Beginner’s Survey of Quantum Hardware:

A comparison of superconducting, trapped ion, and neutral atom quantum computers.

Introduction

Quantum computing is rapidly evolving, with three architectures emerging as the leading contenders: superconducting qubits, trapped ions, and neutral atoms. Superconducting qubits are the most mature, backed by significant funding and extensive development, making them the most proven to date. Trapped ions, while offering the highest fidelity and connectivity, face challenges in scaling to larger systems. Neutral atom quantum computers, the youngest of the three architectures, show promising potential for scalability, positioning them as a strong candidate for future breakthroughs. These platforms represent the forefront of quantum computing research and development.

This paper focuses on critical performance factors that currently limit quantum computers from achieving real-world applicability: fault tolerance—encompassing noise and decoherence—and scalability. Overcoming these challenges is crucial for transitioning beyond the NISQ (Noisy Intermediate-Scale Quantum) era into the realm of practical, large-scale quantum computing.

We begin by examining the high-level physical implementations of these technologies, including the mechanisms behind their basic building blocks and the operation of single- and two-qubit quantum gates. We then compare the trade-offs across the three architectures, emphasizing the sources and solutions to noise and decoherence and scalability. For a more complete picture, we then follow with a discussion of the overarching strengths and limitations of each system, highlighting their potential trajectories in the quest for practical quantum computing.

Basic Building Blocks

Basics of Superconductors

Superconducting quantum computers rely on tiny electrical circuits that exploit the quantum properties of superconductors - materials that exhibit zero electrical resistance and create magnetic fields when cooled below a critical temperature. The building blocks of these systems are superconducting qubits, which are modeled as nonlinear oscillators and are typically formed by two components: the LC Circuit and the Josephson junction.

An LC circuit, composed of inductors (L) and capacitors (C), acts as a quantum harmonic oscillator with evenly spaced spaced energy levels. However, the even spacing prevents the differentiation of the computational states |0⟩ and |1⟩. To overcome this limitation, a Josephson junction is introduced. [2] Introducing nonlinearity to the system, the Josephson alters the harmonic potential into an anharmonic one, creating unevenly spaced energy levels. The two lowest energy levels are then used as the basis of the superconducting qubit. [2]

The Josephson junction consists of a thin insulating barrier sandwiched between two superconducting layers. This structure enables the quantum tunneling of Cooper pairs—pairs of electrons that move without resistance through the lattice. These junctions behave as artificial atoms with energy levels controlled by the capacitances, inductances, the junction barrier, and resonators, which serves as a readout circuit. Unlike qubits represented by single particles (such as atoms, electrons, or photons), superconducting qubits are macroscopic systems, a key differentiator from other quantum computing platforms. [5] The macroscopic nature leads to inherent behavior variability in the superconducting qubits arising from fabrication. Calibrations become necessary to address the variability and achieve the highest attainable uniformity. [3]

To improve robustness to noise, we can construct a transmon qubit by shunting the Josephson Junction with a large capacitor. This reduces the capacitance contribution of the junction relative to the capacitance of the shunt, allowing the Josephson energy to dominate the capacitance energy. [2] The Josephson energy is the energy stored in a Josephson junction due the phase difference between the superconducting wave functions on either side of the junction. The capacitance energy is the sum of the capacitance of the junction and the shunt. Doing so minimizes charge sensitivity and increases robustness against charge noise while improving coherence time. [2] Superconducting quantum computers can incorporate various qubits designs, each with their own tradeoffs. Examples include symmetric and asymmetric transmons, C-Shunted Flux Qubit, Fluxonium Qubit, [eng guid to supercond] and bosonic qubits [5].

Superconductor quantum implementations introduce tunability to qubits with a dc Superconducting Quantum Interference Device (dc-SQUID). This is done by replacing the single Josephson junction with two parallel Josephson junctions in a loop. [2] The external magnetic flux through the SQUID loop allows us to tune the Josephson energy and in turn the qubit frequency. dc-SQUID achieves tunability at the cost of introducing random fluctuations in magnetic flux (flux noise) that affects qubit stability. The asymmetric transmon design is used to reduce flux noise sensitivity while preserving tunability. [2] The tunability enables quantum gates and minimizing qubit interactions.

