PRX QUANTUM 3, 030316 (2022)

Editors’ Suggestion

Single-Atom Trapping in a Metasurface-Lens Optical Tweezer

T.-W. Hsu,1,2 W. Zhu ,3 T. Thiele,1,2 M. O. Brown,1,2 S. B. Papp,4 A. Agrawal ,3 and
C. A. Regal 1,2,*
1
JILA, National Institute of Standards and Technology and University of Colorado, Boulder, Colorado 80309,
USA
2
Department of Physics, University of Colorado, Boulder, Colorado 80309, USA
3
National Institute of Standards and Technology, Gaithersburg, Maryland 20899, USA
4
National Institute of Standards and Technology, Boulder, Colorado 80305, USA

(Received 9 February 2022; revised 6 May 2022; accepted 23 June 2022; published 1 August 2022)

Optical metasurfaces of subwavelength pillars have provided new capabilities for the versatile definition
of the amplitude, phase, and polarization of light. In this work, we demonstrate that an efficient dielectric
metasurface lens can be used to trap and image single neutral atoms with a long working distance from
the lens of 3 mm. We characterize the high-numerical-aperture optical tweezers using the trapped atoms
and compare with numerical computations of the metasurface-lens performance. We predict that future
metasurfaces for atom trapping will be able to leverage multiple ongoing developments in metasurface
design and enable multifunctional control in complex quantum information experiments with neutral-atom
arrays.

DOI: 10.1103/PRXQuantum.3.030316

I. INTRODUCTION active components, from acousto-optic devices to spatial

light modulators, is important for moving and addressing
Arrays of single trapped neutral atoms are a burgeon-
individual atoms [13,14]. Static components that reduce
ing platform for quantum simulation, computing, and
the reliance on large conventional optics for trapping
metrology [1–3]. With ground-up control similar to that
and focusing will also reduce constraints in increasingly
of trapped ions, individual atoms can be prepared and
complex vacuum chambers and improve scalability. The
entangled [4–6] and increasingly hold promise for scal-
advancement of these specialized optical systems will
able quantum computing [7–9]. However, practical quan-
benefit from connection to the forefront of integrated
tum computing requires substantial advances in reducing
photonics development.
error rates and scaling the qubit number. One upcom-
Metasurfaces are planar photonic elements composed
ing outstanding challenge for neutral-atom arrays is the
of a periodic array of subwavelength dielectric or metal-
development of scalable and multifunctional optical com-
lic nanostructures that have made a significant impact
ponents that enable site-selection manipulation of hyper-
on photonic systems in recent years. Contrary to tradi-
fine states and Rydberg excitations, operate in constrained
tional optical elements that rely on refraction for phase
environments, and achieve low scattering and crosstalk.
shift, the nanostructures constituting a metasurface cou-
In ion-trap experiments, long-standing efforts in
ple resonantly or off resonantly and reradiate the incom-
developing integrated optical components have enabled
ing light with a transformed phase, polarization, and
improved parallelism and addressing capabilities [10–12].
amplitude determined by the nanostructure shape, size,
Neutral atoms will require a similar trajectory and have
and material composition [15]. Electromagnetic modeling,
many unique requirements. For example, control of sin-
device optimization, and fabrication of nanostructures with
gle neutral atoms relies heavily on optical potentials for
unprecedented complexity and resolution have enabled
trapping, either in lattices or arrays of tightly focused
multifunctional control of the optical wave front [15,16].
laser beams, termed optical tweezers. The development of
By spatially varying the constituent nanopillar width in a
pattern similar to a Fresnel lens, the reradiated light can
* converge at the far field to form a focal point, i.e., create a
regal@colorado.edu
metasurface lens (Fig. 1).
Published by the American Physical Society under the terms of
the Creative Commons Attribution 4.0 International license. Fur- An important performance metric for utilizing meta-
ther distribution of this work must maintain attribution to the surface optics is the transmission efficiency, which is
author(s) and the published article’s title, journal citation, and governed by multiple factors, including the choice of low-
DOI. loss and high-index dielectric thin films and the ability to

2691-3399/22/3(3)/030316(11) 030316-1 Published by the American Physical Society

T.-W. HSU et al. PRX QUANTUM 3, 030316 (2022)

(a) (b) (c) Vacuum

Metasurface
Chamber

NA = 0.55
Metalens

λ = 852 nm (d)
13.2 μm
Count

85–185 nm λ = 780 nm

Two-axis
- AOD

660 nm
Si
1 μm
Incident wave Focused wave front Single-atom-array image
CCD camera

FIG. 1. Metasurface optics for optical-tweezer trapping. (a) A scanning electron micrograph of the fabricated metasurface lens
containing a periodic array (lattice constant = 280 nm) of amorphous-Si (a-Si) nanopillars (height 660 nm) of width ranging from 85
nm to 185 nm (dark blue) on top of a 500-μm-thick fused-silica substrate (light blue). The inset shows the varying nanopillar width to
achieve the desired phase shift [see Fig. 4(a)]. (b) A notional illustration of the metasurface-lens operation, showing light propagation
(pink), wave fronts (dashed lines), and the secondary wavelets (black semicircles) reemitted by the nanopillars that interfere to create
the focusing wave front. (c) The optical setup for trapping (pink) and fluorescence imaging (green) of single atoms in an array created
with multiple input beams generated using a two-axis acousto-optic deflectors. (d) An image of a trapped 87 Rb array created by
averaging over multiple experiment iterations (100 in this case) with approximately 52% probability of a single atom in each trap per
image. The variation in the averaged intensity is caused by trap depth and shape variations that affect the relative loading probability
and imaging signal in the array.

pattern high-aspect-ratio nanostructures with high fidelity. explorations in metasurfaces for atomic experiments have
Recently, metasurface lenses have been reported with effi- been used to create beam-shaping and -splitting elements
ciencies ranging from 60% to 92% [17–19], utilizing a in magneto-optical traps [25,26]. In this work, we open
range of materials such as TiO2 , HfO2 , Si, and GaN, and up the use of metasurfaces to optical dipole traps, in the
operating from the deep ultraviolet to the near infrared. form of tightly focused optical tweezers, and hence to
Further, the use of inverse design, machine learning, and impact increasingly complex quantum information exper-
multilayers can improve the performance and expand the iments with neutral atoms. We use a high-NA dielectric
versatility of metasurface optics [20,21]. metasurface lens to trap and image single atoms (Fig. 1)
It is informative to compare the spatial wave-front and obtain tight trap confinement. We form an atom array
control of metasurfaces with, for example, commercial by combining the metasurface lens with tunable acousto-
spatial-light-modulator (SLM) technologies based on liq- optic deflectors and characterize the tweezer foci using
uid crystals (LCs) or digital-micromirror devices (DMDs). the trapped atoms. Our successful trapping is an indication
LC SLMs and DMDs have been used in combination with that potential deleterious effects of metasurface optics—for
conventional high-numerical-aperture (high-NA) lenses in example, scattered light, the presence of undeflected zero-
atom-array experiments to create arbitrary reconfigurable order light, or deformations due to absorption and heating
potentials through spatial modification of the optical wave of the lens—make negligible contributions to the trap-
front using pixels larger than the optical wavelength. Meta- ping performance of large-spacing tweezers. We predict
surfaces, in contrast, consist of deep-subwavelength pillars that future optimized photonic metasurfaces that leverage
and operate based upon a waveguide effect that provides ongoing advances in element-design libraries and multi-
large bend angles that can be used for high-NA optics layer design will enable advanced future high-NA designs
and aggressive wave-front shaping. To tailor the wave with multifunctional performance.
front, the pillars have a controlled size, spacing, and shape,
which further enables capabilities such as polarization [22]
and wavelength multiplexing [23]. While the metasurface II. REQUIREMENTS OF HIGH-NA OPTICAL
used for atom trapping in this work is a static metasur- TWEEZERS
face, active wave-front shaping using metasurfaces is an
area of active research [24] and has the potential to yield In optical tweezers, high-NA optics are key for the
a novel class of SLMs offering capabilities complementary creating of trapping potentials, the optical addressing of
to counterparts based on LCs or DMDs. individual atoms in quantum gate protocols, and imaging
In atomic physics, metasurface optics are intriguing to the fluorescence of single atoms [27,28]. Often, multiele-
explore, given their combination of high performance, ment objective lenses are required to achieve the requi-
multifunctional response, and low form factor. Initial site performance [27–29], although single aspheric lenses

