In neutral atom quantum computers, readout and preparation of the atomic qubits are usually based on fluorescence imaging and subsequent analysis of the acquired image. For each atom site, the brightness or some comparable metric is estimated and used to predict the presence or absence of an atom. Across different setups, we can see a vast number of different approaches used to analyze these images. Often, the choice of detection algorithm is either not mentioned at all or it is not justified. We investigate several different algorithms and compare their performance in terms of both precision and execution run time. To do so, we rely on a set of synthetic images across different simulated exposure times with known occupancy states, which we generated using a previously validated imaging simulation. Since the use of simulation provides us with the ground truth of atom site occupancy, we can easily state precise error rates and variances of the reconstructed property. However, knowing the relative performance of these algorithms is not sufficient to justify their use, since better ones can exist that were not compared. To investigate this possibility, we calculated the Cramer-Rao bound in order to establish an upper limit that even a perfect estimator cannot outperform. As the metric of choice, we used the number of photonelectrons that can be contributed to a specific atom site. Every estimator that reconstructs a different property can simply be scaled accordingly. Since the bound depends on the occupancy of neighboring sites, we provide the best and worst cases, as well as a half filled one, which should represent an averaged bound best. Our comparison shows that of our tested algorithms, a global nonlinear least-squares solver that uses the optical system's point spread function (PSF) to return a global bias and each sites' number of photoelectrons performed the best, on average crossing the worst-case bound for longer exposure times. Its main drawback is its huge computational complexity and, thus, required calculation time. We manage to somewhat reduce this problem, suggesting that its use may be viable, leading us to a novel group of algorithms that present a compromise between speed and precision. However, our study also shows that for cases where utmost speed is required, simple algorithms, like summing up pixel values around atom sites, may be preferable.
In recent years, neutral atom quantum computers (NAQCs) have attracted a lot of attention, primarily due to their long coherence times and good scalability. One of their main drawbacks is their comparatively time-consuming control overhead, with one of the main contributing procedures being the detection of individual atoms and measurement of their states, each occurring at least once per compute cycle and requiring fluorescence imaging and subsequent image analysis. To reduce the required time budget, we propose a highly-parallel atom-detection accelerator for tweezer-based NAQCs. Building on an existing solution, our design combines algorithm-level optimization with a field-programmable gate array (FPGA) implementation to maximize parallelism and reduce the run time of the image analysis process. Our design can analyze a 256$\times
256−pixelimagerepresentinga10\times$10 atom array in just 115
μs on a Xilinx UltraScale+ FPGA. Compared to the original CPU baseline and our optimized CPU version, we achieve about 34.9
× and 6.3
× speedup of the reconstruction time, respectively. Moreover, this work also contributes to the ongoing efforts toward fully integrated FPGA-based control systems for NAQCs.
Imagine a new generation of computers that dramatically surpasses the most powerful supercomputers, being able to tackle the most challenging of computational problems. These problems arise, for example, in quantum physics where the computational power required to model a many-particle system increases exponentially with the number of its constituents, and as a consequence the calculations become prohibitively time-consuming. Nobel laureate Richard Feynman proposed this new type of computer in the early 1980s: ‘Let the computer itself be built of quantum mechanical elements which obey quantum mechanical laws.’ Such a quantum simulator would allow researchers to tackle challenging problems in condensed-matter physics. In particular many-body fermionic quantum systems are very difficult to compute due to the antisymmetric nature of the wave function, and the resulting sign problem for quantum Monte Carlo methods. This is why a quantum simulator will be an excellent test bed to investigate phenomena and properties of strongly correlated fermionic quantum systems. A powerful experimental platform to realize a quantum simulator is laser-cooled atoms [1] held in an artificial crystal of light, generated by superimposing laser beams. The atoms are kept in this optical lattice similar to marbles in the hollows of an egg carton, and they can mimic the behaviour of electrons in a real crystal. These systems interface atomic physics with condensed matter physics in a new way, as the tools of atomic physics allow for exquisite control of many experimental parameters, which are often inaccessible in solid state experiments. In-situ imaging techniques [2] using ‘quantum-gas microscopes’ constitute a new approach to the study of many-body quantum systems in an optical lattice with unprecedented resolution at the single-particle and single-site level [3,4]. This new method is complementary to earlier optical lattice experiments measuring distributions in momentum space after a time-of-flight expansion. A quantum-gas microscope uses an optical imaging system (Fig. 1a) for collecting the fluorescence light of ultracold atoms in an optical lattice with high spatial resolution comparable to the lattice spacing of ≈ 500 nm. (a) Atoms in optical lattice atoms are imaged with a high-resolution microscope objective [4]. (b) Single-site addressing is achieved by focussing a laser beam through the microscope onto selected atoms, bringing them into resonance with microwave radiation which produces spin flips [12]. (c) and (d) Single-atom-resolved fluorescence images of bosonic Mottinsulators [4,9]. (e) Deterministically prepared atom distribution using single-site addressing [12]. (f) Direct observation of single-atom quantum walks [15]. Fluorescence imaging of atoms in a quantum-gas microscope has made it possible to directly observe bosonic Mott insulators with single-atom resolution [4–6], giving access to in-situ measurements of temperature and entropy distributions. Subsequently, ground-breaking experiments in many-body quantum dynamics have been performed, such as the first direct observation of correlation spreading after quenches of the system to lower lattice depths [7]. Other experiments include the quantum simulation of an antiferromagnetic spin-1/2 chain [8], the realization of algorithmic cooling [9], the direct measurement of correlations across the superfluid-to-Mott-insulator transition [10], or the observation of the ‘Higgs’ amplitude mode [11]. The high-resolution imaging system is also capable of optically addressing individual atoms with sub-diffraction-limited resolution. For this purpose, a laser beam is focussed though the microscope objective onto selected lattice sites, bringing the atoms into resonance with microwave radiation which produces spin flips [12,13]. The preparation of arbitrary atom distributions in the lattice not only opens perspectives for neutral-atom quantum information processing, but also allows the deterministic creation of specific out-of-equilibrium situations, and their ensuing system dynamics can be studied with single-atom resolution. Examples include the creation of spin impurities and the study of their dynamics in a many-body system [14], the observation of quantum walks [15] and magnon-bound states [16], and the measurement of entanglement entropy [17]. The technique used for single-site-resolved detection of individual atoms in the optical lattice is fluorescence imaging. It requires laser cooling of the atoms to sub-Doppler temperatures while detecting the fluorescence photons emitted during this process. A challenge is to achieve a low particle loss rate and to prevent the atoms from hopping between lattice sites. In the case of the first quantum-gas microscopes, this has been implemented with bosonic rubidium-87 atoms, and also recently for ytterbium-174. However, it has proven more challenging for fermionic alkaline species such as potassium-40 and lithium-6, as their low mass and small excited state hyperfine splitting-prevented standard sub-Doppler cooling from working. As a result, it took several years to extend the quantum-gas microscope technique to fermionic species, and in 2015 several groups have demonstrated single-atom-resolved imaging of fermions [18]. In these setups, the direct observation of band insulators [19], fermionic Mott insulators [20,21] and even antiferromagnetic correlations [22] has been reported most recently. These developments provide fascinating opportunities to study many-body fermionic quantum systems with single-particle access. The fermionic nature of the electron is vital to understanding a range of phenomena within solid-state physics, such as electron pairing in superconductivity, and quantum magnetism. In this context, the quantum simulator will allow, e.g. for direct measurement of ordered quantum phases and out-of-equilibrium dynamics, with access to quantities such as spin--spin correlation functions and the growth of many-body entanglement. The high-resolution imaging systems could also be used to project tailored time-varying light fields onto the atoms, allowing the removal of high-entropy regions by locally modifying the confining potential to attain very low-entropy states. Attaining these low-entropy many-particle states is one of the key challenges in our field. In summary, quantum-gas microscopes offer unprecedented access to many-body quantum systems, and they will allow us to gain an even deeper insight into the intriguing world of quantum physics that governs the properties of strongly correlated materials. Using a well-controlled quantum system to simulate other quantum systems could help resolve some of the most challenging problems scientists face in condensed matter physics and materials science, but it will also provide new insights in quantum chemistry, high-energy physics and biology. SK acknowledges funding from the EU (ERC FERMILATT, SIQS, QuProCs) and SUPA.