Basics of Trapped Ion

Trapped ion quantum computers have a vacuum chamber with small micro-fabricated chips embedded with electrodes as it’s main components. The electrodes generate a static electric field known as the ion trap. When the static electric field is combined with a static magnetic field or an oscillating electric field, a Penning trap and a Paul trap is formed, respectively. The Paul trap is the more common of the two and it uses a time-dependent electric field to force ions toward the trap center, minimizing their electric potential.[3] The ions are arranged into linear chains or 2-D arrangements. Interestingly, completed trapped ion quantum computers can fit into two 19-inch industry racks [4].

The process of trapped-ion implementations begin by generating neutral atomic vapor of one species of neutral atoms. Multiple species are also possible, but significantly increases complexity.[3] These neutral atoms are ionized by absorbing laser beams. Typically two laser are used. The trapped ions are then cooled, first with lasers that reduce its kinetic energy and velocity via Doppler Cooling, followed by a sub-Doppler cooling method such as Raman sideband cooling or sympathetic cooling. The cooling step is very time-intensive and significantly limits the runtime of the quantum computer.[3] After cooling, the ions are initialized to specific states with optical pumping, a technique that uses light with specific polarizations to repeatedly excite the ions to higher states and allow them to decay back to a desired lower state via spontaneous emissions until the ion is no longer reactive.

Ions are confined in the linear chains along the trap axis by balancing the Coulomb repulsion, pushes ions apart, and the electric potential created by the endcaps, pulls the ions together. Managing this balance allows the computer to control distance between ions in the chain. The trapped ions naturally oscillate in their equilibrium positions and the oscillatory motion is described by vibrational modes. These vibrational modes allow for ions to be coupled over long distances which is important for multi-qubit gate operations.[3] The coupling via vibrational modes also gives trapped ion quantum computers all-to-all connectivity, a unique feature not shared by superconductor nor neutral atom implementations.

The vibrational mode and the electronic state of the ions define the qubit’s quantum states (e.g., |0>, |1>, etc). State transitions are controlled by monochromatic laser pulses at particular polarizations. There are use four qubit variants that can be implemented by trapped ion quantum computers: Zeeman, Hyperfine, Optical, and Fine-structure.[3] Fine-structure qubits are possible, but not used in practice. We will briefly examine some differences when considering the single-qubit gates, but an in-depth comparison of the variants is beyond the scope of this paper.

The qubit’s state is measured by electron shelving, a state-dependent fluorescence technique. A laser transitions the qubit between |0> and a short-lived auxiliary state |a>. As the ion decays from |a> back to |0>, it will scatter the laser photons allowing for detection of a “bright state“. If the ion is in |1>, the detector will register no signals and find a “dark state.” Mid-circuit measurements are also possible by relocating ions to separate measurement zones or by using different ion species with distinct transition frequencies. [3]

Basics of Neutral Atom

Neutral atom quantum computers typically use atoms from the first two families of the periodic table because of their favorable electronic structure for cooling and trapping.[1] The atoms are first cooled down using a multi-stage cooling process which includes a stochastic Doppler cooling with three lasers oriented to target atoms in any arbitrary direction. A magnetic field then applies a velocity- and space-dependent force to keep atoms from drifting out of the cooling area. After cooling, lasers are used to move and keep the atoms in placed at fixed locations. The lasers that “trap” the atoms in place are the optical traps.