030316-2
SINGLE-ATOM TRAPPING... PRX QUANTUM 3, 030316 (2022)

have also been instrumental in state-of-the-art experiments The design wavelength of the lens is the trapping wave-
studying interacting Rydberg atoms [30]. length of λ = 852 nm, which is sufficiently far off res-
Optical-tweezer experiments require both low aberra- onance for the 87 Rb atoms to avoid recoil heating. The
tions to achieve tight confinement and a high focusing 4 mm × 4 mm square lens is illuminated with a circular
efficiency to achieve a sufficient trap depth for a given Gaussian beam with a 1/e2 radius of 2 mm. The lens is
trapping power and to efficiently image single atoms. also used to collect fluorescence on the 87 Rb D2 line at
Achromatic properties are needed for simultaneously col- 780 nm. Given the singlet properties of the lens and the
lecting atom fluorescence, conservative trapping in a far design space offered by the square-shaped nanopillars used
off-resonance trap, and often also the delivery of exci- in this work, it is not optimized to be diffraction limited
tation light that controls the atomic state in individual at 780 nm. The metalens is comprised of a thin film of
tweezers [4]. Broadband operation is especially important amorphous-Si (a-Si) (refractive index n = 3.62 + i0.004 at
for multispecies or molecular optical-tweezer experiments λ = 852 nm) deposited and patterned on a fused-silica sub-
[31–33]. Further, arbitrary and clean polarization control is strate (n = 1.45) [Fig. 1(a)] (for fabrication details, see the
increasingly desired. Appendix). The materials used for the metalens are fully
A long working distance (WD) is required to allow ultrahigh-vacuum (UHV) compatible and can resist UHV
access for laser cooling beams, to maintain a sufficient dis- baking temperatures without any change of properties. The
tance between the lens substrate and atoms in high-energy lens is mounted on a sample holder inside an antireflection
Rydberg states that are sensitive to surface electric dipoles, (AR)-coated glass cell.
and to focus light into a complex vacuum chamber or cryo-
genic environments [34]. In addition, stability of the optics
IV. METASURFACE MODELING AND
is crucial, for example, in registration of optical tweezers
CHARACTERIZATION
and lattices or for in-vacuum applications. Further, pertur-
bations to the trap focus due to multibeam interference or We carry out a full numerical simulation of the expected
scattered light need to be minimized, especially if they are metalens properties using the finite-difference time-domain
not static, as these fluctuations can drive atom motion. (FDTD) method (see the Appendix). The theoretical focus-
ing efficiency, estimated as the fraction of the incident light
that is diffracted toward the focal spot, is 78%. The loss is
expected to be derived from a combination of reflection
(14%), light remaining in the zeroth order (6%), and light
III. METASURFACE-LENS OVERVIEW
scattered into higher orders (2%).
For the demonstration presented in this work, we use To optically characterize the fabricated metalens, we
a high-contrast transmission-mode metasurface lens (met- perform a number of experimental tests of the device used
alens) with a NA of 0.55, a measured focusing efficiency in the atom-trapping experiment. First, to characterize the
of 58% at the design wavelength of 852 nm (56% for the focal spot, we image a pair of 300-nm-diameter pinholes
atom-imaging wavelength at 780 nm), and a focal length separated by 6.86 μm using the metalens. We find that the
of 3 mm (equivalently, a WD of 3 mm for the thin lens) lens is diffraction limited at 852 nm [Fig. 2(f)] by measur-
(Fig. 1). Using the trapped atoms, we measure the Gaus- ing the imaged point-spread function (PSF) and fitting it to
sian 1/e2 radius (waist) of the focused tweezer spot to a Gaussian to find a waist of 0.72 μm. At the focus for 780
be w0 = (0.80 ± 0.04) μm, which is consistent with NA nm, we find a Gaussian waist of 1.1 μm. Further, the metal-
of the designed lens. Further, we create an array of traps ens images the 780-nm atom fluorescence out of focus and
with our focusing metasurface lens by introducing multiple we use the pinholes to also analyze and predict the diver-
beams with tunable angles derived from an acousto-optic gence of the imaging system. Specifically, we find 780 nm
deflector pair and we demonstrate a field of view (FOV) of with a chromatic focal shift of +300 μm compared to 852
±11 μm (±0.2◦ ) [Fig. 1(d)], which is consistent with a full nm. We also find, as expected for this in-plane square-pillar
theoretical model of the metalens. The FOV is defined as design, that there is negligible polarization dependence in
the distance (angle) at which the size is 10% larger (Strehl the focal-spot positions.
ratio > 0.8). We are able to observe the atoms by measur- Second, we characterize the efficiency relevant to both
ing through the same metasurface lens, which is a stringent trapping the light throughput at 852 nm and the collec-
test of the efficiency of the system. Recently, metasurface- tion efficiency at 780 nm. We assess the combined loss
lens trapping and detection of dielectric nanoparticles has from all factors by measuring the optical-tweezer power
been demonstrated but note that these experiments have transmitted through a 300-μm-diameter spatial filter and
not required the efficiencies demonstrated in our work, we also measure the zeroth-order transmission contribu-
because the dielectric particles have been detected using tion directly (see the Appendix). The measured focusing
scattered trap light with a much larger magnitude than efficiency, defined as the ratio of power that passes through
atomic fluorescence signals [35]. the 300-μm spatial filter placed at the lens focus to the total

030316-3
T.-W. HSU et al. PRX QUANTUM 3, 030316 (2022)

(a) the increase of zeroth-order light in comparison to the-

No atom ory are likely due to fabrication imperfections resulting in
(b) Count
Loading threshold nonideal nanopillar cross sections and side-wall tapering.
The amount of zeroth-order undiffracted light can
Occurrence

One atom potentially be an issue if it is large enough to alter the

trapping potential by interfering with the focused light.
However, from the efficiency measurement of our tightly
Gaussian fit focused optical tweezers, the intensity at the focused
Two atoms
(no occurrence)
tweezer spot is more than 8 orders of magnitude larger
than the zeroth-order intensity at the trap center. Hence,
CCD camera count the amplitude ratio, which is important for interference
(c) (d) effects, is calculated to be 2 × 104 times smaller. In the
future, the zeroth-order light contribution can be reduced
by approaching the theoretical number through better fab-
Survival

Survival

rication or intentionally diverging or deflecting the zeroth-

order light in the design.

