Letter https://doi.org/10.

1038/s41586-018-0450-2

Synthetic three-dimensional atomic structures

assembled atom by atom
Daniel Barredo1,2*, Vincent Lienhard1,2, Sylvain de Léséleuc1,2, Thierry Lahaye1 & Antoine Browaeys1

A great challenge in current quantum science and technology reach depths of U0/kB ≈ 1 mK, where kB is the Boltzmann constant, and
research is to realize artificial systems of a large number of radial (longitudinal) trapping frequencies of around 100 kHz (20 kHz).
individually controlled quantum bits for applications in quantum We produce highly uniform microtrap potentials (with peak inten-
computing and quantum simulation. Many experimental sities differing by less than 5% root mean square) via a closed-loop
platforms are being explored, including solid-state systems,
such as superconducting circuits1 or quantum dots2, and atomic,
molecular and optical systems, such as photons, trapped ions or a
neutral atoms3–7. The latter offer inherently identical qubits that are
well decoupled from the environment and could provide synthetic
structures scalable to hundreds of qubits or more8. Quantum-gas
microscopes9 allow the realization of two-dimensional regular
lattices of hundreds of atoms, and large, fully loaded arrays of
about 50 microtraps (or ‘optical tweezers’) with individual control
are already available in one10 and two11 dimensions. Ultimately,
however, accessing the third dimension while keeping single-atom
control will be required, both for scaling to large numbers and for
extending the range of models amenable to quantum simulation.
Here we report the assembly of defect-free, arbitrarily shaped
three-dimensional arrays, containing up to 72 single atoms. We use
holographic methods and fast, programmable moving tweezers to
arrange—atom by atom and plane by plane—initially disordered
arrays into target structures of almost any geometry. These results
present the prospect of quantum simulation with tens of qubits
arbitrarily arranged in space and show that realizing systems of b c d
hundreds of individually controlled qubits is within reach using
current technology.
Ly = 103 μm

Three-dimensional atomic arrays at half filling have been obtained

using optical lattices with large spacings12, which facilitate single-site
addressability and atom manipulation13. As an alternative approach,
here we use programmable holographic optical tweezers to create m
three-dimensional (3D) arrays of traps. Holographic methods offer Lx = 103 μm 9μ
=9
Lz
the advantage of higher tunability of the lattice geometry because the
design of optical potential landscapes is reconfigurable and only limited Fig. 1 | Experimental setup and trap images. a, We combine a spatial light
by diffraction14–16. In our experiment14, arbitrarily designed arrays of modulator (SLM) and a high-numerical-aperture aspheric lens (AL) under
up to about 120 traps are generated by imprinting a phase pattern on a vacuum to generate arbitrary 3D arrays of traps. The intensity distribution
in the focal plane is measured with the aid of a second aspheric lens, a
dipole trap beam at 850 nm with a spatial light modulator (Fig. 1a). This
mirror (M) and a diagnostics CCD camera (d-CCD). The fluorescence of
phase mask is calculated using the 3D Gerchberg–Saxton algorithm, the atoms in the traps at 780 nm is separated from the dipole trap beam
simplified for the case of point traps17. The beam is then focused with with a dichroic mirror (DM) and detected using an electron-multiplying
a high-numerical-aperture (0.5) aspheric lens under vacuum, creating CCD camera (EMCCD). For atom assembly we use moving tweezers
individual optical tweezers with a measured 1/e2 radius of about 1.1 μm superimposed on the trap beam with a polarizing beam splitter (PBS).
and a Rayleigh length of approximately 5 μm. After recollimation with This extra beam is deflected in the plane perpendicular to the beam
a second aspheric lens, the intensity of the trapping light is measured propagation with a 2D acousto-optical deflector (AOD), and its focus
using a standard charge-coupled device (CCD) camera. An electrically can be displaced axially by changing the focal length of an electrically
tunable lens (ETL1) in the imaging path allows us to acquire series of tunable lens (ETL3). The remaining electrically tunable lenses (ETL1
stack images along the optical axis z, from which we reconstruct the and ETL2) in the camera paths allow imaging of different planes along
z. The inset depicts the intensity distribution of the trap light forming a
full 3D intensity distribution. The imaging system covers a z-direction
bilayer array (red) and the action of the moving tweezers on an individual
scan range of 200 μm. atom (purple). b–d, Intensity reconstructions of exemplary 3D patterns
Figure 1b–d shows some examples of patterns suitable for experiments obtained from a collection of z-stack images taken with the diagnostics
with single atoms. The images are reconstructed using a maximum- CCD camera. The regions of maximum intensity form a trefoil knot (b),
intensity projection method18 from 200 z images obtained with the a 5 × 5 × 5 cubic array (c) and a C320 fullerene-like structure (d). The
diagnostics CCD camera. With about 3.5 mW of power per trap we dimensions, Lx, Ly, Lz, of the images are the same in all the examples.