The lasers induce a small separation of positive and negative charges in the atom called a dipole moment. Dipole moments have a dipole force that is used to “trap” the atom in specific locations. Readers interested in the theoretical framework explaining the interaction between dipole forces, atom attraction, and light can refer to [10] as deeper details are beyond the scope of this paper. Arrays of optical traps, often called “optical tweezers,” are formed by arranging using highly focused laser to arrange the atoms in arbitrary two or three dimensional geometric arrangements. Current optical traps successfully trap about 50-60% leaving unfilled spots in the array.[11] The traps are placed with enough distance between them to allow for laser beams to execute single-qubit gate operations on single atoms.[1]

To complete the geometric arrangements, the filled atoms must be identified. This can be done by shining a light with a frequency causing fluorescence. A charge-coupled camera then catches an image, which shows bright spots for present atoms and dark spots for empty traps.[1] Acousto optical deflectors (AODs) then generate mobile traps to move the filled atoms into the desired arrangements, filling gaps or adjusting the arrangements as needed. The quantum gates are then applied and the results are measured.[1]

The measurement happens by applying a light with an energy level that will kick the atom in a |1> state out of the trap, but not an atom in a |0> state, leaving atoms in the |1> state as dark spots and atoms in the |0> state as bright spots. Interestingly, while this method appears destructive, mid-circuit measurements strategies have been developed such as electromagnetically induced transparency or transporting atoms to isolated imaging zones.[9] In case the reader is interested in the details of such strategies, [9] goes into depth on a strategy for single alkali species.

Gate Construction

Superconductor Based Gates

Superconductor quantum computers implement single-qubit gates by capacitative coupling a microwave line to the qubit. Specifically, an arbitrary waveform generators (AWG) parameterizes and outputs the “in-phase” (I) and “quadrature” (Q) signals which the IQ mixer use to modulate the microwave carrier signal. The IQ Mixer then generates microwaves at different pulse with specific shapes, amplitudes, and phases, determining a qubit’s evolution on the Bloch sphere.[2] The amplitude controls the angle of rotation and the phase controls the axis of gate rotation. The microwave drive line delivers the signal to a capacitor that allows the electric field from the microwave signal to drives the qubit. Higher capacitance drives a stronger signal and lower capacitance drives a weaker signal. By controlling these two parameters, superconductor quantum computers can execute arbitrary single qubit rotations allowing for any single-qubit gate.[2]

While various methods exist for implementing the two-qubit quantum gate, tunable coupling implementation have emerged as the dominant approach in recent platforms. Google’s Sycamore, the Chinese counterparts, and IBM’s 2023 Heron all use tunable couplers. [5] This implementation places a coupling circuit between two qubits to mediate their interaction. The coupling circuit can be tuned by directly applying magnetic flux to adjust the interaction between two qubits or by modifying the coupler’s resonant frequency enabling to a time dependent control of the coupling parameter. [2]

Trapped Ion Based Gates

Trapped-ion quantum computers implement single-qubit gates by shooting laser beams at individual ions. For Zeeman and Hyperfine qubits, two laser beams with a frequency difference matching the qubit transition energy shoot a single ion to perform a single-qubit gate.[3] One beam is sent through a multi-channel acousto-optic modulator (AOM) which outputs multiple beams. The AOM controls each beam’s amplitude and phase, allowing for different ions to execute different single-qubit gates in parallel. The second beam is applied globally. For Optical qubits, both beams go through a fiber-coupled AOM or an acousto optical deflector (AOD). The AOMs and the AODs split and control the amplitude and phase of the output beams, enabling arbitrary single-qubit gates. Another approach without lasers is to vary the energy level of each ion with a magnetic field gradient and use a global microwave pulse on all ions with specific frequencies targeting the individual ions.[3]

Trapped-ion quantum computers effectively couple the qubit states of ions by using shared vibrational modes of the ion chains as a medium of interaction, enabling entanglement and two-qubit gates.[3] Similar to single-qubit gates, this can be done with laser lights or a microwave pulse with magnetic gradient. The common two-qubit gate is the Mølmer-Sørensen gate, as it does not require the ion chain’s vibrational mode to be cooled to ground-state prior to starting as is required by the older Cirac-Zoller gate.[3] In the Mølmer-Sørensen gate, lasers with two tones are symmetrically detuned to ensure the target ions will either both receive photons or neither does. The case where only one target ion receives a photon is suppressed by the interference of the two lasers. This links the qubit states of the target ions while decoupling the gate operation from the vibrational state.[3]