Δ (MHz) Modulaon frequency (kHz) V. METASURFACE-LENS ATOM TRAPPING

(e) (f)
Atoms are captured in the optical tweezers by overlap-
Normalized intensity

ping the focus of the metalens with a magneto-optical trap

vtrap (kHz)

(MOT) and applying polarization-gradient cooling (PGC)

for 15 ms while the optical-tweezer traps are on [36].
Light-assisted collisions are used to ensure that only one
atom remains in the trap [27]. To image the atoms, we
use a 1-mm-diameter probe beam that avoids scattering off
U/kB (mK) x (μm)
of the metasurface by propagating parallel to the substrate
[see Fig. 5(e)]. This beam, in PGC configuration, illumi-
FIG. 2. Single-atom trapping in a metalens optical tweezer. (a)
nates the atoms for 25 ms, the fluorescence is collected
A camera-count histogram indicating the presence of either zero
or one atoms in a tweezer trap. A threshold is chosen to determine by the metalens, and the slightly diverging fluorescence is
if an atom is trapped and to calculate the loading efficiency. (b) reflected by a dichroic mirror, passed through an imaging-
A typical single-shot fluorescence image of a single atom imaged lens system, and focused onto a charge-coupled-device
through the metalens with PGC imaging. (c) A typical light-shift (CCD) camera [see Figs. 1(c) and 5]. Figure 1(d) shows
measurement with a Gaussian fit (red line) to the shifted atomic an example single-atom array averaged over 100 loading
resonance. The dashed line corresponds to the free-space 87 Rb iterations.
D2 F = 2 to F  = 3 transition. (d) A typical parametric heating We first analyze in detail one trap that is at the center
measurement with a Gaussian fit (red line) to extract the trap fre- of the metalens FOV. We plot a histogram of the fluores-
quency (νtrap ) from the modulation resonance (2νtrap ). Each point
cence counts collected through the metalens and registered
is an average of 100 trap-loading sequences. (e) The measured
trap frequency versus the trap depth (light shift) obtained from on the camera (CCD counts) versus the occurrence from a
multiple measurements similar to (c) and (d). The solid red line single pixel at the atom location [Figs. 2(a) and 2(b)]. The
is a model fit (see main text) to extract the effective Gaussian lower-count histogram peak corresponds to background
tweezer waist seen by the trapped atom. (f) The peak-normalized signal with no atom and the second higher CCD-count
cross section of the intensity transmission through a 300-nm- peak corresponds to the fluorescence of the single atom.
diameter pinhole imaged at the 852-nm trapping wavelength by Collisional blockade prevents the loading of more than
the metalens. The solid blue line is an Airy function fitted to the one atom into the tweezer, as reflected in the absence of
data to extract the spot size and effective NA. [The error bars in a two-atom peak [27]. We find a loading probability of
(c) and (d) represent the standard deviation and the error bars in
(47 ± 5)%. However, due to the limited imaging-beam
(e) are the standard error of the fitted Gaussian centers.]
geometry [see Fig. 5(e)], the atom loss during imaging
is (10 ± 2)%. Taking this into account, a loading proba-
power incident on the substrate, is determined to be 58% bility of (52 ± 5)% is comparable to the typical loading
at 852 nm and 56% for 780 nm, somewhat smaller than efficiency from other optical-tweezer experiments [37].
the theoretical value. We find zeroth-order light transmit- We determine the length of time for which a single atom
ted through the lens (which is conveniently used for system remains trapped in the optical-tweezer focus, with no cool-
alignment) to be 13%, somewhat larger than the theoreti- ing light present, by holding the atom with variable time
cal estimation. The reduction of the overall efficiency and between two consecutive images. The measurement gives

030316-4
SINGLE-ATOM TRAPPING... PRX QUANTUM 3, 030316 (2022)

a lower bound of the exponential decay lifetime of 10 s; heats the atom out of the trap. For trap-frequency mea-
atom-lifetime assessment in a metalens trap beyond this surements, we modulate the trap between 5% and 10% of
scale will require additional investigation of background its depth around the nominal value to parametrically heat
gas collision rates due to finite vacuum level and potential at twice the trap frequency (νtrap ) and subsequently lower
atom-loss contributions due to inelastic light scattering the trap depth to eject hot atoms. Figures 2(c) and 2(d)
from residual trapping light. show the typical light shift and trap-frequency measure-
Next, we characterize the effective tweezer focus size ments. The trap waist can be deduced from the slope of
by measuring both the trap depth and the trap frequency a graph that plots the√trap frequency versus the depth as
(the harmonic oscillator strength of the atom moving in per νtrap (U, w0 ) = 2π
1
4U/w0 mRb [Fig. 2(e)]. We extract
the optical-tweezer light). The measurements are made a 1/e2 Gaussian radius of w0 = (0.80 ± 0.04) μm at 852
by determining atom survival following perturbations that nm, which is consistent with the value determined from the
depend upon a parameter of interest. For measuring the optical-lens characterization [Fig. 2(f)]. With the clipped
trap depth U, we make use of the fact that the trap light Gaussian-beam illumination used for the optical-tweezer
induces an ac Stark effect that shifts the atomic reso- trapping (versus the uniform illumination during character-
nance by 28.8 MHz/mK compared to that in free space ization), we expect the tweezer to have a waist of 0.78 μm,
and we determine the frequency at which resonant light consistent with the measured value.
An important metric for creating and imaging large atom
arrays is the lens FOV. Figure 3 illustrates a study of the
(a) (b) metalens tweezer off axis. For this, we create four traps
Count with the lower-left tweezer at the center of the FOV (the
optical axis) and characterize the traps (with various spac-
Survival

ing) in analogy to Figs. 2(c) and 2(d). In the presence

13.6 μm

of aberrations, the traps become asymmetric, resulting in

μ

nondegenerate frequencies in the radial dimensions of the

Opcal axis trap. This will manifest as a double-peak structure in the
Modulaon frequency (kHz) trap-frequency measurement [Fig. 3(b)]. We characterize
(c) the FOV by plotting the waist determined from the trap
frequency and depth measurements as a function of the
distance from the optical axis [Fig. 3(c)] and find that
the aberrations are consistent with FDTD calculations of
w0 (μm)

tweezer intensity from our metalens optical-field distribu-

tion [blue lines, Fig. 3(c)]. Here, the FOV is defined as the
distance to a point at which the average waist is 10% larger
(Strehl ratio > 0.8) than at the center and we find a FOV
of ±11 μm (±0.2◦ ).
r (μm)

