Articles

https://doi.org/10.1038/s41565-020-0754-x

High quality factor phase gradient metasurfaces

Mark Lawrence 1,6 ✉, David R. Barton III 1,6 ✉, Jefferson Dixon 2, Jung-Hwan Song 3
,
Jorik van de Groep 3,5, Mark L. Brongersma 3 and Jennifer A. Dionne 1,4 ✉

Dielectric microcavities with quality factors (Q-factors) in the thousands to billions markedly enhance light–matter interac-
tions, with applications spanning high-efficiency on-chip lasing, frequency comb generation and modulation and sensitive
molecular detection. However, as the dimensions of dielectric cavities are reduced to subwavelength scales, their resonant
modes begin to scatter light into many spatial channels. Such enhanced scattering is a powerful tool for light manipulation, but
also leads to high radiative loss rates and commensurately low Q-factors, generally of order ten. Here, we describe and experi-
mentally demonstrate a strategy for the generation of high Q-factor resonances in subwavelength-thick phase gradient meta-
surfaces. By including subtle structural perturbations in individual metasurface elements, resonances are created that weakly
couple free-space light into otherwise bound and spatially localized modes. Our metasurface can achieve Q-factors >2,500
while beam steering light to particular directions. High-Q beam splitters are also demonstrated. With high-Q metasurfaces, the
optical transfer function, near-field intensity and resonant line shape can all be rationally designed, providing a foundation for
efficient, free-space-reconfigurable and nonlinear nanophotonics.

D
ue to their small mode volumes and extremely high qual- meeting or exceeding the performance of their bulk optical counter-
ity factors (Q-factors), microcavities have transformed both parts. Unfortunately, the Q values of metasurface elements have so
the way we manipulate light and how it is used to interro- far been limited to a few tens. Consequently, nonlinear behaviour
gate physical systems. Whispering gallery resonators in particular, has been seen only when pumping the devices with high-power
with Q-factors ranging from thousands to billions1–3, correspond- femtosecond pulses28–30. Similarly, with weakly resonant scattering
ing to resonant lifetimes from picoseconds to microseconds, have being rather insensitive to changes in refractive index, dynamic
produced unprecedented light–matter coupling strengths. For modulation of phase gradient structures has been difficult; to date,
example, whispering gallery resonators including microtoroids achieving metasurface modulation depths exceeding a few decibels
and microdroplets have enabled parametric oscillation4, stimulated has proved challenging31,32.
Raman scattering5, cascaded Raman emission6 and Kerr bistability;7 Here, we unveil a new approach combining the power of high-Q
while these optical nonlinearities were previously measurable only resonances, seen previously in microcavities, with the full wave-
with ultra-high-power pulsed lasers, high-Q cavities made them front control enabled by subwavelength diffractive flat optics. By
observable with continuous-wave sources with power levels as low judicious arrangement of isolated silicon nanoantennas supporting
as microwatts. Important devices and applications have also been specific Mie and guided-mode resonant distributions, we measure
realized with chip-integrated high-Q photonic crystal defect cavi- Q-factors as high as 2,500 in beam-steering and beam-splitting
ties8 and ring/whispering gallery resonators9,10, including Raman phase gradient metasurfaces. We also numerically extend the con-
lasing11 and nonlinear non-reciprocity12 with pump thresholds as cept of high-Q beam steering to other spectral transfer functions,
low as 1 µW, electrically pumped lasing with nA threshold current13, including narrowband and slow-light beam steering. In all cases,
frequency comb generation14 and modulation15, quantum entangled with silicon having an extinction coefficient <1 × 10–9 for wave-
photon generation16, single-photon switching17 and isolation18 and lengths >1.3 µm, the quality factor is limited only by long-range
even virus detection at the single-particle level19. uniformity and lithographic resolution.
As the dimensions of an optical cavity are reduced to the sub-
wavelength scale, they take on an entirely different character. In this Design of guided-mode resonant beam-steering
size regime, resonant modes start to resemble point sources, scat- metasurface
tering an incident wave in many different directions20. While the Throughout this article, all structures are based on arrays of
large number of scattering channels generally translates into a high 600-nm-tall silicon bars atop a sapphire substrate. We will show
radiative loss rate and thus low Q-factor, these nanoscale objects, that this simple platform can be engineered to exhibit unique
or nanoantennas, have proved to be excellent building blocks for light-trapping and -scattering properties by combining two distinct
the construction of phase gradient metasurfaces21–23. Consisting of design strategies. The first of these relies on the fact that each bar sup-
non-uniform arrays of nanoscale dielectric or metallic resonators, ports a pair of spectrally overlapping electric and magnetic dipole Mie
metasurfaces sculpt the spatial distribution of scattered phase24, resonances when illuminated with transverse-magnetic-polarized
amplitude25 and polarization26,27 of an impinging light wave with light (Supplementary Fig. 1). This overlap ensures that transmission
subwavelength resolution. Such flat optical devices provide a versa- remains high while the phase delay through a given bar can be varied
tile and highly compact route to free-space light manipulation, often by adjusting its width (Supplementary Fig. 2). When combined into

1
Department of Materials Science and Engineering, Stanford University, Stanford, CA, USA. 2Department of Mechanical Engineering, Stanford University,
Stanford, CA, USA. 3Geballe Laboratory for Advanced Materials, Stanford University, Stanford, CA, USA. 4Department of Radiology, Stanford University,
Stanford, CA, USA. 5Present address: Van der Waals-Zeeman Institute for Experimental Physics, Institute of Physics, University of Amsterdam, Amsterdam,
Netherlands. 6These authors contributed equally: Mark Lawrence, David R. Barton III. ✉e-mail: [email protected]; [email protected];
[email protected]