1
Laboratoire Charles Fabry, Institut d’Optique Graduate School, CNRS, Université Paris-Saclay, Palaiseau, France. 2These authors contributed equally: Daniel Barredo, Vincent Lienhard, Sylvain de
Léséleuc. *e-mail: [email protected]

6 S E P TE M B ER 2 0 1 8 | V O L 5 6 1 | N A T U RE | 7 9
© 2018 Springer Nature Limited. All rights reserved.
 RESEARCH Letter

a Hyperboloid (90 sites) b Möbius strip (85 sites) is negligible. This distance can be further reduced to about 14 μm by
operating the moving tweezers with less power, without any degrada-
tion in the performance of the sorting process. In a complementary
experiment, where we fully assembled small arrays, we also checked
that the assembling efficiency is not affected by slight changes (below
about 3 μm) in the exact axial position of the moving tweezers.
We now demonstrate full loading of arbitrary 3D lattices using plane-
by-plane assembly. We start by creating a 3D trap array that can be
decomposed in several planes normal to z. In each plane we generate
c C84 fullerene-like (84 sites) d Cone (100 sites) approximately twice the number of traps that we need to load, so that
we can easily load enough atoms to assemble the target structure. The
sequence used to create fully loaded patterns (see Fig. 3b) starts by
loading the MOT and monitoring the atoms entering and leaving the
traps by sequentially taking a fluorescence picture for each plane. We
trigger the assembler as soon as there are enough atoms in each plane
to fully assemble it. We then freeze the loading by dispersing the MOT
cloud and record the initial positions of the atoms by another series of
z-stack images. Analysis of the images reveals which traps are filled with
e Torus (120 sites) f Eiffel tower (126 sites) single atoms. We use this information to compute (in about 1 ms) the
moves needed to create the fully loaded target array and perform plane-
by-plane assembly by changing the z position of the moving tweezers
after the assembly in each plane is completed. Finally, we detect the final
Ly = 87 μm

3D configuration with another series of z-stack images.