FIG. 3. The atom array and the metalens FOV. (a) A PGC
fluorescence image of an atom array trapped with metasurface VI. COMPARATIVE AND FUTURE POTENTIAL
optical tweezers. The image is averaged over 100 experimen- As one comparison, we discuss the performance of a
tal cycles. The bottom-left tweezer is on the optical axis of the
typical commercial asphere that has been used in optical-
metalens. The off-axis tweezer sites typically have a lower load-
ing probability and nonoptimal PGC imaging detuning, resulting tweezer experiments. Reference [30] uses an aspheric lens
in a dimmer single-atom signal. (b) An example of typical trap- with NA = 0.5, a working distance of 5.7 mm, a focal
frequency measurement data at approximately 13.6 μm from the shift of −40 μm from 852 nm to 780 nm, and a focal
FOV center where asymmetric aberrations in the trap are present, length of 8 mm. This aspheric lens has a transverse FOV
along with a double Gaussian (green line) fit. (c) The extracted of ±25 μm (±0.18◦ ), an inferred beam waist of 1 μm for
Gaussian waist as determined from the atom trapping as a func- the trapping wavelength, and a 0.9 μm waist for the imag-
tion of the distance (r) to the metalens optical axis (the center of ing wavelength. The metasurface studied here has a worse
the FOV). The average waist is extracted from a single Gaussian focal shift than a standard asphere but, as discussed below,
fit to the trap-frequency data (red) and the major waist (green
this is not of primary concern in our experiments given
circle) and minor waist (green diamond) are extracted from data
similar to (b) when the two trap frequencies are distinguishable. the prospects for future control, for example, using wave-
We compare with the theory of the major and minor Gaussian length polarization multiplexing. The singlet metasurface
waists fitted from FDTD simulation [see Fig. 4(c)]. [The error here achieves similar or better performance as the repre-
bars in (b) represent the standard deviation and the error bars in sentative asphere for the focal-length-to-FOV ratio, i.e.,
(c) are the standard error.] the angle, and the effective NA.

030316-5
T.-W. HSU et al. PRX QUANTUM 3, 030316 (2022)

In comparison, a complex high-NA objective lenses APPENDIX

used for atom trapping and imaging can have a FOV of
1. Metasurface phase profile
a few 100 μm (approximately ±3◦ ) combined with achro-
matic operation over a wide range of wavelengths [3,28]. The metalens used in this study consists of a square
While the singlet metalens described in this work does not lattice (lattice constant a = 280 nm) of a-Si nanopillars
yet achieve these metrics, we now discuss the horizon of on a fused-silica substrate. Each nanopillar unit cell, of in-
prospects for design features of future metasurfaces. plane square cross section (side length L) and height H =
As discussed previously, with a metasurface it is pos- 660 nm, acts as a phase-delay waveguide [Fig. 1(a)]. The
sible to achieve a focusing response that is either polar- in-plane side lengths L of the nanopillars vary between 85
ization selective [38] or that transforms the polarization nm and 185 nm, forming a library of metasurface unit-cell
[39], which are functions not offered by traditional optical designs offering transmittance close to unity and relative
lenses. For example, polarization multiplexing provides a phase shift φ covering the full 0–2π span [Fig. 4(a)]. This
method to trap and collect fluorescence at the diffraction design library is then used to implement the phase profile
limit for two different wavelengths using a singlet metasur- of an ideal lens [45], given by
face lens and may find utility in combining additional mul-
tifunctional beams in complex trapping experiments. To 2π   
φ(x, y) = f − x2 + y 2 + f 2 , (A1)
illustrate this prospect, we design and optically test a sam- λ
ple with in-plane rectangular shape pillars that achieves
equal focal lengths for 780-nm and 852-nm light of orthog- where λ is the design wavelength (here, λ = 852 nm), f is
onal polarization (see the Appendix). This concept can be the focal length (here, f = 3 mm), and x, y are the coor-
used to trap at 852 nm and collect fluorescence at 780 nm, dinates of the sampled phase position relative to the lens
with a 50% efficiency due to the random polarization of the center.
scattered light from atoms.
More functionality can be achieved by expanding the 2. Metasurface theoretical estimation of efficiencies
number of surfaces offered in the design. To focus on the
FOV as one metric, an enhanced FOV up to ±25◦ has been We use the grating averaging technique [46] to esti-
achieved by patterning both sides of the substrate to cre- mate the focusing efficiency, the fraction of incident light
ate a double-layer metasurface [40]. We estimate that by remaining as undeflected zeroth-order light, and the total
using design components similar to the singlet lens pre- reflectance of the millimeter-scale diameter metalens. Fol-
sented here, expanding to a doublet can improve the field lowing this technique, we approximate the metalens as
angle to beyond ±5◦ at 0.55 NA. a combination of aperiodic beam deflectors. For an ape-
Further design improvements can be achieved through riodic beam deflector with a deflection angle θD (θD =
the use of an expanded unit-cell library to include cross, sin−1 (λ/Na), where N is chosen to calculate θD between 0◦
donut, and other exotic nanopillar shapes [41] or via and 50◦ ), the deflection efficiency, η1 , the fraction of light
inverse design [42]. The choice of optimal materials in the zeroth order, η0 , and the reflectance, ηrefl , for unpo-
and designs that are robust to nanofabrication variation larized input light, are calculated [circles in Fig. 4(b)] using
is expected to offer higher efficiencies that exceed that rigorous coupled-wave analysis (RCWA) and fitted with
achieved in the experiments presented here [16]. Further, parabolic functions [solid lines in Fig. 4(b)]. Finally, the
a hybrid lens design consisting of a curved refractive focusing efficiency of the metalens, T1 , the total fraction in
surface and a metasurface patterned on the same sub- the undeflected zeroth order, T0 , and the total reflectance,
strate will offer additional prospects for enhanced design Trefl , are estimated as the area average of η1 , η0 , and ηrefl ,
space [43,44]. respectively, using

ACKNOWLEDGMENTS   R

This work was supported by the Defense Advanced Ti = 1/(π R )

2
ηi ds = 2/R 2
ηi (r)r dr, (A2)
S 0
Research Projects Agency (DARPA) A-PhI program
under Grant No. FA9453-19-C-0029, the Office of Naval where i = 1, 0, or refl; r = f tanθD ; and R is the radius of
Research (ONR) under Grants No. N00014-17-1-2245 and the metalens.
No. N00014-21-1-2594, and the National Science Foun-
dation (NSF) Quantum Leap Challenge Institute (QLCI)
under Office of Multidisciplinary Activities (OMA) Grant 3. Metasurface theoretical estimation of FOV
No. 2016244, and NSF Grant No. PHYS 1914534. We The beam waist at the focal spot as a function of the
acknowledge helpful input from Christopher Kiehl and distance from the metalens optical axis or, equivalently, the
Zhenpu Zhang and technical expertise from Yolanda incident angle of the input beam, is calculated using the
Duerst. FDTD technique, with a minimum mesh size of 4 nm. Due

030316-6
SINGLE-ATOM TRAPPING... PRX QUANTUM 3, 030316 (2022)

(a) (b) (c)

0°

First order

Zeroth order
0.2° 0.3° 0.4°

D(°)

FIG. 4. The metalens design and simulations. (a) The transmittance t and phase shift φ as a function of the nanopillar side length L.
(b) The calculated deflection efficiency η1 , the fraction of light in the undeflected zeroth-order η0 , and the reflectance ηrefl of aperiodic
metasurface beam deflectors as a function of the deflection angle θD . The circles are data from RCWA simulations and the solid lines
are parabolic fits. (c) The FDTD simulated beam profiles of the focal spots as a function of the angle of the incident light.

