Spin models are the prime example of simplified manybody Hamiltonians used to model complex, real-world strongly correlated materials 1 . However, despite their simplified character, their dynamics often cannot be simulated exactly on classical computers as soon as the number of particles exceeds a few tens. For this reason, the quantum simulation 2 of spin Hamiltonians using the tools of atomic and molecular physics has become very active over the last years, using ultracold atoms 3 or molecules 4 in optical lattices, or trapped ions 5 . All of these approaches have their own assets, but also limitations. Here, we report on a novel platform for the study of spin systems, using individual atoms trapped in two-dimensional arrays of optical microtraps with arbitrary geometries, where filling fractions range from 60 to 100% with exact knowledge of the initial configuration. When excited to Rydberg D-states, the atoms undergo strong interactions whose anisotropic character opens exciting prospects for simulating exotic matter 6 . We illustrate the versatility of our system by studying the dynamics of an Ising-like spin-1/2 system in a transverse field with up to thirty spins, for a variety of geometries in one and two dimensions, and for a wide range of interaction strengths. For geometries where the anisotropy is expected to have small effects we find an excellent agreement with ab-initio simulations of the spin-1/2 system, while for strongly anisotropic situations the multilevel structure of the D-states has a measurable influence 7,8 . Our findings establish arrays of single Rydberg atoms as a versatile platform for the study of quantum magnetism.Rydberg atoms have recently attracted a lot of interest for quantum information processing 9 and quantum simulation 10 . In this work, we use a system of individual Rydberg atoms to realize highly-tunable artificial quantum Ising magnets. By shining on the atoms lasers that are resonant with the transition between the ground state |g and a chosen Rydberg state |r , we implement the Ising-like Hamiltonianwhich acts on the pseudo-spin states |↓ i and |↑ i corresponding to states |g and |r of atom i, respectively. Here, Ω is the Rabi frequency of the laser coupling, the σ i α (α = x, y, z) are the Pauli matrices acting on atom i, and n i = (1 + σ i z )/2 is the number of Rydberg excitations (0 or 1) on site i. FIG. 1|: Experimental platform. a: An array of microtraps is created by imprinting an appropriate phase on a dipole-trap beam. Siteresolved fluorescence of the atoms, at 780 nm, is imaged on a camera using a dichroic mirror (DM). Rydberg excitation beams at 795 and 475 nm are shone onto the atoms. The inset shows the measured light intensity for an array of Nt = 19 traps. b: Sketch of an experimental sequence. During loading, the camera images are analyzed continuously to extract the number of loaded traps. As soon as a triggering criterion is met, the loading is stopped and an image of the initial configuration is acquired. After Rydberg excitation, a final image is ...

Large arrays of individually controlled atoms trapped in optical tweezers are a very promising platform for quantum engineering applications. However, to date, only disordered arrays have been demonstrated, due to the non-deterministic loading of the traps. Here, we demonstrate the preparation of fully loaded, two-dimensional arrays of up to ∼ 50 microtraps each containing a single atom, and arranged in arbitrary geometries. Starting from initially larger, half-filled matrices of randomly loaded traps, we obtain user-defined target arrays at unit filling. This is achieved with a real-time control system and a moving optical tweezers that performs a sequence of rapid atom moves depending on the initial distribution of the atoms in the arrays. These results open exciting prospects for quantum engineering with neutral atoms in tunable geometries.The last decade has seen tremendous progress over the control of individual quantum objects [1, 2]. Many experimental platforms, from trapped ions [3] to superconducting qubits [4], are actively explored. The current challenge is now to extend these results towards large assemblies of such objects, while keeping the same degree of control, in view of applications in quantum information processing [5], quantum metrology [6], or quantum simulation [7]. Neutral atoms offer some advantages over other systems for these tasks. Besides being well isolated from the environment and having tunable interactions, systems of cold atoms hold the promise of being scalable to hundreds of individually controlled qubits. Control of the atomic positions at the single-particle level can been achieved with optical potentials. In a 'top-down' approach using optical lattices and quantum gas microscopes, hundreds of traps can now be created and addressed individually [8]. By making use of the superfluid to Mott-insulator transition, single atom filling fractions exceeding 90% are achieved [9], albeit at the expense of relatively long experimental duty cycles and constraints in the lattice geometries.Single atoms can also be trapped in 2d arrays of microscopic optical tweezers with single-site resolution using holographic methods [10][11][12]. This bottom-up approach offers faster preparation and a higher degree of tunability of the underlying geometry. However, achieving unit filling of the arrays is hampered by the stochastic nature of the loading and has remained so far elusive. Although proof-of-principle demonstrations of quantum gates [13] and quantum simulations [14] using this latter platform have been reported [15], this non-deterministic loading poses a serious limitation for applications where large-scale ordered arrays are required. To solve this problem, several approaches have been considered, exploiting the Rydberg blockade mechanism [16], or using tailored light-assisted collisions [17]. To date, despite those efforts, loading efficiencies of around 90% at best for a single atom in a single tweezers could be achieved [18,19], making the probabilities to fully load large arrays still...