956 Nature Nanotechnology | VOL 15 | November 2020 | 956–961 | www.nature.com/naturenanotechnology

Nature NanOtecHnOlOgy Articles
a c e
1

650

2π/k (nm)

∣E ∣/ ∣E 0 ∣
550
500 nm

450 0 0 110
1,400 1,500 1,600 ∣E ∣/ ∣E 0 ∣
Wavelength (nm)

b d 0.4 f
+1st
0th
–1st +1st
Diffraction efficiency
0.3
θ

0.2 Si Q = 8,200

E
0th
0.1 kinc Λ

–1st 500 nm d 2 µm
0
1,380 1,420 1,460 1,500
Wavelength (nm)

Fig. 1 | Conceptual and numerical design of high-Q phase gradient metasurfaces. a, Schematic illustrating broadband beam steering with a Mie resonant
phase gradient metasurface. b, Schematic illustrating strong light localization and modified diffraction after periodic perturbation of phase gradient metasurface
shown in a. c, Waveguide dispersion for phase gradient metasurface shown in a. Left: guided-mode wavelength (2π/k) plotted against free-space wavelength
(λ), with dashed line denoting 2π/k = 570 nm. Right: electric field distributions, colour coded to match dispersion plot. Arrows represent electric field
polarization. |E|/|E0| denotes the amplitude of the electric E field normalized by the incident field amplitude. d, Simulated diffraction spectra for periodically
perturbed phase gradient metasurface. Inset: schematic showing illumination and diffraction configuration, with kinc representing the incident wavevector and
diffraction arrows colour coded to match spectra. Plots of the 0th and −1st orders are reproduced in Supplementary Fig. 4 to better visualize the GMR line shape
in these orders. e, Top: SEM image of metasurface fabricated with nominal dimensions matching those in d. Bottom: simulated electric near-field distribution
at the GMR centre wavelength corresponding to that in d. Arrows represent electric field polarization. f, Angled SEM images, with enlargement in inset, of
fabricated phase gradient metasurfaces demonstrating uniform patterning and minimal sidewall tapering. Perturbation period (Λ) and depth (d) are labelled.

a single array, known as a Huygens metasurface33–35, bars of differ- If a sufficient number of notches is used, light will leak back out
ent width then act as phase ‘pixels’, collectively generating a desired to free space appearing as a guided-mode resonance (GMR) in the
wavefront for transmitted light. Figure 1a illustrates the optical scat- diffraction spectra when Λ equals the guided-mode wavelength36,37.
tering from a periodic metasurface consisting of three equally spaced This contrasts with grating couplers where, instead, energy leaves
nanobars of width 190, 260 and 350 nm, repeated every 2,121 nm. through the end of the waveguide. A numerical example of such
This choice of width corresponds to a constant relative phase delay a resonance, for 100-nm-deep, 100-nm-long notches placed within
between neighbouring elements equal to 2π/3, approximating a lin- the largest bar every 570 nm, is given in Fig. 1d. As expected from
ear phase gradient (Supplementary Fig. 3). Such a device is known the phase gradient design, efficient beam steering occurs across
to steer a normally incident beam to the angle θt, according to the most of the plotted spectral range, indicated by the dominance
generalized Snell’s law θt = arcsin(λ/ntp)21, where p is the metasur- of the +1st diffraction order (red curve). Close to λ = 1,440 nm,
face period, λ is the free-space wavelength and nt is the transmis- a GMR can be seen to interrupt the broadband response, reduc-
sion refractive index. In our case, p = 2,121 nm placing θt in the range ing transmission and creating a narrow reflective band. The high
40–45° for λ = 1,380–1,500 nm. Q-factor, approximately 8,200, of this resonance indicates that the
Our second and most important design concept, illustrated sche- notch dimensions used represent a weak perturbation to the con-
matically in Fig. 1b, involves subtle perturbation of individual nano- tinuous nanoantenna. In the near field, the long resonant lifetime
antennas to generate high-Q dipole resonances. To this end, we note translates to a field enhancement factor >100 within the perturbed
that each nanoantenna maintains a degree of translational symme- element while the unperturbed elements show little enhancement
try (Fig. 1a). Therefore, in addition to modification of free-space (Fig. 1e). Unlike most microcavities where the injection efficiency
optical signals, each metasurface element can also act as a wave- for each mode depends on the specific mode shape, GMRs combine
guide. This can be seen from Fig. 1c, which shows the calculated feedback and coupling to the external environment into a single
waveguide dispersion for the first four modes within the metasur- mechanism. This means that, in the absence of parasitic dissipa-
face. The lowest- and third-lowest-order modes, represented by tion effects, the conditions for critical coupling are satisfied almost
black and red curves, respectively, correspond to guided solutions automatically38. Importantly for our high-Q metasurface design,
of the widest bar (see near fields in Fig. 1c). All these modes possess the absorption losses of silicon in the near infrared are negligible
larger momentum than free-space radiation and so are fully bound and so the resonant lifetime is dominated by radiative loss; indeed,
to their respective nanoantennas. This momentum mismatch can for small notch depths the radiative loss is entirely dependent on
be bridged by introducing a series of periodic notches, of period Λ. the notch structure. For example, after reproducing Fig. 1d with

Nature Nanotechnology | VOL 15 | November 2020 | 956–961 | www.nature.com/naturenanotechnology 957