to the millimeter-scale size of the metalens, a cylindrical 5. Metasurface optical characterization

metalens is simulated instead, implemented by using one To verify that the lens is diffraction limited at 852 nm,
unit cell along the y axis with periodic boundary condi- we image a pair of pinholes spaced by 6.86 μm and 300 nm
tions. All the unit cells along the x axis are included in the in diameter with the metalens. The pinholes are sufficiently
simulation and the oblique incidence angle is only applied small to be treated as point sources. The magnification of
along the x direction. For a given incident angle, a near- the system is calibrated by using the known separation of
field monitor records the electric and magnetic fields of the the pinholes. Fitting an Airy function to the imaged PSF,
output beam at a distance of 50 nm from the exit surface of a 1/e2 Gaussian waist of (0.72 ± 0.02) μm and an effec-
the metasurface. A near-field to far-field projection is then tive NA of 0.55 ± 0.01 are extracted [Fig. 2(f)], which is
used to calculate the focal-spot intensity profile at the focal consistent with the diffraction limit.
plane [Fig. 4(c)]. The major and minor waists of the focal To measure the focusing efficiency, a spatial filter is used
spot are obtained as the distance from the intensity peak to to exclude the zeroth-order transmission from the focused
1/e2 of the peak intensity along the x axis. light. A collimated 852-nm laser beam of 4 mm in diame-
ter illuminates the metalens. A pinhole of dimensions that
allow the focused light to be transmitted (300 μm pinhole
4. Metasurface fabrication details of 300 μm thickness) is then placed at the metalens focus.
The metasurface optics are fabricated by depositing a A power meter is placed 7 mm away from the metalens
layer of 660-nm-thick a-Si on a 500-μm-thick fused-silica (4 mm from the metalens focus) and the pinhole is trans-
wafer using plasma-enhanced chemical vapor deposition lated in x, y, and z to maximize the power transmitted.
(PECVD). A 300-nm-thick layer of electron-beam resist The input power and the transmitted power are compared
(ZEP 520A) followed by a 20-nm-thick layer of anticharg- to extract the focusing efficiency. The procedure is then
ing conductive polymer (ESpacer 300Z) are spin coated repeated for 780 nm and for other input polarizations. The
onto the a-Si film. A 100-keV electron-beam lithography focusing efficiency is found to be 58% at 852 nm and 56%
system is used to expose the nanopillar pattern, followed for 780 nm and insensitive to polarization rotation for both
by ESpacer removal with deionized water at room temper- wavelengths.
ature and resist development with hexyl acetate at 4 ◦ C.
The developed pattern in the resist layer is transferred 6. Sample mounting and vacuum chamber
to an electron-beam-evaporated 70-nm-thick Al2 O3 layer The metasurface sample is mounted in a Pyrex cell (sci-
using the lift-off technique. By using the patterned Al2 O3 ence cell) with an antireflection coating on the outside
layer as an etch mask, inductively coupled plasma reactive- [Fig. 5(a)]. A sample holder machined from a fused-
ion etching (ICP RIE: gas mixture, SF6 and C4 F8 ; ICP silica wedge (0.5◦ ) with faces polished to better than
power, 1750 W; radio-frequency (rf) power, 15 W) is per- λ/8 is epoxied to the inside of the cell with ultralow-
formed to etch the underlying a-Si layer at 15 ◦ C, to create outgassing high-temperature epoxy (Epotek-353ND). The
high-aspect-ratio a-Si nanopillars. The metasurface-optics epoxy absorbs any minor thermal expansion mismatch
fabrication is finalized by soaking the wafer in a mixture between the Pyrex and the fused-silica substrated. The
of hydroperoxide and ammonia hydroxide solutions (80 ◦ C metalens sample [Fig. 5(b)] is then optically contacted to
for 30 min) to remove the Al2 O3 etch mask and any etch the sample holder [Fig. 5(a)]. The optical-contact bond-
residue. ing ensures that the metalens substrate remains optically

030316-7
T.-W. HSU et al. PRX QUANTUM 3, 030316 (2022)

(a) (b) (c)

y Vacuum cell Zeroth order
z (top view) transmission
Alignment Forward beam Dichroic mirror x
mirror Alignment
y camera 2
NA = 0.55
x
z 4 mm metalens
Substrate Substrate
back reflecon back reflecon
(d) 111 mm Infinite conjugate
Airy-disk
Iris
150 mm
focus adjustment Prealigned Alignment (e)
Obj: on axis PBS Vacuum cell
camera 1
Probe (end view)
Tweezer beam y
L1 L2 CCD camera beam
Metalens CCD camera x
f = 150 mm f = 150 mm (atom imaging) z
10 μm off axis 3 mm

FIG. 5. The metalens in-vacuum mounting and tweezer alignment. (a) A photograph of the metalens sample optically contacted onto
the wedged fused-silica sample holder that is epoxied onto the AR-coated glass cell. (b) The fabricated metalens sample with a NA of
0.55 designed for 852-nm tweezer light. (c) A schematic illustrating how the tweezer light and the science CCD camera are aligned
to the metalens sample via substrate back reflection. (d) A schematic illustration of the optical-tweezer imaging path with lenses
that compensate for out-of-focus imaging due to the chromatic focal shift introduced by the metalens. The insets are the ray-tracing
simulation of the imaging system (object on axis and 10 μm off axis) assuming that the metalens only has a chromatic focal shift and
no other aberration. The result shows that L1 and L2 do not introduce additional aberrations. The black circle is the diffraction-limited
Airy disk. (e) The end view of the vacuum cell, showing the probe beam (also the resonant heating beam) orientation in relation to
the metalens sample and the tweezer beam. The probe beam is 1 mm in diameter, shines vertically up, and is 3 mm away from the
metalens, overlapping with the optical-tweezer focus.