Figure 3c–h shows a gallery of fully loaded 3D atomic arrays arbi-
trarily arranged in space. We can create fully loaded 3D architectures
m with up to 72 atoms distributed in several layers with different degrees
8μ
Lx = 110 μm =1
1 of complexity. The selected structures include simple cubic lattices
Lz (Fig. 3d), bilayers with square or graphene-like22 arrangements (Fig. 3c,
Fig. 2 | Single-atom fluorescence in 3D arrays. a–f, Maximum-intensity- e, g), lattices with inherent geometrical frustration such as pyrochlore23
projection reconstruction of the average fluorescence of single atoms (Fig. 3f) and lattices with cylindrical symmetry (Fig. 3h), which are
loaded stochastically into exemplary arrays of traps. The x, y, z scan range suitable, for example, for studying quantum Hall physics with neutral
of the fluorescence (Lx, Ly, Lz) is the same for all the 3D reconstructions. atoms24. The arrays are not restricted to periodic arrangements, and the
positions of the atoms can be controlled with high accuracy (<1 μm).
optimization14. Rubidium-87 atoms are then loaded in the traps from The minimum interlayer separation that we can achieve depends on
a magneto-optical trap (MOT), with a final temperature of 25 μK. We the type of underlying geometry. This is illustrated in Fig. 3e, which
detect the occupancy of each trap by collecting the fluorescence of the shows the full 3D assembly of a bilayer square lattice (with a layer sep-
atoms at 780 nm with an electron-multiplying CCD camera for 50 ms. aration of dz = 5 μm). There, sites corresponding to the second layer
A second tunable lens (ETL2) in the imaging path is used to focus the are displaced by half the lattice spacing. Because traps belonging to
fluorescence of different atom planes. neighbouring layers do not have the same (x, y) coordinates, there is
In Fig. 2 we show the fluorescence of single atoms trapped in no limitation to the minimum interlayer distance that we can produce.
various complex 3D structures, some of which are relevant, for instance, to In both images we can observe a defocused fluorescence at intersite
the study of non-trivial properties of Chern insulators19–21. Each exam- positions due to atoms trapped in the neighbouring layer. By contrast,
ple is reconstructed from a series of 100 z-stack images covering an whenever traps are aligned along the z axis (for example, in Fig. 3d), we
axial range of about 120 μm. With no further action, these arrays are set a minimum axial separation of about 17 μm to avoid any disturbance
randomly loaded with a filling fraction of about 0.5; we thus average from the moving tweezers on the atoms while assembling neighbouring
the fluorescence signal over 300 frames to reveal the geometry of the planes. However, for some trapping geometries this constraint can be
structures. overcome by applying a small global rotation of the 3D trap pattern
For deterministic atom loading, we extend our two-dimensional around the x or y axis, so that neighbouring traps do not share the
(2D) atom-by-atom assembler11 to 3D geometries. For that, we super- same (x, y) coordinates. The minimum interlayer spacing ultimately
impose a second 850-nm laser beam (with 1/e2 radius of about 1.3 μm) depends on the Rayleigh range of our trapping beam (about 5 μm) and
on the trapping beam, which can be steered in the x–y plane using a could be further reduced, for example, by using an aspheric lens with
2D acousto-optical deflector and in the z direction by changing the higher numerical aperture. The range of interatomic distances that we
focal length of a third tunable lens (ETL3). Combined with a real-time can achieve (3– 40 μm) is suitable for implementing fast qubit gates25
control system, the moving tweezers can perform single-atom transport or simulating excitation transport26 and quantum magnetism with
with fidelities exceeding 0.993, as shown in ref. 11, and produce fully Rydberg atoms, because interaction energies between Rydberg states
loaded arrays by using independent and sequential rearrangement of at those distances are typically in the megahertz range.
the atoms for each of the np planes in the 3D structures. To illustrate this possibility, we performed a proof-of-principle
To explore the feasibility of plane-by-plane atom assembly, we first experiment with two atoms belonging to the cylindrical lattice dis-
determine the minimal separation between layers so that each tar- played in Fig. 3h. The atoms are separated by a total distance of
get plane can be reordered without affecting the others. To quantify R12 = 20 μm (dx = 10 μm, dz = 17 μm); see Fig. 4. We first initialize the
this, we perform the following experiment in a 2D array containing atoms in state |g〉 = |5S1/2, F = 2, mF = 2〉, where F and mF are the hyper-
46 traps. We randomly load the array with single atoms and demand fine and magnetic quantum numbers, respectively, by optical pumping
the atom assembler to remove all the atoms. We average over about in a 47-G magnetic field that defines the quantization axis and is
50 realizations and then repeat the experiment for different axial sep- aligned perpendicular to the internuclear axis. Then, the dipole trap is
arations between the position of the moving tweezers and the trap switched off and a two-photon Rydberg stimulated Raman adiabatic
plane. The result is shown in Fig. 3a, where we see that for separations passage27 excites both atoms to the |↑〉 = |60S1/2, mj = 1/2〉 Rydberg
beyond about 17 μm the effect of the moving tweezers on the atoms state, where mj is the spin projection along the magnetic field direction.

8 0 | N A T U RE | V O L 5 6 1 | 6 S E P TE M B ER 2 0 1 8
© 2018 Springer Nature Limited. All rights reserved.
 Letter RESEARCH

a b Initial images Final images

1.0 1 np 1 np
dz
0.8 EMCCD
Recapture probability

0.6
AL Atom ETL2
plane
0.4 1 np

0.2 2D assembler

0.0
–15 –10 –5 0 5 10 15 ETL3
Axial distance, dz (μm)
(50 ms × np) (60 ms × np) t

c AA-stacked square lattices d Cubic lattice

(dz = 40 μm; 72 atoms) (dz = 20 μm; 36 atoms)