Articles Nature NanOtecHnOlOgy

a 1,470 e
Maximum ~30 nm
0th 0.4
–1st +1st
1,470 θ

+1st diffraction efficiency

Wavelength (nm)

Si 600 nm 0.3

Counts
1,439 1,439
E 707 nm
Λ (nm) = 550
kinc
0.2
1,410
–50 –35
570
590
610
0
1,410 0.1
–50 –40 –30 –20 –10 0 10 20 30 40 50 1,420 1,460 1,500 1,540
Diffraction angle (°) Wavelength (nm)

b 0.4 c d 2.6 f 0.3

0.4
2.2
0.3
+1st diffraction efficiency

+1st
+1st
Diffraction efficiency

Diffraction efficiency
0.2

Q-factor (×103)
0.3 1.8
0.2

50
1.4 0.1
0.2 70
0.1 kinc
–1st 0th 0th
100 1.0 –1st
125
d (nm) = 150
0 0.1 0
1,380 1,420 1,460 1,500 1,410 1,430 1,450 1,470 50 100 150 1,520 1,540 1,560 1,580 1,600
Wavelength (nm) Wavelength (nm) Notch depth (nm) Wavelength (nm)

Fig. 2 | Experimental demonstration of high-Q phase gradient metasurface beam steering. a, Fourier plane spectral image of beam-steering phase gradient
metasurface, with Λ = 570 nm and d = 100 nm. Left inset: enlargement of −1st diffraction order, with value amplified 10×. Right inset: schematic showing
illumination and diffraction configuration, with diffraction arrows colour coded to match spectra in b. b, Diffraction spectra for sample used in a and normalized
to transmission through the bare substrate, with each diffraction order measured separately. c, +1st diffraction spectra for metasurfaces, with perturbation
depth d decreasing from 150 to 50 nm and measurement performed as in b. d, Extracted quality factors from c, colour coded to match spectra. The dashed line
was drawn as a visual guide. e, +1st diffraction spectra for metasurfaces with d = 70 nm and varying perturbation period Λ, spanning 550–610 nm in 20-nm
increments, with measurements performed as in b. Note that the purple curve in c and orange curve in e are from structures of nominally identical dimensions
but were patterned on different samples. Extracted resonant wavelengths for the data in c,e are included in Supplementary Fig. 9, and extracted quality factors
for data in e are provided and compared to theory in Supplementary Fig. 10. f, Lower-frequency GMR measured in metasurface, with Λ = 570 nm and d = 100 nm
and measurements performed as in b. Inset schematically compares dominant diffraction on-resonance (solid arrows) and off-resonance (dashed arrow).

the perturbation depth reduced to 50 nm, Q increases to 39,000 imaging spectrometer. The Fourier plane spectral map of a meta-
(Supplementary Fig. 5). We note that GMRs have previously been surface with Λ = 570 nm and d = 100 nm shows strong preferential
investigated in two-dimensionally periodic nanostructures36,39–41. In scattering into the +1st diffraction order with minimal diffraction
these studies, lattice interference across a one- or two-dimensional into other directions, verifying the linear phase gradient design
grating suppresses free-space scattering, giving rise to very sharp (Fig. 2a). A narrow dip in the +1st diffraction near 1,440 nm reveals
spectral features under free-space illumination. Considerable the presence of a GMR. This GMR can also be seen to modulate the
work has been directed towards locating and implanting these other two orders, most clearly visible in the inset of Fig. 2a showing
resonances in momentum space, and to understanding the phys- a close-up of the −1st diffraction, albeit with a smaller amplitude
ics governing their formation and collapse42–45. Instead, crucially, and distinct line shape. This direction-dependent line shape arises
by not relying simply on modes that are de-localized throughout a from interference between the weakly coupled waveguide mode and
two-dimensional plane, our silicon nanobars here maintain a dipole directly transmitted Mie mode, which we will show later allows for
radiation pattern in the plane of diffraction, giving much finer con- diverse functionality. To ensure efficient free-space coupling and
trol over scattered wavefronts. reduce finite-size effects, we fabricated metasurfaces of large area
(300 × 300 µm2). We note, however, that the effect does not rely on
Measurement of high-Q beam steering large areas (Supplementary Fig. 6) and have fabricated metasurfaces
Based on the results shown in Fig. 1d, we fabricated a series of phase of only 7 µm in the phase gradient direction (corresponding to a
gradient metasurface samples. A representative scanning electron total of ten metasurface elements) without influencing the high-Q
microscope (SEM) image of one such sample is shown in Fig. 1f, resonant feature (Supplementary Fig. 7).
annotated with the two parameters that we varied experimentally— To better quantify the response of our metasurfaces, we also
notch depth d and period Λ. We then characterize these devices by recorded the real-space spectral image of each diffraction order
illuminating them at normal incidence through the substrate with independently. Using this approach, Fig. 2b shows the normalized
a collimated white-light laser and measuring transmission with an spectra from the sample used in Fig. 2a. Apart from the constant

958 Nature Nanotechnology | VOL 15 | November 2020 | 956–961 | www.nature.com/naturenanotechnology

 Nature NanOtecHnOlOgy Articles
a b tune the spectral position of the GMR. Fixing d = 70 nm, Fig. 2e
Broadband Resonant
0.3 +1 displays the relevant portions of the +1st order diffraction effi-
+1st ciency for Λ = 550 nm (yellow curve) to 610 nm (purple curve), in
Diffraction efficiency