flat after an UHV bake (up to 220 ◦ C). The adhesive-free 8. Metalens and CCD camera alignment
optical contact also allows the cell to be reused indefinitely. To ensure optimal tweezer performance from the high-
The materials used for the metalens (a-Si and fused-silica) NA metalens, the input light is aligned perpendicular to
are UHV compatible and can be baked to high temperature and centered onto the metalens [Fig. 5(c)]. The back reflec-
(> 200 ◦ C). tion of the substrate is used to align the tweezer input light.
The atomic source is a MOT glass cell that is located 40 The tweezer light (1/e2 waist of 2 mm) is passed through
mm from the science cell and connected through a differ- a polarizing beam splitter (PBS) and an iris apertures the
ential pumping orifice with a vacuum conductance of 0.05 beam down to 0.5 mm diameter. Alignment camera 1
L/s. The science cell connects to an ion pump with a pump- [Fig. 5(c)] is placed on the reflection port of the PBS to
ing speed of 25 L/s, resulting in a vacuum environment of monitor the back reflection from the metalens substrate.
< 10−10 hPa measured at the ion pump. A valve between This iris allows < 0.25 mrad angular-alignment resolution
the source MOT cell and the rest of the system isolates between the input beam and the substrate. Alignment cam-
the source MOT while the system is vented for sample era 2 [Fig. 5(c)] is placed behind the glass cell to monitor
exchange. The compact construction of the vacuum cham- the zeroth-order metalens transmission. The shadow cast
ber allows the chamber to be moved elsewhere for sample by the structure on the metalens allows the input beam to
mounting and UHV baking. be centered on the metalens. The input beam is walked
while monitoring the image on both alignment cameras
until the input is both perpendicular and centered. The
residual reflection of the back-reflected tweezer beam from
7. Acousto-optic deflectors the dichroic mirror [Fig. 5(c) light pink] is used to align
To generate an array of optical tweezers, a 1/e2 waist the position of the science camera and the imaging system.
2 mm collimated beam at 852 nm [pink shaded beams Finally, a band-pass filter centered at 780 nm (Semrock
in Fig. 1(c)] is launched into a two-axis acousto-optic LL01-780-12.5) is placed in the imaging path to block any
deflector (AOD) (AA Opto-electronics Model: DTSXY- residual tweezer light.
400-850.930-002). This produces a series of deflected
beams with variable angle controlled by the AOD-input
rf frequencies. This array of angled collimated light is 9. Imaging path compensation
then imaged with a 1:1 relay lens onto the back aperture Because the metalens is only designed to be diffrac-
of the metalens substrate. The relay lens ensures that all tion limited at 852 nm, it is important to characterize the
the deflected beams coincide on the metalens to minimize imaging performance of the lens at the atomic fluores-
asymmetric beam clipping. cence wavelength of 780 nm. To measure the chromatic

030316-8
SINGLE-ATOM TRAPPING... PRX QUANTUM 3, 030316 (2022)

(a) (b) the imaging system by only adjusting one optical element.
φy @ 850 nm
Metasurface The inset of Fig. 5(d) shows the ray-tracing simulation of
the imaging system for both on axis and 10 μm off axis on
φx @ 780 nm
the tweezer plane, verifying that the compensation lens and
850 nm tube lens do not introduce aberrations. The ray-tracing sim-
high-index
a-silicon

ulation does not include aberration inherent to the metalens

780 nm design.
To characterize the compensated imaging system, the
same 300-nm-diameter double pinhole is imaged again
with the pinhole positioned at the 852-nm focus of the
Incident wave Focused wave front
metalens and illuminated with 780-nm light. The resulting
(c) (d) PSF has a waist of (1.1 ± 0.07) μm, which is not diffrac-
780 nm, x-polarized 850 nm, y-polarized tion limited (due to the metalens spherical aberration at
780 nm) but sufficient for single-atom detection, and the
effective solid angle for light collection is equivalent to
0.55 NA.

1 μm 1 μm
10. Loading and detection optical parameters
Normalized intensity

Normalized intensity

The single-atom loading starts with the three-dimensional

science MOT. The atoms from the dispenser in the source
cell are cooled in the transverse direction with the MOT
laser red detuned from the 87 Rb D2 F = 2 to F  = 3
Radial distance (μm) Radial distance (μm)
transition (free-space atomic resonance) by 14 MHz and
transported to the science cell via a push laser beam. The
FIG. 6. The polarization-multiplexed metalens. (a) An a-Si collimated atom beam has a flux up to 108 s−1 . The science
nanopillar (dark blue) with a rectangular cross section on a fused- MOT loading lasts for 500 ms with a typical MOT size of
silica substrate (light blue) creates a phase delay for two orthog- 3 × 107 atoms and a density of 109 cm−3 . After loading,
onal polarizations independently (polarization multiplexing). (b) the source MOT lasers are shut off, the magnetic field gra-
An illustration of the polarization-multiplexed metasurface-lens dient is turned off, and the MOT lasers are changed to 144
operation. The input wave front (pink dash) with orthogonal MHz red detuned from the free-space atomic resonance to
polarization for 780 nm (green) and 852 nm (blue) propagates
perform PGC with a σ+ -σ− configuration for 15 ms. Dur-
and interacts with the metasurface. Secondary wavelets (black
semicircles) reemitted by the nanopillars interfere and create ing the PGC, the optical tweezer is turned on to load atoms
identical focusing wave fronts for both 780 nm and 852 nm. into the tweezer. The typical free-space PGC temperature
(c),(d) The experimental PSF focused by the metalens for 780- is between 30 μK and 50 μK and the tweezer trap depth is
nm x-polarized and 852-nm y-polarized input light, imaged with typically at 1.3 mK during loading. During the PGC load-
an 0.95-NA microscope objective without changing the focus. ing, the laser is red detuned from the atomic resonance,
Two-dimensional cuts of the PSF show the fitted Airy function resulting in light-assisted collision that assures that only a
from which the NA is extracted. single atom is trapped [27].
To image a single atom in the tweezer, we utilize PGC
focal shift, the metalens is illuminated with a collimated imaging. The PGC configuration with less detuning cools
tunable laser source and the focused spot is imaged with the atom while scattering photons. The trapped atom is
an apochromatic microscope objective with a NA of 0.9. illuminated with a 500-μm waist and a 150-μW PGC and
By changing the microscope focus, we determine the chro- probing beam [beam geometry shown in Fig. 5(e), approx-
matic focal shift to be +300 μm between 852 nm and 780 imately 10Isat , 47 MHz red detuned from free-space atomic
nm. We then calculate that the signal of an atom trapped at resonance] for 25 ms and the fluorescence is collected by
852-nm focus and emitting 780-nm fluorescence diverges the metalens (Isat is the saturation intensity of the 87 Rb
with an effective focal length (EFL) of −39 mm after pass- D2 F = 2 to F  = 3 transition). After passing through the
ing through the metalens [Fig. 5(d)]. To compensate, a lens metalens, the slightly diverging fluorescence is reflected by
of EFL 150 mm [L1 in Fig. 5(d), Thorlabs AC254-150-B] a dichroic mirror and passed through the compensation and
is placed 111 mm from the metalens. The combined opti- imaging lenses [L1 and L2 in Fig. 5(d)] and focused onto
cal system (metalens + L1) becomes infinitely conjugate, a Princeton Instruments PIXIS 1024B CCD camera. The
so the tube lens [L2 in Fig. 5(d), Thorlabs AC254-150-B] imaging loss rate is higher than typical PGC imaging due
is decoupled from the compensated imaging system. L2 to the probe beam being perpendicular to the metalens sub-
is prealigned to the camera and L1 is translated to focus strate, so no axial cooling is present during PGC imaging.

