal reflection
[26,27]. Therefore, as with common confocal detection
[28] and widefield imaging [29] approaches, good sens-
ing performance can be achieved without the need of
light exposure through the sample. However, the thickness
of the diamond sample can be much larger in our case.
Ultimately, the architecture can be extended to multiple,
individually addressable waveguides, thus forming a two-
dimensional sensing array with potential for time-resolved
and spatially correlated magnetometry in material and life
sciences.
In the following, we discuss the working principle of our
sensing device and characterize the waveguide-assisted
optical access. Then, we perform proof-of-principle mag-
netic field and temperature sensing with ensembles of
N-V
−
centers. We compare the sensor performance with
respect to the state of the art.
II. WORKING PRINCIPLE
Type-II waveguides are written in a chemical vapor
deposition (CVD) grown electronic-grade diamond slab
(
2mm× 2mm×0.3 mm
)
with pulsed laser illumination
at 515 nm, with a repetition rate of 500 kHz, pulse width
of 300 fs and laser power of 100 mW. The laser beam is
focused into the sample through a high-numerical-aperture
(high-NA) objective (1.25 NA, 100×) to create type-II
waveguides of 2 mm length, according to the size of
the diamond, in depths ranging from 5 to 25 μm below
the top diamond surface. The femtosecond laser writing
damages the diamond lattice structure within the mod-
ification lines serving as waveguide walls. This causes
a reduced refractive index at the waveguide walls and
an increased refractive index between two walls due to
stress in the diamond lattice. Hence, the waveguide mode
is confined to the center between two waveguide walls
for light guiding [30]. The waveguide depths are mea-
sured from the surface to the center of the modification.
The type-II waveguide width of 15 μm [center-to-center
054059-2
INTEGRATED MAGNETOMETRY PLATFORM WITH... PHYS. REV. APPLIED 15, 054059 (2021)
transverse spacing between the two laser modification
tracks as shown in Fig. 1(a) front view] is optimized for
single-mode, low-loss light transmission between 630 nm
and 740 nm according to the N-V
−
-center sideband emis-
sion. In order to functionalize the waveguides with N-V
−
centers, we shallow implant Nitrogen ions into the front
facet of the waveguides followed by subsequent annealing
at 1000
◦
C. Further details on the fabrication can be found
in Fig. 1(b) and in the methods section.
The spin state of the N-V
−
center, a spin-1 system
in its ground and excited states, can be detected opti-
cally via magnetic resonance spectroscopy (ODMR) [31],
as sketched in Fig. 1(c). Shelving from m
s
=±1 states
induces optical pumping and allows to read out the state via
measurement of the fluorescence intensity. The microwave
(MW) resonance frequency decreases from 2.88 GHz at
cryogenic temperatures to 2.87 GHz at room temperature
thereby enabling, in principle, temperature sensing over a
wide temperature range although the sensitivity decreases
at low temperatures. In a magnetic field, the degeneracy of
the m
s
=±1 states lifts due to a magnetic-field-dependent
splitting of = 2γ B
z
, with the gyromagnetic ratio γ =
28 GHz T
−1
, of the two states. As a result, the ODMR
resonance dip splits into two, one for each spin state,
thus enabling inference of the magnetic field strength.
Figure 1(d) illustrates the magnetic field dependent split-
ting. Here, the degeneracy is already lifted at zero-field
owing to residual strain in the diamond lattice [32,33]. For
continuous-wave ODMR (CW ODMR) measurements, the
overall sensitivity to dc magnetic fields [34],
η
dc
=
g
e
μ
B
√
N τ
,(1)
depends on the number of N-V
−
centers N with coherence
time τ. Thereby, g
e
= 2.0 denotes the electron g factor, μ
B
the Bohr magneton and the Planck constant. One way to
improve the overall sensitivity is to increase N . This can
either be done by increasing the density of N-V
−
centers or
by enlarging the sensing volume. However, progressing to
large sensing volumes remains challenging. The efficient
extraction of photons out of the diamond host from a large
ensemble of N-V
−
centers distributed over a large sensing
area is an outstanding problem. We therefore begin with
the benchmarking of the device detection efficiency with
respect to conventional, well-established confocal ODMR
measurements. We then extrapolate the overall sensitiv-
ity of the device by taking into account the large NV
−
ensemble that is addressable via the waveguide mode.