20-nm increments. Here, we see that we can shift the resonant cen-
0.2 0th
tre wavelength by approximately 30 nm for every 20-nm increase

Hz / ∣Hz,max ∣
in period without substantially impacting the background phase
–1st
0.1 gradient profile.
y
Since the perturbations are inserted into the largest bar, a sec-
0 –1
ond GMR exists at a longer wavelength, corresponding to free-space
x
1,485 1,490 1,495 z coupling into a vertically polarized waveguide mode (see the black
Wavelength (nm)
dispersion curve in Fig. 1c). Figure 2f compares the diffraction
c d Broadband Resonant spectra of a metasurface near λ = 1,560 nm. As can be seen, the dif-
80
75 ferent phase relationship between the broad background and local-
0.4
ized mode produces a more asymmetric Fano line shape in the +1st

∣E|/ ∣E0 ∣
+1st
60
diffraction order46. There is little modification to the light scattered
Diffraction efficiency

0.3 700 nm
0 into the −1st diffraction order, though there is a marked increase
+1
∣E|/ ∣E0 ∣

0.2
40 in the 0th order. In other words, on-resonance, the narrowband
response is dominated by direct transmission while off-resonance

Hz / ∣Hz,max ∣
0.1 20 the structure steers light to ∼45°.
0th
–1st
y
0 0 Narrowband and slow-light beam steering
1,485 1,490 1,495
Wavelength (nm) x
–1 Having experimentally verified the possibility of embedding high-Q
z
resonances within the diffraction spectrum of a linear phase gradi-
ent metasurface, we next explored additional opportunities enabled
Fig. 3 | Narrowband and slow-light beam steering. a, Simulated diffraction by this design principle. First, we show how relative weights asso-
spectra of narrowband beam-steering metasurface with asymmetric ciated with the available diffraction orders can be independently
perturbations placed in the thinnest bar. Curve colours are chosen to match tailored both on and off resonance. Figure 2f shows that high-Q
the schematic in the inset of Fig. 1d. b, Magnetic field maps corresponding scattering into the directly transmitted beam can be increased while
to a, at λ = 1,495 nm (broadband) and λ = 1,489.9 nm (resonant). Hz/|Hz,max| the first diffracted order is suppressed. This behaviour, which is dis-
denotes the z component of the magnetic H field normalized by the tinct from the diffraction seen in Fig. 2a–e, arises from the particu-
maximum field strength. c, Simulated diffraction spectra of slow-light lar symmetry of the GMR and its corresponding radiation pattern.
beam-steering metasurface, with identical geometry as in a for the thinnest In Fig. 2a–e, the GMR is approximately an in-plane electric dipole
bar but different neighbouring bar widths. The orange curve represents (Supplementary Fig. 11). In contrast, the response shown in Fig. 2f
peak electric field amplitude (right y axis), while remaining curves is associated with a vertically polarized electric dipole, as seen in the
represent diffraction orders with colours chosen to match the schematic bottom field map in Fig.1c. When compared to an in-plane dipole, a
in the inset of Fig. 1d. d, Electric field amplitude (top) and magnetic vertical dipole couples uniquely to both the incident wave and notch
field (bottom) maps corresponding to c, at λ = 1,495 nm (broadband) symmetry, giving us a new degree of freedom for moulding diffrac-
and λ = 1,490 nm (resonant). Shorter arrows overlaid with magnetic tion. We expect that the behaviour found in Fig. 2b,f will be use-
fields display the local Poynting vector, while longer arrows highlight the ful in switching applications where very small optical, electrical or
dominant diffraction orders. Detailed metasurface dimensions are given in thermal changes to the refractive index can produce large changes
the text. Colour bar label |E|/|E0| in d denotes the amplitude of the electric in diffraction efficiency.
E field normalized by the incident field amplitude. While the radiation pattern of a GMR depends predominantly on
the notch dimensions, the overall metasurface scattering is decided
by the interplay between the GMR and the background phase pro-
oscillations arising from Fabry–Perot resonance in the substrate, we file. In Fig. 3, we harness this interplay to numerically demonstrate
find excellent agreement between Fig. 2b and the numerical data two additional spectral transfer functions of the three-bar design:
presented in Fig. 1d. A quality factor of 1,500 has been extracted narrowband beam steering and slow-light beam steering. To achieve
for this mode. In Fig. 2c we fix Λ = 570 nm and plot relevant por- these functions, we switch to a vertically polarized electric dipole
tions of the +1st diffraction spectra with the notch depth d swept GMR. We also decrease the width of the notched bar to 210 nm,
from 150 nm (yellow curve) to 50 nm (black curve). As d drops to thereby changing the background Mie scattered phase associated
50 nm we observe a red shift of approximately 60 nm as the GMR with the narrow resonance, adjusting its line shape. A key property
approaches the band crossing point of an ideal, or unnotched, wave- of a vertical electric dipole is that it radiates symmetrically about
guide (∼1,500 nm) (Fig. 1c). This red shift makes sense, because a the vertical axis. This means that such a mode cannot be excited by
smaller perturbation will produce less pronounced Bragg scatter- a plane wave travelling in the vertical direction unless the metasur-
ing and subsequently less band splitting in guided-mode dispersion face breaks that inversion symmetry. In Supplementary Fig. 12 it is
(Supplementary Fig. 8). More importantly, from the narrowing line seen that the phase of light emitted to the left and right can be con-
width of the resonant dips in Fig. 2c, as well as the retrieved val- trolled by tuning the relative depth of notches placed on the left and
ues plotted in Fig. 2d, Q is seen to increase from 900 to 2,500. This right sides of the bar. This allows interference between the resonant
increase clearly demonstrates the strong connection between notch scattering and background diffraction to be engineered separately
depth and GMR radiative coupling strength. Unfortunately, mode for different diffracted directions. In contrast, the in-plane mode,
contrast also decreases with decreasing perturbation depth, proba- shown in Supplementary Fig. 11, exhibits very different behaviour
bly caused by scattering loss. Nevertheless, while lower than numer- with the angular intensity of emission, rather than the angular phase
ically predicted due to fabrication imperfections and finite-size distribution, depending on the notch symmetry.
effects, our measured values are still two orders of magnitude higher Taking advantage of these insights, we placed periodic notches of
than reported phase gradient metasurface Q values to date. Aside period 635 nm into both sides of the 210-nm-wide bar. The notches
from line width, we also investigated the ability to systematically were 36-nm deep, 150-nm long on the left (negative x direction)