Neutral Atom Based Gates

Neutral atom computers use lasers and microwave pulses to execute instructions to perform gate sequences or a general Hamiltonian.[1] Similar to the trapped ion implementation, the transition of the qubits can be controlled using two laser beams that are tuned with to frequencies with a difference equal to the qubit transition frequency. Here, the lasers can target a single qubit or a string of multiple qubits by dividing the beam with AODs, allowing for parallel execution. Controlling the frequency, intensity, and phase of various beams creates superpositions of qubit states, facilitating arbitrary rotations on the Bloch sphere.[1] While microwaves can be combined with lasers to achieve similar effects, microwaves pulses alone can not individual atoms because of their large wavelengths.[1]

To achieve entanglement and two qubit gates, the neutral atom computers excite atoms to a Rydberg state.[1] In this highly excited state, the valance electron are in a “large weakly bound orbit” resulting in unique properties such as extremely large dipole moments. [8] These properties enable long range interactions between Rydberg atoms. These interactions create Rydberg blockades.

Rydberg blockades occurs because the energy required to excite a pair of atoms to the Rydberg state simultaneously is inversely proportional to the distance between the two.[1] This means only one of neighboring atoms can enter the Rydberg state. This conditional interaction is the key to two qubit gates. Exploiting this property, two or three pulse lasers generate a Controlled-Z gate, effectively flipping the sign of the target atom depending on the state of the control atom.[1] Movable optical traps can reposition atoms allowing for non-local connections in a larger system.

Noise and Decoherence

Challenges arising from noise and decoherence are one of the primary hurdles preventing our quantum computers from transcending the NISQ era. These challenges prevent the attainment of ideal coherence time and gate fidelities. The paper will now review the noise and decoherence faced by superconducting, trapped ion, and neutral atom quantum computers.

Superconductor Quantum Computers

In superconducting quantum computers, the primary sources of noise include charge noise, magnetic flux noise, photon number fluctuations, quasiparticles, and cross talk.[2] Charge noise, a challenge in early superconducting implementations, stem primarily from fluctuating charges in the material. Efforts to reduce sensitivity to low-frequency charge fluctuations led to the development of capacitively shunted transmon.[2] Another issue, magnetic flux noise is caused by random flipping of magnetic dipoles on the surface of the superconducting metals resulting in unpredictable changes in the magnetic field. While the cause is unclear, adsorbed molecular oxygen on surface is the prime suspect.[2] Magnetic flux noise is stronger at low frequencies, but can be mitigated with techniques such as dynamical decoupling. In some architectures, photon number fluctuations will cause thermal radiation leaking from warmer areas of the computer. The number of photons stored in microwave cavities then fluctuate randomly and shift the qubit’s energy unpredictably.[2] Quasiparticles, unpaired electrons, naturally arise because of thermal energy tunnel through the Josephson junction also disrupting the qubits. Quasiparticles are theoretically low at cryogenic temperatures, but experiments show that there are more than expected possibly because of non-thermal processes or bottlenecks in the superconductor that prevents the electrons from pairing. They can be temporarily pumped away to suppress their effects. [2] Cross-talk, another source of noise and decoherence, is unwanted interactions between qubits or between qubits and control signals. They can be suppressed with techniques such as dynamical decoupling. [12]

Neutral Atom Quantum Computers

Cross talk also disrupt neutral atom quantum computers, although not as severely as superconducting systems.[1] The amount is typically moderate because the lasers are focused enough to minimize undesired effects. In neutral atom systems, cross talk occurs when gate pulses unintentionally affect nearby qubits or when parallel operations interfere with each other. [1]

Neutral atom quantum computers’ main source of noise are inaccuracies and fluctuations in the intensity and phase of the lasers used for qubit manipulations. The imprecisions creates inaccuracies causing noise and degradation in fidelity.[1] Another source of noise is from qubits colliding with leftover atoms in the vacuum. These collisions shorten the qubit lifespan. Furthermore, the external influences, such as Earth’s magnetic field or changing electric and magnetic fields from lab devices contribute to the noise experienced by neutral atom computers.[1]