III. WAVEGUIDE-ASSISTED OPTICAL ACCESS
We first compare the detected signal through the waveg-
uide with the confocal detection. Therefore, we excite a
confocal spot of the N-V
−
ensemble and simultaneously
measure the transmission signal through the waveguide
and the confocal signal in reflection. We examine the trans-
mission properties for waveguides at a depth of 20 μm
below the diamond surface with well-isolated waveg-
uide modes. A confocal scan of the waveguide front
facet, depicted in Fig. 2(a), shows the bright areas where
N-V
−
centers are created. The areas around waveguide
walls appear brighter indicating an increased number of
N-V
−
centers. In this study, we focus on the leftmost
waveguide (green box). To characterize the transmission
of the waveguide, we map the mode excitation shown in
Fig. 2(b) by laterally scanning a 738 nm laser while record-
ing the maximum transmission signal. The 2D-Gaussian
mode profile is projected onto the horizontal and ver-
tical coordinate axes. We obtain a similar mode profile
when exciting the N-V
−
centers with 532 nm laser light
and collecting the fluorescence in transmission through
the waveguide while laterally scanning the excitation laser
[see Fig. 2(c)]. We note that the transmitted fluorescence
resembles the mode profile, although the front facet is
homogeneously covered with N-V
−
centers [see the con-
focal scan in Fig. 2(a)]. We conclude that N-V
−
emission
is guided through the waveguide.
In order to benchmark the detection efficiency of the
waveguide-assisted optical access, we first compare the
photoluminescence (PL) spectrum from the N-V
−
ensem-
ble in confocal configuration (reflection) with the spec-
trum acquired in transmission through the waveguide (see
Fig. 2(d) for experiment schematic) as shown in Fig. 2(e).
Note that in order to clearly determine the N-V
−
emis-
sion we cooled the sample to cryogenic temperatures.
The resulting PL spectra, both in transmission and reflec-
tion, reveal the characteristic zero-phonon line (ZPL) and
phonon-sideband (PSB) emission of N-V
−
centers. Here,
intensities are normalized with respect to the ZPL emission
for sake of clarity. The first thing to note is that the polar-
ization pattern of the waveguide mode, shown in the inset
of Fig. 2(b), yields a distinct polarization contrast which
matches the polarization of the waveguided N-V
−
fluores-
cence within ≈ 16.5
◦
. From the measured transmission and
reflection spectra we extract the ODMR-signal detection
efficiency of our waveguide-assisted sensor with respect to
conventional confocal measurements.
Figure 2(f) illustrates the ratio of transmitted signal
versus confocal detection which we refer to as relative
detection efficiency. The relative efficiency evidences the
challenge to extract photons from a large detection volume
with respect to a resolution-limited spot. The waveguide
transmission losses are disregarded in this comparison
and will be considered later together with the absolute
detection efficiency. The green curve marks the detection
efficiency for a N-V
−
subensemble that is not perfectly
centered compared with the waveguide mode (see Fig. 2(c)
green circle and dashed white cross for the excitation
point), resulting in a reduction of the intensity by the factor
054059-3
MICHAEL HOESE et al. PHYS. REV. APPLIED 15, 054059 (2021)
z
y
150 µm
y
z
10 µm
y
z
10 µm
640 660 680 700 720
Wavelength (nm)
0
10
20
30
Detection efficiency (%)
η = 0.61
η =1/e
2
0.2
0.4
0.6
0.8
1.0
1.2
Intensity (arb. units)
Reflection
Transmission
0°
7.0°
90°
180°
187.0°
270°
0°
90°
170.5°
180°
270°
-9.5°
0.25
0.50
0.75
1.00
Intensity (arb. units)
0.25
0.50
0.75
1.00
Intensity (arb. units)
(a)
(b)
(c)
(e)
(d)
(f)
FIG. 2. (a) Confocal scan showing the fluorescence spots of the shallow-implanted N-V
−
centers on the front facet of the sample. The
leftmost waveguide (marked with a green box) is studied in the following. (b) Mode shape of the waveguide (WG). The shape of the
WG mode is revealed by laterally scanning a 738 nm laser while recording the collected counts in transmission. The two intersections