030316-9
T.-W. HSU et al. PRX QUANTUM 3, 030316 (2022)

While a full analysis of the fluorescence collection effi- and that for 850-nm y-polarized light is 0.82 ± 0.01. The
ciency requires calibration of the probe light intensity, trap errors quoted are the standard error of the fitted Airy func-
depth, and imaging path efficiency, we can roughly com- tion. The focusing efficiencies for x-polarized light at λ =
pare the expected and measured CCD counts. The total 780 and for y-polarized light at λ = 850 are 42% and 45%,
fluorescence collected in the experiment is around 3% respectively. In order to work toward atom trapping in a
and the expected efficiency is 4.5% when only account- polarization-multiplexed lens, a lens must be fabricated
ing for the solid angle at NA 0.55 and the efficiency of with more perfect rectangular pillars to slightly improve
the metalens but not other loss in the imaging path. The the efficiency and with a larger diameter to increase the
data presented are in CCD counts and are not converted working distance. This presents more design and fabrica-
to photon counts. The intensity variation in the averaged tion challenges and such studies will be a subject of future
atom-array image presented in Figs. 1(d) and 3(a) stems work.
from varying trap depths and aberrations that affect both
loading and imaging. In the array-trapping experiment, the
optical powers of the traps are equalized to within 5% rel-
ative difference at the Fourier plane in between the relay [1] H. Labuhn, D. Barredo, S. Ravets, S. d. Leseleuc, T. Macri,
T. Lahaye, and A. Browaeys, Tunable two-dimensional
lens but due to aberrations the actual trap depth deviates arrays of single Rydberg atoms for realizing quantum Ising
from the on-axis trap. models, Nature 534, 667 (2016).
[2] D. S. Weiss and M. Saffman, Quantum computing with
11. Trap depth and frequency measurement neutral atoms, Phys. Today 70, 44 (2017).
parameters [3] M. A. Norcia, A. W. Young, W. J. Eckner, E. Oelker, J.
Ye, and A. M. Kaufman, Seconds-scale coherence on an
To measure the tweezer trap depth, a resonant heating optical clock transition in a tweezer array, Science 366, 93
beam is used. Between the two consecutive PGC imag- (2019).
ing sequences, the heating-beam intensity is set to 100 μW [4] X. L. Zhang, L. Isenhower, A. T. Gill, T. G. Walker, and M.
(approximately 6Isat ) and is pulsed on for 60 μs. The pulse Saffman, Deterministic entanglement of two neutral atoms
time and intensity of the resonant heating beam are care- via Rydberg blockade, Phys. Rev. A 82, 030306(R) (2010).
fully chosen such that the atom only heats out near the [5] T. Wilk, A. Gaëtan, C. Evellin, J. Wolters, Y. Miroshny-
chenko, P. Grangier, and A. Browaeys, Entanglement of
atomic resonance. The atom survival versus the heating-
Two Individual Neutral Atoms Using Rydberg Blockade,
beam detuning is recorded by measuring the atom survival Phys. Rev. Lett. 104, 010502 (2010).
between the two PGC images [Fig. 2(c)]. [6] H. Levine, A. Keesling, G. Semeghini, A. Omran, T. T.
To measure the trap frequency, the tweezer trap depth Wang, S. Ebadi, H. Bernien, M. Greiner, V. Vuletic, H.
is modulated around the nominal value between the con- Pichler, and M. D. Lukin, Parallel Implementation of High-
secutive PGC imaging and the atom survival is recorded Fidelity Multiqubit Gates with Neutral Atoms, Phys. Rev.
as a function of the modulation frequency [Fig. 2(d)]. The Lett. 123, 170503 (2019).
modulation depth ranges between 5% and 10% of the trap [7] D. Bluvstein, H. Levine, G. Semeghini, T. T. Wang, S.
depth and the modulation time ranges from 30 ms to 50 ms. Ebadi, M. Kalinowski, A. Keesling, N. Maskara, H. Pich-
ler, M. Greiner, and V. Vuletic, A quantum processor based
on coherent transport of entangled atom arrays, Nature 604,
12. Polarization-multiplexed metalens 451 (2022).
To illustrate the flexibility of metalens functionalities, [8] I. S. Madjarov, J. P. Covey, A. L. Shaw, J. Choi, A. Kale,
A. Cooper, H. Pichler, V. Schkolnik, J. R. Williams, and
we fabricate and test a polarization-multiplexed metal-
M. Endres, High-fidelity entanglement and detection of
ens that can provide diffraction-limited focusing at both alkaline-earth Rydberg atoms, Nat. Phys. 16, 857 (2020).
the trapping and fluorescence wavelengths with NA = 0.8. [9] N. Schine, A. W. Young, W. J. Eckner, M. J. Martin, and A.
This metalens is designed to implement an ideal-lens phase M. Kaufman, Long-lived Bell states in an array of optical
function for x-polarized light at λ = 780 nm and for y- clock qubits (2021), arXiv preprint ArXiv:2111.14653.
polarized light at λ = 850 nm, simultaneously, targeting a [10] E. W. Streed, B. G. Norton, A. Jechow, T. J. Weinhold, and
focal length of 0.5 mm and NA of 0.8, for both wavelength D. Kielpinski, Imaging of Trapped Ions with a Microfab-
[Fig. 6(b)]. This is achieved by assembling the metalens ricated Optic for Quantum Information Processing, Phys.
with a library of a-Si birefringent nanopillar unit cells, each Rev. Lett. 106, 010502 (2011).
[11] K. R. Brown, J. Kim, and C. Monroe, Co-designing a
having an in-plane rectangular cross section that can simul-
scalable quantum computer with trapped atomic ions, npj
taneously impose the desired local phase shift φx for x- Quantum Inf. 2, 1 (2016).
polarized light at λ = 780 nm and φy for y-polarized light [12] K. K. Mehta, C. Zhang, M. Malinowski, T.-L. Nguyen,
at λ = 850 nm [Fig. 6(a)]. The measured intensity distribu- M. Stadler, and J. P. Home, Integrated optical multi-ion
tions at the targeted focal plane reveal diffraction-limited quantum logic, Nature 586, 533 (2020).
focal spots for both wavelengths [Figs. 6(c) and 6(d)]. The [13] D. Kim, A. Keesling, A. Omran, H. Levine, H. Bernien,
measured NA for 780-nm x-polarized light is 0.8 ± 0.01 M. Greiner, M. D. Lukin, and D. R. Englund, Large-scale

030316-10
SINGLE-ATOM TRAPPING... PRX QUANTUM 3, 030316 (2022)