Trapped Ion Quantum Computers

Trapped ion quantum computers also suffer from various types of noise, with motional heating being the primary culprit.[3] Even after ions are cooled to quantum ground state, the residual energy can combine with small electric noises to push the ions to higher motional states, causing degradation in fidelity and coherence. The electrical devices of the computer give off noise that can be mitigated by using low noise electronics and by electronically filtering the connection lines to the vacuum cell.[3] Another source of noise is nyquist noise. This type of noise arises from thermal fluctuations of the resistors and can be reduced by using trap components made with low resistance materials.[3] The metallic and dielectric surfaces of the trap cause another source of noise, surface noise. Surface noise can be reduced by cleaning and cooling the trap to cryogenic temperatures (about 10K). [3]

Beyond noise, a significant cause of errors in trapped ion quantum computers is ion loss and leakage. Ion loss occurs because interactions between ions and background gas molecules allow ions to escape the trap.[3] Leakage happens when the ions that are supposed to stay in the two energy levels that make up the computational basis of quantum states unintentionally transition to other energy levels because of shortcomings in the equipment such as poor laser calibrations that result in imperfect ion manipulation. Leakage can be mitigated with correction protocols, topological quantum memories, error detection, and restoration protocol.[3]

Scalability

In order to go beyond NISQ and achieve fault-tolerant quantum computation, the scalability of the quantum computer platform must accounted for. In this section, we will first consider the scalability of superconductor based implementation, followed by trapped ion, and then neutral atom.

Superconducting Quantum Computers

Scaling superconducting quantum computers to the range of 100-1,000 qubits faces two main challenges. First, as number of qubits increase, certain qubits lose coherence significantly more, up 30-100 times more, and disproportionately affect the system reliability. [6] Experimenters have shown that improving fabrication will decrease this degradations, but stress the need for improved benchmarking and process control, especially for cryogenic devices. [6] The second intermediate scale challenge is that noise and decoherence currently force recalibrations at a rate that cannot scale. Full recalibrations for a quantum computer of around 100 qubits are required approximately once a day, taking up to two hours. [6] Scaling this rate up to thousands of qubits would be simply impractical.

At an even larger scale of 1,000+ qubits, the challenges shift toward wiring, packaging, and noise mitigation.[6] Improving wiring and packaging of superconductor chips in the dilution refrigerators is critical as current wiring techniques cost millions for the wiring alone. For example, a $5M cryostat for a 150-qubit processor will allocate $4M just for wiring.[6] The challenge lies in mitigating crosstalk as many designs reduce wires and fit more qubits into a single dilution refrigerator, but at cost of introducing crosstalk.[6] As superconductor quantum computers scale, cross talk becomes an increasingly large hurdle. Scaling will also demand improved thermal noise management especially for components in cryogenic environments. As more qubits are packed in greater density, thermal noise increases, which in turn requires more noise control components.[6] The components add to wiring complexity, take up valuable real estate, and increase noise probability. Ultimately this increases the cost of scaling. Proposed solutions include a modularized, distributed system of smaller quantum computing units working in sync or to use compact, low power technologies such as cryoCMOS.[6]

Technologies such as cryoCMOS showcase an overall advantage that the superconductor quantum platform benefits from. They can leverage the existing semiconductor industry, which has been built on decades of work, investment, and efforts by brilliant scientists and industrialists, entire governments, and multi national corporations. The semiconductor industry offers advancement such as nanoscale fabrications and cryogenic wafer-scale integration techniques used in AI chips can be used to improve fabrication of quantum devices and scalability as well as to reduce noise. [6]

Trapped Ion Quantum Computers

Trapped ion quantum computers also face significant scaling challenges. One such challenge is the errors from the ion loss as previously mentioned.[3] Ion loss increases when the number of ions and residual gas pressure increases in the vacuum. With estimated loss rate of one ion per second for current systems, trapped ion computers will have difficulty scaling to the several thousands of ions with lifetimes of tens of hours necessary for practical applications.[3] Another consequence of scaling ions in the trap is increased instability caused by greater heat and increased collisions between ions and background gas molecules.[3]