at the side show the corresponding 2D Gaussian fit to the WG mode. The corresponding mode-polarization contrast is shown in the
lower right corner. Note that the image is rotated by 90
◦
with respect to panel (a). (c) N-V
−
transmission through the waveguide. We
excite the N-V
−
centers off-resonantly with 532 nm laser light from the front and collect the transmitted signal. Laterally scanning the
excitation laser resembles the waveguide mode in transmission. The intersection of the dashed lines denotes the position which is at
61% of the maximum of the WG mode (green ellipse). (d) Experiment schematic. The sketch illustrates the two detection channels,
reflection and transmission. (e) The PL spectra in reflection and transmission are compared. The exposure time of the transmission
measurement is ten times larger than for the measurement in reflection. Both spectra are normalized to the ZPL. The inset shows
the ZPL polarization measured in transmission through the waveguide, which resembles the waveguide mode polarization (marked
as a green line). (f) We extract the wavelength-specific optical detection efficiency of the N-V
−
ensemble to the waveguide mode by
comparing the transmitted PL spectrum with the fluorescence in reflection and by accounting for losses that occur until detection. The
green curve is the inferred efficiency for a N-V
−
ensemble located at 61% of the optimal coupling corresponding to the green ellipse
in (c). The black lines marks the 1/e
2
-profile of the Gaussian mode.
0.61 compared with the maximum at the optimal position.
We calculate relative detection efficiencies ranging from
8% at 660 nm to more than 15% at 735 nm. The gray curve
gives the efficiency for an ideally coupled N-V
−
center
located at the maximum of the waveguide mode, yielding
ideal relative detection efficiencies up to 30% at 735 nm.
The gray shaded area indicates all possible detection effi-
ciencies from ideally to completely uncoupled ensembles.
The black line marks the 1/e
2
profile of the Gaussian mode
that will be excited when operating the sensor through the
waveguide. The figure illustrates the large sensing area of
up to 105 μm
2
per waveguide mode with an average rela-
tive detection efficiency of 10.4% for the large ensemble of
N-V
−
centers which is addressed via the waveguide mode.
We now quote the absolute detection efficiency taking
into account the waveguide transmission losses including
the outcoupling efficiency of up to 79.8%, or 6.95 dB,
which is comparable to values reported for type-II waveg-
uide in diamond [20,35]. The absolute detection efficiency
of the confocal collection behind the objective is 2.5%
054059-4
INTEGRATED MAGNETOMETRY PLATFORM WITH... PHYS. REV. APPLIED 15, 054059 (2021)
corresponding to an absolute detection efficiency of 0.05%
through the waveguides when averaged over the 1/e
2
area
of the Gaussian mode. The waveguide detection efficiency
corresponds to a waveguide coupling efficiency of 0.3%
in agreement with estimated coupling efficiencies for indi-
vidual NV
−
centers in laser-written waveguides [24]of
0.1%. In future designs, shorter waveguides and improved
outcoupling performance will increase transmission, thus
leading to higher detection efficiencies.
Although the detection efficiency of a confocally excited
N-V
−
-center ensemble, as expected, decreases with respect
to conventional confocal detection, the large number
of addressable N-V
−
centers via the waveguide mode
increases the achievable overall sensitivity. We extrapolate
the increased number of N-V
−
centers that is accessible via
the waveguide mode by comparing the mode field area of
105 μm
2
with the confocal spot size of 0.062 μm
2
. Taking
into account the homogeneous distribution of N-V
−
centers
over the entire waveguide mode, we expect an increase in
the number of addressed N-V
−
centers N by a factor of
1690. Compared with conventional confocal detection, the
sensitivity thus improves by a factor of 41 according to the
√
N dependency.