The power of quantum computation derives from algorithmic methods that exploit the availability of quantum superposition and entanglement to perform computations that are intractable with classical devices. The race is on to develop hardware that will unleash the promise of quantum algorithms. A handful of different types of hardware are currently being developed with the greatest efforts directed at superconducting, quantum-dot, trapped-ion, photonic, and neutral-atom approaches [1]. While all approaches have strengths and weaknesses, and are at different stages of development, the challenge of creating a practical design that can be scaled to a million or more qubits has not yet been met with any of the existing platforms. One may wonder if we really need a million qubits for a useful quantum computer. The answer depends on our ambition level. On the one hand it is widely believed that a quantum computer with less than 100 reliable qubits will be able to demonstrate a quantum advantage for a well defined computational problem [2]. Demonstration of a quantum computational advantage will generate great excitement in the halls of academia, but will have little broader import. A problem of great practical interest such as ab initio design of chemicals for increasing the yield of fertilizers for crop production, to name but one example, may also require just a few hundred qubits, provided computational sequences with circuit depth of 1010 or more operations can be reliably performed [3]. However, given the current performance of qubit hardware, which has demonstrated error rates not less than ∼10−3, computations with depth 1010 will only produce random outcomes of negligible value. The solution lies in the construction of error-correction circuitry operating on logical qubits, each composed of many physical qubits, which will enable logical error rates many orders of magnitude smaller than physical error rates. Although error correction is remarkably efficient for classical computing devices—Hamming codes detect and correct errors while only requiring roughly 10% overhead in the number of physical bits—quantum error correction is notoriously expensive. New approaches to error correction are the subject of intensive research with one promising direction being the development of codes that are optimized to suppress the dominant error mechanism in a particular physical platform [4,5]. Nevertheless, with physical error rates of 10−4, a plausible target for current qubit systems, the overhead required to reach logical error rates of 10−10 may be measured in thousands of physical qubits for each logical qubit. Thus a quantum computer able to perform deep calculations on a few hundred logical qubits could require a million physical qubits. A consequence of the cost of quantum error correction is that any viable approach to large-scale quantum computing needs to combine high-fidelity quantum logic operations with a capability for integrating large numbers of physical qubits. From this perspective neutral-atom qubits appear particularly promising [6] as they have already demonstrated control of more qubits than any other platform with recent experiments in 1D [7], 2D [8,9], and 3D [10] geometries showing control of up to 50 atomic qubits. Large qubit arrays can potentially be prepared in 1D, 2D, or 3D geometries. The number of proximal qubits increases with the dimensionality, which is advantageous for implementation of error correction code words, and points towards a preference for 2D or 3D arrays. On the other hand, the added overhead of compensating crosstalk incurred when targeting atoms inside a 3D volume makes parallel operations on multiple qubits challenging. A 2D geometry, as shown schematically in Fig. 1, may be an optimal choice since qubit control and measurement can be performed with lasers propagating normal to the plane of the array so that individual qubit sites can be addressed independently. With demonstrated array periods of a few micrometers we require a footprint of less than 1 mm2 for 104 qubits. Conceptual picture of a 2D neutral-atom qubit array. a) High-numerical-aperture lenses project several different wavelengths of light for control and measurement of the qubits. Specific choices of optical wavelengths depend on the atomic species. Not shown are optics for initial laser cooling, transport to the array, beam scanning, and imaging for qubit measurement. b) Averaged fluorescence image of 49 atomic qubits on a 2D grid with |
3.8μm| spacing [8]. From a fundamental perspective the prospects for scaling neutral-atom systems are particularly promising due to the exceptionally large ratio between coherent and incoherent coupling rates. Neutral-atom qubits are encoded in hyperfine ground states for which coherence times of the order of 10 s have been demonstrated [10]. This implies a dephasing rate |
γ∼0.1s−1|. The coherent coupling that enables quantum logic operations is switched on by exciting atoms to high-lying Rydberg states with principal quantum number close to 100 [11,12]. Rydberg excited atoms at spacings of a few micrometers in an optical lattice interact coherently with rates |