uniform optical focus array generation with a phase spatial A. Browaeys, and P. Grangier, Diffraction-limited optics
light modulator, Opt. Lett. 44, 3178 (2019). for single-atom manipulation, Phys. Rev. A 75, 013406
[14] D. Barredo, V. Lienhard, S. De Leseleuc, T. Lahaye, and A. (2007).
Browaeys, Synthetic three-dimensional atomic structures [31] L. R. Liu, J. D. Hood, Y. Yu, J. T. Zhang, K. Wang, Y.-W.
assembled atom by atom, Nature 561, 79 (2018). Lin, T. Rosenband, and K.-K. Ni, Molecular assembly of
[15] S. M. Kamali, E. Arbabi, A. Arbabi, and A. Faraon, A ground-state cooled single atoms, Phys. Rev. X 9, 021039
review of dielectric optical metasurfaces for wavefront (2019).
control, Nanophotonics 7, 1041 (2018). [32] K. Singh, S. Anand, A. Pocklington, J. T. Kemp, and H.
[16] W. T. Chen, A. Y. Zhu, and F. Capasso, Flat optics with Bernien, A Dual-Element, Two-Dimensional Atom Array
dispersion-engineered metasurfaces, Nat. Rev. Mater. 5, with Continuous-Mode Operation, Phys. Rev. X 12, 011040
604 (2020). (2021).
[17] M. Khorasaninejad, W. T. Chen, R. C. Devlin, J. Oh, A. [33] L. Anderegg, L. W. Cheuk, Y. Bao, S. Burchesky, W. Ket-
Y. Zhu, and F. Capasso, Metalenses at visible wavelengths: terle, K.-K. Ni, and J. M. Doyle, An optical tweezer array
Diffraction-limited focusing and subwavelength resolution of ultracold molecules, Science 365, 1156 (2019).
imaging, Science 352, 1190 (2016). [34] K.-N. Schymik, S. Pancaldi, F. Nogrette, D. Barredo,
[18] B. H. Chen, P. C. Wu, V. C. Su, Y. C. Lai, C. H. Chu, I. C. J. Paris, A. Browaeys, and T. Lahaye, Single Atoms
Lee, J. W. Chen, Y. H. Chen, Y. C. Lan, C. H. Kuan, and with 6000-Second Trapping Lifetimes in Optical-Tweezer
D. P. Tsai, GaN metalens for pixel-level full-color routing Arrays at Cryogenic Temperatures, Phys. Rev. Appl. 16,
at visible light, Nano Lett. 17, 6345 (2017). 034013 (2021).
[19] C. Zhang, S. Divitt, Q. Fan, W. Zhu, A. Agrawal, Y. Lu, T. [35] K. Shen, Y. Duan, P. Ju, Z. Xu, X. Chen, L. Zhang, J. Ahn,
Xu, and H. J. Lezec, Low-loss metasurface optics down to X. Ni, and T. Li, On-chip optical levitation with a metalens
the deep ultraviolet region, Light: Sci. Appl. 9, 55 (2020). in vacuum, Optica 8, 1359 (2021).
[20] M. M. R. Elsawy, S. Lanteri, R. Duvigneau, J. A. Fan, and [36] T.-W. Hsu, T. Thiele, W. Zhu, M. O. Brown, S. B. Papp,
P. Genevet, Numerical optimization methods for metasur- A. Agrawal, and C. A. Regal, in CLEO: QELS Fundamen-
faces, Laser Photon. Rev. 14, 1900445 (2020). tal Science, Optical Society of America (OSA, San Jose,
[21] Z. Lin, C. Roques-Carmes, R. E. Christiansen, M. Soljačć, California, 2020), p. JW2A–12.
and S. G. Johnson, Computational inverse design for ultra- [37] B. J. Lester, N. Luick, A. M. Kaufman, C. M. Reynolds, and
compact single-piece metalenses free of chromatic and C. A. Regal, Rapid Production of Uniformly Filled Arrays
angular aberration, Appl. Phys. Lett. 118, 041104 (2021). of Neutral Atoms, Phys. Rev. Lett. 115, 073003 (2015).
[22] M. Liu, W. Zhu, P. Huo, L. Feng, M. Song, C. Zhang, [38] Y. Hu, X. Wang, X. Luo, X. Ou, L. Li, Y. Chen, P. Yang, S.
L. Chen, H. J. Lezec, Y. Lu, A. Agrawal, and T. Xu, Wang, and H. Duan, All-dielectric metasurfaces for polar-
Multifunctional metasurfaces enabled by simultaneous and ization manipulation: Principles and emerging applications,
independent control of phase and amplitude for orthogonal Nanophotonics 9, 3755 (2020).
polarization states, Light Sci. Appl. 10, 107 (2021). [39] C. Chen, S. Gao, X. Xiao, X. Ye, S. Wu, W. Song, H. Li, S.
[23] E. Arbabi, J. Li, R. J. Hutchins, S. M. Kamali, A. Arbabi, Y. Zhu, and T. Li, Highly efficient metasurface quarter-wave
Horie, P. Van Dorpe, V. Gradinaru, D. A. Wagenaar, and A. plate with wave front engineering, Adv. Photon. Res. 2,
Faraon, Two-photon microscopy with a double-wavelength 2000154 (2021).
metasurface objective lens, Nano Lett. 18, 4943 (2018). [40] B. Groever, W. T. Chen, and F. Capasso, Meta-lens doublet
[24] J. Yang, S. Gurung, S. Bej, P. Ni, and H. W. H. Lee, in the visible region, Nano Lett. 17, 4902 (2017).
Active optical metasurfaces: Comprehensive review on [41] S. Shrestha, A. C. Overvig, M. Lu, A. Stein, and N. Yu,
physics, mechanisms, and prospective applications, Rep. Broadband achromatic dielectric metalenses, Light: Sci.
Prog. Phys. 85, 036101 (2022). Appl. 7, 85 (2018).
[25] L. Zhu, X. Liu, B. Sain, M. Wang, C. Schlickriede, Y. [42] T. Phan, D. Sell, E. W. Wang, S. Doshay, K. Edee, J.
Tang, J. Deng, K. Li, J. Yang, M. Holynski, and S. Zhang, Yang, and J. A. Fan, High-efficiency, large-area, topology-
A dielectric metasurface optical chip for the generation of optimized metasurfaces, Light: Sci. Appl. 8, 48 (2019).
cold atoms, Sci. Adv. 6, eabb6667 (2020). [43] W. T. Chen, A. Y. Zhu, J. Sisler, Y.-W. Huang, K. M. A.
[26] W. R. McGehee, W. Zhu, D. S. Barker, D. Westly, A. Yousef, E. Lee, C.-W. Qiu, and F. Capasso, Broadband
Yulaev, N. Klimov, A. Agrawal, S. Eckel, V. Aksyuk, and achromatic metasurface-refractive optics, Nano Lett. 18,
J. J. McClelland, Magneto-optical trapping using planar 7801 (2018).
optics, New J. Phys. 23, 013021 (2021). [44] D. K. Nikolov, A. Bauer, F. Cheng, H. Kato, A. N.
[27] N. Schlosser, G. Reymond, I. Protsenko, and P. Grangier, Vamivakas, and J. P. Rolland, Metaform optics: Bridging
Sub-Poissonian loading of single atoms in a microscopic nanophotonics and freeform optics, Sci. Adv. 7, eabe5112
dipole trap, Nature 411, 1024 (2001). (2021).
[28] W. S. Bakr, J. I. Gillen, A. Peng, S. Fölling, and M. Greiner, [45] W. T. Chen, A. Y. Zhu, V. Sanjeev, M. Khorasaninejad,
A quantum gas microscope for detecting single atoms in a Z. Shi, E. Lee, and F. Capasso, A broadband achromatic
Hubbard-regime optical lattice, Nature 462, 74 (2009). metalens for focusing and imaging in the visible, Nat.
[29] A. M. Kaufman, B. J. Lester, and C. A. Regal, Cooling a Nanotechnol. 13, 220 (2018).
single atom in an optical tweezer to its quantum ground [46] A. Arbabi, E. Arbabi, M. Mansouree, S. Han, S. M. Kamali,
state, Phys. Rev. X 2, 041014 (2012). Y. Horie, and A. Faraon, Increasing efficiency of high
[30] Y. R. Sortais, H. Marion, C. Tuchendler, A. M. Lance, M. numerical aperture metasurfaces using the grating averag-
Lamare, P. Fournet, C. Armellin, R. Mercier, G. Messin, ing technique, Sci. Rep. 10, 1 (2020).

030316-11