Increasing the size and density of the ion chain to fit more qubits increases the difficulty in distinguishing chain oscillations, resulting in greater decoherence, and the difficulty in preventing cross talk. While crosstalk in trapped ion computers is not as severe as superconductor based implementations, it remains a concern with respect to scaling.[3] Managing space to fit more components to accommodate the demands of cooling, repumping, detection, and manipulation poses a challenge. Larger trap arrays also have increased radio frequency power dissipation, making it difficult to scale.[3]

Upon scaling to the 1,000+ qubits range, the larger trap arrays can no longer be arranged in a linear fashion, so specific zones with different purposes are used, such as a zone dedicated to entanglement. Ions are transported or “shuttled” between the zones. Shuttling is one of the primary blockers for scaling as 58% of a typical circuit budget is provisioned for this purpose compared to 1% and 41% for operations and cooling respectively.[3] Shuttling presents a trade-off between velocity and fidelity because increases in velocity means increases in heat and ion loss that result in decreases in gate fidelity.

To address these challenges, the guiding principles of quantum charge-coupled device (QCCD) architectures have been proposed.[3] Implementations of QCCD architecture include a grid based surface Paul trap and the race-track architecture. The grid based Paul trap is a 2D chip design that uses minimal control signals to simplify wiring and ion organization. The race-track architecture uses “parking zones,” for storage, special zones for operations, and a closed loop for shuttling.[3] To reduce shuttling, photons can link several QCCDs together for entanglement with a technique called photon-mediated entanglement. This is done with a beam splitter interfering the photons sent in an optical fiber from two ions from separate QCCDs. However, this approach creates another scaling challenge because of low collection efficiency of the emitted photons.[3]

Neutral Atom Quantum Computers

With respect to the three platforms, neutral atom quantum computers demonstrate the greatest potential for scalability. Despite being relatively young, neutral atoms quantum computers have already scaled to the order of hundreds of qubits and scaling further to the order of thousands seems to be within reach because the primary constraint is laser power availability and optical system performance.[1] This is not to say there are no major scaling challenges. One challenge is the complexity of rearranging increasingly large number of atoms prior to computation. More atoms require more atom transportation and increases required time and atom loss probability.[1] This is not easy because the preparation and computation must all be completed within the lifetime of the trapped atom. Increased lifetimes come at the cost of more complex infrastructure used to lower the temperature of the atoms.[1]

Besides the lifetime of the trapped atom, the relatively short lifetime of the Rydberg state is another major obstacle. The Rydberg state experiences natural decay to lower energy states and undesired transition between energy states caused by thermal radiations creating errors. These effects can be reduced with cooling, again at the cost of more complex infrastructure.[1] Another path is to increase the degree of parallelism as most implementations only allow for the parallel execution of a single qubit gates. Parallel execution of multiple gates would mean the compute time scales slower relative to the scaling in the number of qubits. [1]

General Discussion

The evaluation of superconducting, trapped ion, and neutral atom implementations of quantum computers, aspects beyond scaling and fundamental building blocks are important for a more a complete comparison. This section compares implementation on various characteristics such as fidelity, connectivity, and execution time.

Fidelity

Neutral atom quantum computers currently achieve a two-qubit gate fidelity near 99.5%, surpassing quantum error correction threshold. This also places their two-qubit gate fidelity in a comparable range to that of superconductor implementations.[1] The primary hurdles for improved fidelity are atomic motion at non-zero temperatures and interference from laser scattering. Because these hurdles are of a technical nature, as opposed to fundamental constraints in the physics of things, solutions and improvements are likely to be discovered in time.[1] Trapped-ion systems currently lead in fidelity with a 99.99% fidelity for the two qubit gate.[3]

Connectivity

Besides having the highest fidelity, the trapped ion implementations also enjoy longer coherence times and superior qubit connectivity in the form of all-to-all connectivity.[3] The all-to-all connectivity is enabled by the unique vibrational modes in trapped ion architectures. Higher connectivity allows for efficient error correction and lowers the number of required physical qubits. [1] Neutral atom implementations can only offer intermediate connectivity by utilizing Rydberg states and the Rydberg Blockade. While the connectivity is not as strong as that of the trapped ion, neutral atom implementations may still be optimal depending on the required execution of two-qubit gates for the problem at hand. The executions of the two-qubit gates determines the qubit connectivity required by the quantum hardware, so it may be advantageous to trade off all-to-all connectivity of trapped ion implementations for the scaling capabilities of neutral atom implementations.[1] Superconducting implementations mainly have local connectivity mediated by couplers, and designs to improve connectivity increase cross-talk. [1,2,3]