g>109s−1|. The ratio between desired coherent coupling and residual incoherent dephasing of g/γ ∼ 1010 establishes a figure of merit that is enabling for scalability. To our knowledge only trapped ions can claim a similar g/γ ratio [13], but without a correspondingly direct path towards controlling thousands of qubits in a single processing unit. Although the g/γ ratio is exceptionally favorable for neutral-atom qubits a complete set of universal gate operations has not yet been demonstrated with high fidelity. Single-qubit gates in large 2D and 3D arrays have reached fidelity |
F∼0.999| [8,10]. However, the highest demonstrated fidelity for two-qubit entanglement mediated by Rydberg-state interactions is |
F<0.8| [6,14,15], a value that is well short of the theoretical limit of |
F>0.9999| [16,17]. There is healthy optimism within the research community working with neutral-atom qubits that combining improved laser sources with reduced noise, better cooling to reduce motional dephasing, and control of stray electric fields that perturb Rydberg atoms will lead to substantially improved gate fidelity in the near future. Recent results with better filtering of laser phase noise have already demonstrated remarkable improvement in ground–Rydberg entanglement [18]. The outlook for the next few years appears very promising. We anticipate that with advances in lasers, optical lattices for atom trapping, deterministic rearrangement and loading of atoms in multi-dimensional arrays [19,20], and optics for fast beam scanning it will be possible to increase the number of physical qubits in a single neutral-atom module to at least 104. A challenging yet plausible path to a million-qubit processor can then be contemplated by connecting 100 such modules with atom–photon entanglement channels [21]. Beyond the potential for advancing the state of the art in computation, development of techniques for single-atom control inside large multi-atom arrays will enable other applications of quantum science including metrology and time-keeping. Rydberg interactions that are at the heart of a neutral-atom quantum computer may lead to multi-atom entanglement and Heisenberg scaling of atomic-clock precision [22]. This work was supported by the NSF and ARL CDQI.
Neutral-atom quantum devices offer a natural and powerful platform for Quantum Reservoir Computing (QRC), a non-variational approach to quantum-enhanced machine learning that is attracting growing interest in the Noisy Intermediate-Scale Quantum (NISQ) era. In this work, we implement a QRC-based learning model tailored to Pasqal neutral-atom hardware, where classical data are encoded into time-dependent detuning waveforms and processed through programmable many-body dynamics. Feature embeddings are extracted from single- and two-qubit Pauli-
Z observables and used within a simple linear readout model. We systematically investigate both global and local detuning encoding strategies on a supervised image classification task, benchmarking performance against a purely classical linear baseline. Our results show that processing the data through the quantum reservoir yields a consistent $\sim$4\% improvement in accuracy, highlighting the system effectiveness as a nonlinear feature map. Both encoding strategies exhibit comparable performance, suggesting that QRC can leverage different control schemes while remaining compatible with near-term and emerging hardware capabilities. These findings provide concrete evidence that QRC can enhance classical machine learning pipelines by injecting non-classical correlations without the need for variational optimization, thereby avoiding issues such as barren plateaus and training instabilities. More broadly, this work establishes neutral-atom platforms as a promising and scalable setting for QRC, opening the way toward practical quantum-enhanced learning on near-term devices.
In recent years, neutral atom quantum computers (NAQCs) have attracted a lot of attention, primarily due to their long coherence times and good scalability. One of their main drawbacks is their comparatively time-consuming control overhead, with one of the main contributing procedures being the detection of individual atoms and measurement of their states, each occurring at least once per compute cycle and requiring fluorescence imaging and subsequent image analysis.To reduce the required time budget, we propose a highly-parallel atom-detection accelerator for tweezer-based NAQCs. Building on an existing solution, our design combines algorithm-level optimization with a field-programmable gate array (FPGA) implementation to maximize parallelism and reduce the run time of the image analysis process. Our design can analyze a 256×256-pixel image representing a 10×10 atom array in just 115 μs on a Xilinx UltraScale+ FPGA. Compared to the original CPU baseline and our optimized CPU version, we achieve about 34.9× and 6.3× speedup of the reconstruction time, respectively. Moreover, this work also contributes to the ongoing efforts toward fully integrated FPGA-based control systems for NAQCs.