IV. SENSOR PERFORMANCE
In the following, we study the sensing performance of
our device. All measurements are performed at room tem-
perature although the sensor can be operated over a wide
temperature range. We excite the N-V
−
ensemble with off-
resonant (532 nm) laser light, which is coupled from the
back side into the waveguide. The N-V
−
signal is also read
out through the waveguide in order to keep the front facet
of the diamond sensing area, which contains the shallow-
implanted N-V
−
ensemble, fully accessible. Figure 3(a)
shows a confocal scan of the diamond backside with the
low-NA objective used for incoupling of excitation light
and read-out of the N-V
−
signal through the waveguide.
The waveguide mode is clearly resolved including the
waveguide walls confining the mode, which appear dark.
The spectrum of the waveguide-assisted emission resem-
bles the spectrum of the N-V
−
ensemble, as highlighted
by comparison with a standard confocal spectrum of the
ensemble in Fig. 3(b). The background spectrum is shown
in red and is recorded spatially shifted from the waveguide
mode.
We now utilize the N-V
−
ensemble as a magnetic field
sensor by exploiting its ODMR signal. Therefore, we apply
a MW field to the sample and vary the external mag-
netic field by approaching the sensor with a permanent
magnet. Varying the MW frequency reveals the character-
istic dips in the fluorescence signal, as shown in Fig. 3(c).
First, we compare the ODMR signal at zero magnetic
field detected in standard confocal configuration with the
transmission through the waveguide. Both signals reveal
two characteristic dips separated by 9 MHz due to strain
induced by the laser-written waveguides [30], residual
strain in the diamond lattice [32,33] and also background
magnetic field from the environment. An applied mag-
netic field increases the splitting between the two dips as
shown in Fig. 3(c), both in transmission and reflection
[see Fig. 2(d) for a schematic of the experiment]. Figure
3(d) shows the increased splitting with increasing magnetic
field strength. At higher magnetic fields, each dip splits
into two, because the N-V
−
centers experience different
magnetic fields depending on their orientation in the crys-
tal lattice [36]. We use the gyromagnetic ratio of an N-V
−
center, γ = (g
e
μ
B
/) = 28 GHz T
−1
, in order to calculate
the splitting due to an applied magnetic field B along the
N-V
−
center axis as = 2γ B
z
(see [37]). From the spin
Hamiltonian of the N-V
−
center,
H = D
S
2
z
−
1
3
[
S
(
S + 1
)
]
+ E
S
2
x
− S
2
y
+ g
e
μ
B
B · S,
(2)
we deduce the magnetic field [5]as
(
g
e
μ
B
B
)
2
=
1
3
ν
2
1
+ ν
2
2
− ν
1
ν
2
− D
2
− E
2
.(3)
The zero-field splitting (ZFS) parameter is given as D and a
nonzero E incorporates additional splitting of the two dips,
located at MW frequencies ν
1
and ν
2
at zero magnetic field
due to strain. The ZFS could be further increased by the
Earth’s magnetic field. In our case, the ODMR measure-
ments yield effective magnetic fields between 0.066 and
1.508 mT along the N-V
−
center axis, respectively.
Our measurements showcase the ability to measure
magnetic fields with a precision better than 6 μT with our
device. The lowest panel of Fig. 3(d) illustrates the fitted
peak positions together with their error margins that are
now used to estimate the precision of magnetic field mea-
surements. The two measurements are performed at 0.066
and 0.214 mT applied magnetic field, respectively, with
two clearly separated peak positions. Their fit error mar-
gins are still significantly separated. We calculate the error
of the magnetic field difference of the two measurements
including their fit errors. With these fit errors as reference,
we estimate that magnetic field differences of 6 μT can be
resolved with our device. The sensitivity η
dc
of our mag-
netometer device to dc magnetic fields for CW ODMR
measurements can be calculated as [38]
η
dc
=
ν
γ C
√
I
PL
,(4)
where ν denotes the full-width-at-half-maximum
(FWHM) linewidth of the ODMR dip, C denotes the
ODMR contrast and I
PL
the detected PL intensity in
054059-5