Ly = 67 μm

Lx = 67 μm
e AB-stacked square lattices f Pyrochlore-like lattice

(dz = 5 μm; 32 atoms) (dz = 25 μm; 42 atoms)

g AB-stacked bilayer graphene h Cylindrical lattice

(dz = 40 μm; 49 atoms)
(dz = 17 μm; 42 atoms)

Fig. 3 | Fully loaded 3D arrays of single atoms. a, Recapture probability analysis of the resulting np images reveals the initial position of the atoms
as a function of the axial distance between the focus of the moving in the traps. The 2D atom assembler, in combination with an electrically
tweezers and the plane of the atoms measured experimentally by trying to tunable lens (ETL3), arranges the atoms plane by plane. Finally, a new set
remove all the atoms from a 46-trap array. Error bars denote the standard of sequential images is collected to capture the result of the 3D assembly.
error of the mean and are smaller than the symbol size. The line is a guide c–h, Fully loaded arrays with arbitrary geometries. All images are single
to the eye. b, Time control sequence of the experiment. We start the shots. The models of the 3D configurations are shown for clarity; the
experiment by recording sequentially an image for each target plane. The colours of the frames around the images encode successive atomic planes.

We further use a resonant microwave field and local addressing28 to as a function of the variable interaction time, T. Finally, a de-excitation
transfer the second atom to the |↓〉 = |60P1/2, mj = −1/2〉 state, while sequence projects the population in |↑〉 to |g〉, but leaves the population
the first atom remains in |↑〉. In these two Rydberg levels, the atoms are in |↓〉 unaffected. After switching the dipole trap on again, atoms in |g〉
coupled by a direct dipole–dipole interaction with a strength of are recaptured, while atoms in the excited state |↓〉 are repelled by the
3
U = C 3/R12 , and a calculated C3 coefficient of C3 = h×1,357 MHz μm3, trapping potential of the optical tweezers and appear as atom losses in
where h is the Planck constant. The prepared pair-state |↑↓〉 evolves the final fluorescence images. The outcome of this experiment is shown
3
under the XY-spin Hamiltonian H = (C 3/R12 ) (σ1+ σ2− + σ1− σ2+) in Fig. 4. We observe coherent ‘flip-flops’ between |↑↓〉 and |↓↑〉 with a
(where σi± denotes the Pauli matrices acting on atom i = {1, 2}) and measured frequency of 2U/h = 333 ± 5 kHz. This value is consistent
undergoes coherent spin-exchange oscillations between |↑↓〉 and |↓↑〉 with the frequency 2U/h = 339 kHz expected from our distance

6 S E P TE M B ER 2 0 1 8 | V O L 5 6 1 | N A T U RE | 8 1
© 2018 Springer Nature Limited. All rights reserved.
 RESEARCH Letter

1 5. Meschede, D. & Rauschenbeutel, A. Manipulating single atoms. Adv. At. Mol. Opt.
P↑↓ P↓↑ Phys. 53, 75–104 (2006).
6. Blatt, R. & Roos, C. F. Quantum simulations with trapped ions. Nat. Phys. 8,