Nature Nanotechnology | VOL 15 | November 2020 | 956–961 | www.nature.com/naturenanotechnology 959

Articles Nature NanOtecHnOlOgy

a metasurface supercell period while maintaining the linear phase

gradient can alter the steering angle. Supplementary Fig. 13 shows
a suite of samples engineered to steer light to 36o, with high-Q
responses similar to those shown in Fig. 2. Figure 4 shows an experi-
mental demonstration of free-space high-Q metasurface beam split-
Λ ting. In this case, light incident on the metasurface is redirected in
approximately equal magnitude to the +1 and −1 diffraction orders
d while cancelling the 0th-order transmission almost entirely. An
angle-view SEM image of one such fabricated structure is shown
1 µm
in Fig. 4a. The metasurface has the same phase gradient period
(2,121 nm) as the 45° beam-steering samples and is composed of
two phase pixels. One pixel consists of a nanoantenna made from
b
+1st
two coupled 240-nm-wide silicon nanobars. Light leaking through
0.4 –1st the bare substrate between nanoantennas acts as the second pixel.
Measured spectra from a beam-splitting metasurface, shown in
Fig. 4b, reveal balanced diffraction in the +1st (red curve) and
0.3
−1st (purple curve) directions, while the 0th-order transmission
Diffraction efficiency

0th (black curve) is almost entirely suppressed at the centre wavelength

–1st +1st
of 1,400 nm. A schematic representation of the working principle
0.2 Q = 1,088
underlying beam splitting is shown in the inset of Fig. 4b, and
numerically in Supplementary Fig. 14. With neighbouring pixels
• •
0.1 ϕ=π ϕ=0 differing in their phase delay by π, destructive interference occurs
in the forward direction while constructive interference occurs for
0th waves deflected to ∼±41°.
0 Unlike our other, beam-steering devices, the guided modes of
1,340 1,360 1,380 1,400 1,420 1,440 1,460 the metasurface pictured in Fig. 4a are not localized within a single
Wavelength (nm)
nanobar. Instead, coupling between the bars generates a collection of
even and odd waveguide modes (Supplementary Fig. 15). Following
Fig. 4 | High-Q metasurface beam splitter. a, Angled SEM image of a our previous prescription, we pattern the inside edge of each element
representative beam-splitting metasurface, with annotated perturbation with 80-nm-long perturbations of period Λ = 690 nm and depth d
dimensions. b, Measured diffraction spectra from metasurface beam (in Fig. 4a, d = 100 nm to best visualize the structure; in Fig. 4b the
splitter with d = 30 nm and Λ = 690 nm. The corresponding Fourier plane notch depth is 30 nm). Two sharp GMRs appear in the diffraction
spectral map is given in Supplementary Fig. 16. Inset: schematic of working spectra of Fig. 4b at wavelengths 1,389 and 1,398 nm with Q of 1,088
principle of metasurface beam splitter. ϕ denotes the scattered phase and 580, respectively (see Supplementary Fig. 17 for top-down SEM
of each antenna, grey curves represent crests of cylindrically expanding images and a parameter sweep of perturbation geometries). Here,
waves emanating from the antennae and dashed lines highlight plane the beam-splitting efficiency is reduced while direct transmission is
wavefronts aligned with the diffraction orders, formed from constructive resonantly increased. Varying perturbation depth, as before, modi-
interference between cylindrical waves. Diffraction arrows are colour coded fies the Q-factor (see Supplementary Fig. 17). Note that although
to match spectra. four guided modes exist for this period in this spectral range, by
maintaining inversion symmetry in the diffraction plane only the
two antisymmetric GMRs can be excited with the polarization we
use for illumination (see Supplementary Fig. 18).
and 44-nm deep, 150-nm long on the right (positive x direction).
We find that placing this structure between bars of width 275 and Conclusions
280 nm produces sharp dips in both −1st and 0th diffraction orders While nanoantennas represent an exciting development for optical
(Fig. 3a, purple and black curves) but a narrow peak in the +1st science, researchers typically face a trade-off between antenna size
diffraction order (Fig. 3a, red curve). As confirmed by the mag- in relation to wavelength and resonant lifetime. In shaping diffrac-
netic field profiles and overlaid Poynting vectors (Fig. 3b), this tion using high-Q nanoantennas of subwavelength cross-section in
combination results in the metasurface exhibiting balanced dif- the diffraction plane, we provide experimental evidence to suggest
fraction off-resonance but steers the incident wave to ∼45° close that this trade-off may not, in fact, be fundamental. These results
to the GMR (λ = 1,489.9 nm). In Fig. 3c, we take the structure from point to the possibility that highly resonant and highly compact
Fig. 3a and replace the 275-nm-wide bar with a 370-nm-wide bar. features, once in the purview only of on-chip photonics, can be
Here, spectral variation in the diffraction around the GMR is almost rationally designed to coincide with an arbitrarily chosen electro-
entirely removed while maintaining the strong spectrally narrow magnetic wavefront. Our proof-of-principle phase gradient meta-
field enhancement associated with high-Q resonance. Similarly, the surfaces, namely beam steerers and beam splitters, can lead to a host
magnetic field profiles show very little change in the beam-steering of high-Q diffractive flat optics including lenses47 and holograms.
response with wavelength, while maps of electric field amplitude We also demonstrated that diffraction order-dependent GMR dis-
reveal a dramatic enhancement (exceeding 75×) within the per- persion can be controlled via the perturbation symmetry and that
turbed bar on-resonance (Fig. 3d). Accordingly, with our high-Q multiplexing can be achieved using coupled elements. The phenom-
metasurface approach, the optical transfer function, near-field enon we present is not exclusive to nanobars of varying width—we
intensity and resonant line shape can all be rationally designed. expect to find similar behaviour in other design strategies, including
isotropic structures and those based on geometric phase. The pres-
High-Q beam splitting ent study has focused on devices that maintain periodicity along
As a general scheme for realizing high-Q phase gradient metasur- one dimension and thus require sufficient spatial extent in that
faces, structures can easily be designed with wavefront-shaping direction to sustain long resonant lifetimes. This, however, is not an
capabilities beyond 45° beam steering. For example, changing the essential requirement. By both reducing group velocity dispersion