Superconductor Quantum Computer Advantages

While losing out on connectivity, superconductor implementations win across the board in terms of gate execution time, meaning shorter algorithm run times. [3] Neutral atoms and trapped ion implementations are in a similar range in terms of execution times [1]. Neutral atoms are particularly slow because of the long rearrangement times before starting computation. Also because measurements are destructive in neutral atom computers, if they do not apply special techniques for mid-circuit measurements, the preparation process of cooling and rearrangement must be restarted.[1] A nontechnical advantage of the superconductor quantum computers is the access to capital and investments. Estimates show superconductor quantum computers are backed by investments of $4.68 billion as well as tech giants such as Amazon, Google, and IBM while trapped ion have $1.536 billion and neutral atoms have $323 million.[5] This disparity in investment also provides an indicator to the relative adoption by industry.

Natural Qubit vs “Artificial” Qubit

Trapped ion and neutral atom quantum computers benefit from having natural qubits.[1,3] The atoms and ions of the same species have a natural uniformity with identical characteristics and identical reactions to stimuli. This eliminates the need for calibration of individual qubits required of superconducting qubits. Another advantage that the trapped ion and neutral atom quantum computers share over superconducting circuits is the native SWAP gate. Because the particle represent the qubits, physically swapping the locations of these particles constitutes a SWAP gate. Native SWAP gates lower gate count, thus reducing errors and execution time.[1,3] Also neutral atom and trapped ion quantum computers can operate at room temperatures, but noise can be reduced and performance improved by lowering temperatures with cryostats. In contrast, superconductor quantum computers require near absolute zero temperatures to operate. This suggests that trapped-ion and neutral atoms are more cost-effective. Another difference is the size of trapped ion and neutral atoms qubits are at the particle level while a superconducting qubit is relatively larger with Google Sycamore chip holding 1 qubit per mm^2. [3].

Future Work and Conclusion

The paper is a cursory survey into the hardware implementations of quantum computers in an attempt to inform a beginner on the near term future of quantum computers. More work needs to be done in understanding the constraints of speeding up and optimizing the execution time of neutral atom quantum computers. Additionally understanding the disparity in execution speed between neutral atom implementations and superconducting implementation is critical to a complete picture. Another missing piece of this paper is understanding the dollar cost of building and manufacturing the three implementation. What follows is a short personal reflection of what is likely to come, an attempt to “gaze into the crystal ball”. The enlightened reader may feel free to skip the following as the section essentially drifts from facts to opinions of a novice.

The NISQ era brings to mind the early era of the classical compute where the use of vacuum tubes was common place and the ENIAC the latest innovation. The applicability was constrained by low reliability and high maintenance demands, size, and scaling challenges. That is to say vacuum tube computers did not provide real world applicability. One notable example was using the ENIAC to build the hydrogen bomb. A perhaps brash, but natural analogy seems to be apparent here.

Superconducting quantum computers seems to be the vacuum tube computers of today. The technology dominated at the time due to its relative maturity. The RCA, Western Labs, and GEs of that era are the IBMs, Google, and Amazons of today both investing largely in the technology. Just like the vacuum tube computers, superconducting quantum computers are limited by the need to maintain and recalibrate. The superconducting quantum computers require complex cooling infrastructure making energy efficiency poor in comparison to other platforms just as the vacuum tube computers required support infrastructure to mitigate vacuum tube heat and to supply enough power. The amount of power required of vacuum tube computers seem quiet substantial as well. The size of the superconducting qubit is much larger than that of trapped ion and neutral atom quantum computers and the other implementations do not require the cooling system from the start. Just as vacuum tube computers had difficulty scaling due to its inherent use of vacuum tubes, superconducting quantum computers are having difficulty because its inherent representation of a qubit introduces a lot of noise.