Probability
277–284 (2012).
7. Gross, C. & Bloch, I. Quantum simulations with ultracold atoms in optical
lattices. Science 357, 995–1001 (2017).
8. Weiss, D. S. & Saffman, M. Quantum computing with neutral atoms. Phys. Today
y 70, 44–50 (2017).
0 9. Kuhr, S. Quantum-gas microscopes: a new tool for cold-atom quantum
x z 0 1 2 3 4 simulators. Natl Sci. Rev. 3, 170–172 (2016).
B Interaction time, T (μs) 10. Endres, M. et al. Atom-by-atom assembly of defect-free one-dimensional cold
atom arrays. Science 354, 1024–1027 (2016).
Fig. 4 | Spin-exchange dynamics between two Rydberg atoms in 11. Barredo, D., de Léséleuc, S., Lienhard, V., Lahaye, T. & Browaeys, A. An
different z layers. Excitation-hopping oscillations between |↑↓〉 and atom-by-atom assembler of defect-free arbitrary two-dimensional atomic
|↓↑〉, observed in the populations P↑↓, P↓↑, driven by the dipole–dipole arrays. Science 354, 1021–1023 (2016).
interaction between two Rydberg states, |↑〉 = |60S1/2, mj = 1/2〉 and 12. Nelson, K. D., Li, X. & Weiss, D. S. Imaging single atoms in a three-dimensional
|↓〉 = |60P1/2, mj = −1/2〉, at a distance of about 20 μm (dx = 10 μm; array. Nat. Phys. 3, 556–560 (2007).
13. Wang, Y., Zhang, X., Corcovilos, T. A., Kumar, A. & Weiss, D. S. Coherent
dy = 17 μm). Error bars represent the standard error of the mean and are addressing of individual neutral atoms in a 3D optical lattice. Phys. Rev. Lett.
mostly smaller than the symbol size. Solid lines are damped sine fits to the 115, 043003 (2015).
data. The direction of the magnetic field, B, is indicated. 14. Nogrette, F. et al. Single-atom trapping in holographic 2D arrays of microtraps
with arbitrary geometries. Phys. Rev. X 4, 021034 (2014).
15. Kim, H. et al. In situ single-atom array synthesis using dynamic holographic
calibration (R12 = 20 ± 1 μm), which was performed by optical means. optical tweezers. Nat. Commun. 7, 13317 (2016).
The finite contrast and the small damping of the oscillations arise from 16. Sturm, M. R., Schlosser, M., Walser, R. & Birkl, G. Quantum simulators by design:
many-body physics in reconfigurable arrays of tunnel-coupled traps. Phys. Rev.
experimental imperfections (errors in state preparation and readout, A 95, 063625 (2017).
residual atomic temperature), as reported in ref. 29. This proof-of- 17. Di Leonardo, R., Ianni, F. & Ruocco, G. Computer generation of optimal
principle experiment demonstrates the feasibility of performing holograms for optical trap arrays. Opt. Express 15, 1913–1922 (2007).
18. Wallis, J. W., Miller, T. R., Lerner, C. A. & Kleerup, E. C. Three-dimensional display
quantum simulations using our defect-free 3D atomic arrays of single in nuclear medicine. IEEE Trans. Med. Imaging 8, 297–303 (1989).
atoms. Excitations hopping under the influence of this Hamiltonian 19. Beugeling, W., Quelle, A. & Morais Smith, C. Nontrivial topological states on a
are equivalent to a system of hard-core bosons. The dipole–dipole Möbius band. Phys. Rev. B 89, 235112 (2014).
interactions observed here can be further exploited to engineer 20. Rüegg, A., Cof, S. & Moore, J. E. Corner states of topological fullerenes. Phys. Rev.
B 88, 155127 (2013).
Hamiltonians containing complex hopping amplitudes, which are 21. Ningyuan, J., Owens, C., Sommer, A., Schuster, D. & Simon, J. Time- and
suitable for the study of, for example, topological insulators30. site-resolved dynamics in a topological circuit. Phys. Rev. X 5, 021031 (2015).
Besides the unique tunability of the geometries that it provides, our 22. Castro Neto, A. H., Guinea, F., Peres, N. M. R., Novoselov, K. S. & Geim, A. K. The
electronic properties of graphene. Rev. Mod. Phys. 81, 109–162 (2009).
atom-assembling procedure is highly efficient: we reach typical filling 23. Bramwell, S. T. & Gringas, M. J. P. Spin ice state in frustrated magnetic pyroclore
fractions of 0.95. This measured efficiency is slightly dependent on the materials. Science 294, 1495–1501 (2001).
number of planes and is mainly limited by the lifetime of the atoms in 24. Łacki, M. et al. Quantum Hall physics with cold atoms in cylindrical optical
lattices. Phys. Rev. A 93, 013604 (2016).
the traps (about 10 s) and the duration of the sequence (we typically need 25. Saffman, M., Walker, T. G. & Mølmer, K. Quantum information with Rydberg
60 ms per plane to acquire the fluorescence images and about 50 ms per atoms. Rev. Mod. Phys. 82, 2313–2363 (2010).
plane to perform atom sorting). The repetition rate of the experiment 26. Browaeys, A., Barredo, D. & Lahaye, T. Experimental investigations of dipole–
dipole interactions between a few Rydberg atoms. J. Phys. B 49, 152001 (2016).
is about 1 Hz. The number of traps and the filling fraction of the arrays 27. Vitanov, N. V., Rangelov, A. A., Shore, B. W. & Bergmann, K. Stimulated Raman
could be further increased with current technology: (i) the volume of adiabatic passage in physics, chemistry, and beyond. Rev. Mod. Phys. 89,
the trap array and the maximum number of traps can be extended by 015006 (2017).
28. de Léséleuc, S., Barredo, D., Lienhard, V., Browaeys, A. & Lahaye, T. Optical
increasing the field of view of the aspheric lens and the laser power; control of the resonant dipole–dipole interaction between Rydberg atoms. Phys.
(ii) the lifetime of the atoms in the traps can realistically be increased Rev. Lett. 119, 053202 (2017).
by an order of magnitude; (iii) the repetition rate of the experiment 29. Barredo, D. et al. Coherent excitation transfer in a spin chain of three Rydberg
can be increased by optimizing the atom assembler11, in particular by atoms. Phys. Rev. Lett. 114, 113002 (2015).
30. Weber, S. et al. Topologically protected edge states in small Rydberg systems.
transferring atoms also between different planes31; and (iv) the initial Quantum Sci. Technol. 3, 044001 (2018).
filling fraction of the arrays could reach values exceeding 0.8 by using 31. Lee, W., Kim, H. & Ahn, J. Three-dimensional rearrangement of single atoms
tailored light-assisted collisions32,33. Therefore, the generation of three- using actively controlled optical microtraps. Opt. Express 24, 9816–9825 (2016).
32. Grünzweig, T., Hilliard, A., McGoven, M. & Andersen, M. Near-deterministic
dimensional structures containing several hundred atoms at unit filling preparation of a single atom in an optical microtrap. Nat. Phys. 6, 951–954 (2010).
seems within reach, opening up many new possibilities in quantum 33. Lester, B. J., Luick, N., Kaufman, A. M., Reynolds, C. M. & Regal, C. A. Rapid
information processing and quantum simulation with neutral atoms. production of uniformally filled arrays of neutral atoms. Phys. Rev. Lett. 115,
073003 (2015).