960 Nature Nanotechnology | VOL 15 | November 2020 | 956–961 | www.nature.com/naturenanotechnology

Nature NanOtecHnOlOgy Articles
of the guided modes with high-contrast index variations and using 18. Scheucher, M., Hilico, A., Will, E., Volz, J. & Rauschenbeutel, A. Quantum
photonic crystal mirrors to reflect light that would otherwise leak optical circulator controlled by a single chirally coupled atom. Science 354,
1577–1580 (2016).
through the ends of the nanoantennas, the scale of our metasur- 19. Vollmer, F., Arnold, S. & Keng, D. Single virus detection from the reactive
faces can be reduced considerably. We note that our highest mea- shift of a whispering-gallery mode. Proc. Natl Acad. Sci. USA 105,
sured Q (~2,500) is limited mainly by scattering losses. Much higher 20701–20704 (2008).
Q-factors (104−105) should be achievable with refined fabrication 20. Kuznetsov, A. I., Miroshnichenko, A. E., Brongersma, M. L., Kivshar, Y. S. &
Luk’yanchuk, B. Optically resonant dielectric nanostructures. Science 354,
processes and improved imperfection-tolerant designs. By enabling
aag2472 (2016).
resonant near-field intensity and line shape to be engineered in con- 21. Yu, N. et al. Light propagation with phase discontinuities: generalized laws of
junction with arbitrary wavefront transformations, we envision an reflection and refraction. Science 334, 333–337 (2011).
impact by high-Q phase gradient metasurfaces on any discipline 22. Yu, N. & Capasso, F. Flat optics with designer metasurfaces. Nat. Mater. 13,
that benefits from efficient diffractive optical switching or tuning 139–150 (2014).
23. Lin, D., Fan, P., Hasman, E. & Brongersma, M. L. Dielectric gradient
and low nonlinear thresholds. Specific applications include light metasurface optical elements. Science 345, 298–302 (2014).
detection and ranging (for example LIDAR), light fidelity (LiFi), 24. Zheng, G. et al. Metasurface holograms reaching 80% efficiency. Nat.
and quantum and non-reciprocal optical communication. Nanotechnol. 10, 308–312 (2015).
25. Liu, L. et al. Broadband metasurfaces with simultaneous control of phase and
Online content amplitude. Adv. Mater. 26, 5031–5036 (2014).
26. Balthasar Mueller, J. P., Rubin, N. A., Devlin, R. C., Groever, B. & Capasso, F.
Any methods, additional references, Nature Research report- Metasurface polarization optics: independent phase control of arbitrary
ing summaries, source data, extended data, supplementary infor- orthogonal states of polarization. Phys. Rev. Lett. 118, 113901 (2017).
mation, acknowledgements, peer review information; details of 27. Yin, X., Ye, Z., Rho, J., Wang, Y. & Zhang, X. Photonic spin hall effect at
author contributions and competing interests; and statements of metasurfaces. Science 339, 1405–1407 (2013).
data and code availability are available at https://doi.org/10.1038/ 28. Li, G. et al. Continuous control of the nonlinearity phase for harmonic
generations. Nat. Mater. 14, 607–612 (2015).
s41565-020-0754-x. 29. Almeida, E., Shalem, G. & Prior, Y. Subwavelength nonlinear phase control
and anomalous phase matching in plasmonic metasurfaces. Nat. Commun. 7,
Received: 7 January 2020; Accepted: 7 July 2020; 10367 (2016).
Published online: 17 August 2020 30. Wang, L. et al. Nonlinear wavefront control with all-dielectric metasurfaces.
Nano Lett. 18, 3978–3984 (2018).
References 31. Wu, P. C. et al. Dynamic beam steering with all-dielectric electro-optic III–V
1. Armani, D. K., Kippenberg, T. J., Spillane, S. M. & Vahala, K. J. Ultra-high-Q multiple-quantum-well metasurfaces. Nat. Commun. 10, 3654 (2019).
toroid microcavity on a chip. Nature 421, 925–928 (2003). 32. Li, S. Q. et al. Phase-only transmissive spatial light modulator based on
2. Gorodetsky, M. L., Savchenkov, A. A. & Ilchenko, V. S. Ultimate Q of optical tunable dielectric metasurface. Science 364, 1087–1090 (2019).
microsphere resonators. Opt. Lett. 21, 453–455 (1996). 33. Decker, M. et al. High-efficiency dielectric Huygens’ surfaces. Adv. Opt.
3. Vernooy, D. W., Ilchenko, V. S., Mabuchi, H., Streed, E. W. & Kimble, H. J. Mater. 3, 813–820 (2015).