If superconducting quantum computers are the vacuum tubes computers of today, the natural question is who are the transistors of today? Neutral atoms seem to be very promising because of it’s great scalability potential, relative energy efficiency, size, and it’s natural qubit which allow it to bypass much the noise troubling superconducting quantum computers. This, however reaches the point of speculation. It is very possible that a better alternative technology appears or the two technologies develop in parallel, finding different use cases.

Citations

1. Wintersperger, K., Dommert, F., Ehmer, T., Hoursanov, A., Klepsch, J., Mauerer, W., Reuber, G., Strohm, T., Yin, M., Luber, S.: Neutral atom quantum computing hardware: performance and end user perspective. EPJ Quantum Technology 10(1), 32 (2023). doi:10.1140/epjqt/s40507-023-00190-1

2. Krantz, P., Kjaergaard, M., Yan, F., Orlando, T. P., Gustavsson, S., & Oliver, W. D. A Quantum Engineer's Guide to Superconducting Qubits. Applied Physics Reviews, 6, 021318 (2019). https://doi.org/10.48550/arXiv.1904.06560

3. Strohm, T., Wintersperger, K., Dommert, F., Basilewitsch, D., Reuber, G., Hoursanov, A., Ehmer, T., Vodola, D., & Luber, S. Ion-Based Quantum Computing Hardware: Performance and End-User Perspective. arXiv (2024). https://doi.org/10.48550/arXiv.2405.11450

4. Pogorelov, I., Feldker, T., Marciniak, C. D., Postler, L., Jacob, G., Kriegelsteiner, O., Podlesnic, V., Meth, M., & Negnevitsky, V. Compact Ion-Trap Quantum Computing Demonstrator. PRX Quantum, 2, 020343 (2021). https://doi.org/10.1103/PRXQuantum.2.020343

5. Ezratty, O.: Understanding quantum technologies 2024. arXiv (2024). https://doi.org/10.48550/arXiv.2111.15352

6. Mohseni, M., Scherer, A., Johnson, K. G., Wertheim, O., Otten, M., Aadit, N. A., ... & Martinis, J. M. How to Build a Quantum Supercomputer: Scaling Challenges and Opportunities. arXiv (2024). https://doi.org/10.48550/arXiv.2411.10406

7. Tripathi, V., Chen, H., Khezri, M., Yip, K.-W., Levenson-Falk, E. M., & Lidar, D. A. Suppression of crosstalk in superconducting qubits using dynamical decoupling. arXiv (2022). https://doi.org/10.48550/arXiv.2108.04530

8. Wu, X., Liang, X., Tian, Y., Yang, F., Chen, C., Liu, Y.-C., Tey, M. K., & You, L. A concise review of Rydberg atom based quantum computation and quantum simulation. arXiv (2021). https://doi.org/10.48550/arXiv.2012.100614v2

9. Graham, T. M., Phuttitarn, L., Chinnarasu, R., Song, Y., Poole, C., Jooya, K., Scott, J., Scott, A., Eichler, P., & Saffman, M. Mid-circuit measurements on a single species neutral alkali atom quantum processor. arXiv (2023). https://doi.org/10.48550/arXiv.2303.10051

10. Grimm, R., Weidemüller, M., & Ovchinnikov, Y. B. Optical dipole traps for neutral atoms. arXiv (1999). https://doi.org/10.48550/arXiv.physics/9902072

11. Schymik, K.-N., Lienhard, V., Barredo, D., Scholl, P., Williams, H., Browaeys, A., Lahaye, T.: Enhanced atom-by-atom assembly of arbitrary tweezers arrays. Physical Review A 102(6), 063107 (2020). doi:10.1103/PhysRevA.102.063107

12. Tripathi, V., Chen, H., Khezri, M., Yip, K.-W., Levenson-Falk, E. M., & Lidar, D. A. Suppression of crosstalk in superconducting qubits using dynamical decoupling. arXiv (2022). https://doi.org/10.48550/arXiv.2108.04530