Data availability Acknowledgements We thank A. Läuchli for discussions. This work benefited
The data presented in the figures and that support the other findings of this study from financial support by the EU (H2020 FET-PROACT Project RySQ), by the
are available from the corresponding author on reasonable request. ‘PALM’ Labex (projects QUANTICA and XYLOS) and by the Région Île-de-
France in the framework of DIM Nano-K.
Online content Reviewer information Nature thanks W. Bakr, N. Lundblad and the other
Any methods, additional references, Nature Research reporting summaries, source anonymous reviewer(s) for their contribution to the peer review of this work.
data, statements of data availability and associated accession codes are available at
https://doi.org/10.1038/s41586-018-0450-2 Author contributions D.B., V.L. and S.d.L. performed the experiments. T.L. and
A.B. supervised the work. All authors made critical contributions to the work,
Received: 7 December 2017; Accepted: 16 July 2018; discussed the results and contributed to the writing of the manuscript.
Published online 5 September 2018. Competing interests The authors declare no competing interests.

1. Devoret, M. H. & Schoelkopf, R. J. Superconducting circuits for quantum Additional information

information: an outlook. Science 339, 1169–1174 (2013). Supplementary information is available for this paper at https://doi.org/
2. Veldhorst, M. et al. A two-qubit logic gate in silicon. Nature 526, 410–414 10.1038/s41586-018-0450-2.
(2015). Reprints and permissions information is available at http://www.nature.com/
3. Kok, P. et al. Linear optical quantum computing with photonic qubits. Rev. Mod. reprints.
Phys. 79, 135–174 (2007). Correspondence and requests for materials should be addressed to D.B.
4. Buluta, S. A. I. & Nori, F. Natural and artificial atoms for quantum computation. Publisher’s note: Springer Nature remains neutral with regard to jurisdictional
Rep. Prog. Phys. 74, 104401 (2011). claims in published maps and institutional affiliations.

8 2 | N A T U RE | V O L 5 6 1 | 6 S E P TE M B ER 2 0 1 8
© 2018 Springer Nature Limited. All rights reserved.