High-Q measurements of fused-silica microspheres in the near infrared. Opt 34. Pfeiffer, C. et al. Efficient light bending with isotropic metamaterial Huygens’
Lett 23, 247–249 (1998). surfaces. Nano Lett. 14, 2491–2497 (2014).
4. Kippenberg, T. J., Spillane, S. M. & Vahala, K. J. Kerr-nonlinearity optical 35. Estakhri, N. M. & Alù, A. Wave-front transformation with gradient
parametric oscillation in an ultrahigh-Q toroid microcavity. Phys. Rev. Lett. metasurfaces. Phys. Rev. X 6, 041008 (2016).
93, 083904 (2004). 36. Wang, S. S. & Magnusson, R. Theory and applications of guided-mode
5. Lin, H.-B., Eversole, J. D. & Campillo, A. J. Continuous-wave stimulated resonance filters. Appl. Opt. 32, 2606–2613 (1993).
Raman scattering in microdroplets. Opt. Lett. 17, 828–830 (1992). 37. Lawrence, M., Barton, D. R. & Dionne, J. A. Nonreciprocal flat optics with
6. Min, B., Kippenberg, T. J. & Vahala, K. J. Compact, fiber-compatible, silicon metasurfaces. Nano Lett. 18, 1104–1109 (2018).
cascaded Raman laser. Opt. Lett. 28, 1507–1509 (2003). 38. Jiang, X. et al. Chaos-assisted broadband momentum transformation in
7. Treussart, F. et al. Evidence for intrinsic Kerr bistability of high-Q microsphere optical microresonators. Science 358, 344–347 (2017).
resonators in superfluid helium. Eur. Phys. J. D 1, 235–238 (1998). 39. Fan, S. & Joannopoulos, J. D. Analysis of guided resonances in photonic
8. Akahane, Y., Asano, T., Song, B. S. & Noda, S. High-Q photonic nanocavity crystal slabs. Phys. Rev. B 65, 235112 (2002).
in a two-dimensional photonic crystal. Nature 425, 944–947 (2003). 40. Wu, C. et al. Spectrally selective chiral silicon metasurfaces based on infrared
9. Lee, H. et al. Chemically etched ultrahigh-Q wedge-resonator on a silicon Fano resonances. Nat. Commun. 5, 3892 (2014).
chip. Nat. Photonics 6, 369–373 (2012). 41. Yang, Y., Kravchenko, I. I., Briggs, D. P. & Valentine, J. All-dielectric
10. Lin, G., Diallo, S., Henriet, R., Jacquot, M. & Chembo, Y. K. Barium fluoride metasurface analogue of electromagnetically induced transparency. Nat.
whispering-gallery-mode disk-resonator with one billion quality-factor. Opt. Commun. 5, 5753 (2014).
Lett. 39, 6009–6012 (2014). 42. Wei Hsu, C. et al. Observation of trapped light within the radiation
11. Takahashi, Y. et al. A micrometre-scale Raman silicon laser with a microwatt continuum. Nature 499, 188–191 (2013).
threshold. Nature 498, 470–474 (2013). 43. Kodigala, A. et al. Lasing action from photonic bound states in continuum.
12. Peng, B. et al. Parity–time-symmetric whispering-gallery microcavities. Nat. Nature 541, 196–199 (2017).
Phys. 10, 394–398 (2014). 44. Ha, S. T. et al. Directional lasing in resonant semiconductor nanoantenna
13. Ellis, B. et al. Ultralow-threshold electrically pumped quantum-dot arrays. Nat. Nanotechnol. 13, 1042–1047 (2018).
photonic-crystal nanocavity laser. Nat. Photonics 5, 297–300 (2011). 45. Yang, Y. et al. Nonlinear fano-resonant dielectric metasurfaces. Nano Lett. 15,
14. Stern, B., Ji, X., Okawachi, Y., Gaeta, A. L. & Lipson, M. Battery-operated 7388–7393 (2015).
integrated frequency comb generator. Nature 562, 401–405 (2018). 46. Limonov, M. F., Rybin, M. V., Poddubny, A. N. & Kivshar, Y. S. Fano
15. Wang, C. et al. Monolithic lithium niobate photonic circuits for Kerr resonances in photonics. Nat. Photonics 11, 543–554 (2017).
frequency comb generation and modulation. Nat. Commun. 10, 978 (2019). 47. Klopfer, E., Lawrence, M., Barton, D. R., Dixon, J. & Dionne, J. A. Dynamic
16. Guo, X. et al. Parametric down-conversion photon-pair source on a focusing with high-quality-factor metalenses. Nano Lett. 20, 5127–5132 (2020).
nanophotonic chip. Light Sci. Appl. 6, e16249 (2017).
17. Sun, S., Kim, H., Luo, Z., Solomon, G. S. & Waks, E. A single-photon switch Publisher’s note Springer Nature remains neutral with regard to jurisdictional claims in
and transistor enabled by a solid-state quantum memory. Science 361, published maps and institutional affiliations.
57–60 (2018). © The Author(s), under exclusive licence to Springer Nature Limited 2020

Nature Nanotechnology | VOL 15 | November 2020 | 956–961 | www.nature.com/naturenanotechnology 961

Articles Nature NanOtecHnOlOgy

Methods The second multiplicative term represents the superposition between a constant
Design. Sample dimensions were designed using COMSOL Multiphysics to model complex background, ar + aii, and a Lorentzian resonance with resonant
both waveguide dispersion and phase response. The phase response to determine frequency f0 and full-width at half-maximum 2γ. The Q-factor of this resonance
each meta-element was designed using periodic boundary conditions with a silicon is then taken to be Q = f0/2γ. The first term accounts for the Fabry–Perot
bar of given dimensions. interference through the substrate of thickness hs and refractive index ns. k is
the free-space wavevector (2π/λ) and F accounts for the reflectivity of the air–
Fabrication. The metasurfaces were fabricated using standard lithographic substrate interfaces.
procedures. First, 600-nm, single-crystal silicon-on-sapphire (MTI Corp.)
substrates were cleaned by sonication in acetone and isopropyl alcohol. Following Data availability
a dehydration bake, HSQ negative tone resist (XR-1541–06, Corning) was The data that support the plots within this paper and other findings of this study
spin-coated to the sample and baked for 45 min at 90 ºC. To reduce charging, a are available from the corresponding authors on reasonable request.
charge dissipation layer was applied (e-spacer, Showa Denko) and the sample was
baked again. The patterns were written using electron-beam lithography (Jeol
6300-FS) and developed in strong base. Following lithography, the pattern was Acknowledgements
transferred to the silicon layer using reactive ion etching. First, a non-selective We thank R. Tiberio and U. Raghuram for helpful discussions regarding fabrication.
etch was performed to break through the native oxide while the main etch step This work was supported by PECASE (grant no. FA9550-15-10006) and NSF EFRI
utilized Cl2, HBr and O2 (Lam TCP 9400), etching anisotropically and stopping on (grant no. 1641109). The device fabrication, performed in part by J.D., was supported
the sapphire substrate. The resist was stripped using 2% hydrofluoric acid in water, by the DOE ‘Photonics at Thermodynamic Limits’ Energy Frontier Research Center
following an organic clean using a Piranha solution heated to 120 ºC. under grant no. DE-SC0019140. J.v.d.G, J.-H.S. and M.L.B. acknowledge funding
from an individual investigator grant from AFOSR (no. FA9550-18-1-0323). Part
Characterization. Representative images were taken using a FEI Magellan 400 of this work was performed at the Stanford Nano Shared Facilities and Stanford
XHR scanning electron microscope with a field emission gun source or FEI Helios Nanofabrication Facilities, which are supported by the National Science Foundation
600i dual-beam SEM/FIB. For top-down images, a conductive polymer (ESPACER) and National Nanotechnology Coordinated Infrastructure under award no.
was applied to reduce charging in the images. For side and tilted views, a ECCS-1542152.
representative sample was coated with a ~3.5-nm film of Au/Pd to reduce charging.
Images were typically acquired with accelerating voltage 2–5 kV.
Author contributions
M.L., D.R.B. and J.D. conceived the idea. M.L. performed full-field simulations.
Optical characterization. Spectroscopic measurements were made using a
D.R.B., J.D. and J.v.d.G. fabricated the devices. M.L. and J.-H.S. conducted optical
home-built, angle-resolved microscope (Supplementary Fig. 19). A collimated
characterization. D.R.B. and M.L. wrote the manuscript. J.A.D. supervised the project,
broadband laser (NKT supercontinuum) was polarized and weakly focused onto the
along with M.L.B. on the relevant portions of the research. All authors contributed to
metasurface at normal incidence. Illumination was perpendicular to the metasurface
preparation of the manuscript.
on an assembly that can set a vertical incident angle relative to the objective. Insertion
of a Bertrand lens on a flip mount focused on the back focal plane of the objective
allowed projection of the Fourier plane onto a camera. Because the metasurface Competing interests
diffracts in one plane, we use a grating to split the spectral response in the other The authors declare no competing interests.
dimension and thereby we simultaneously obtain the energy and momentum of
the transmitted light. Light is collected with a 0.42-numerical aperture objective
(at a half-angle of 24°). To capture all three diffraction orders, we performed Additional information
two measurements with the sample/illumination angle set to ±20° relative to the Supplementary information is available for this paper at https://doi.org/10.1038/
objective. One measurement contains the 0th and +1st diffraction orders while the s41565-020-0754-x.
other contains the 0th and −1st. The two spectral maps are aligned, and overlapping
Correspondence and requests for materials should be addressed to M.L., D.R.B. or
data are removed (Fig. 2a). Throughout the paper the resonant spectral features were
J.A.D.
analysed by fitting the diffraction efficiency data with the function
  2 Peer review information Nature Nanotechnology thanks Lan Yang and the other,
 1  b  anonymous, reviewer(s) for their contribution to the peer review of this work.
T ¼  ar þ ai i þ
 
 ð1Þ
2
1 þ F sin ðns khs Þ f � f0 þ γi Reprints and permissions information is available at www.nature.com/reprints.

Nature Nanotechnology | www.nature.com/naturenanotechnology