up:: Quantum Computing MOC

Quantum Hardware Modalities

A quantum hardware modality is the physical technology a qubit is actually built from, and the field is running several different bets at once because none has clearly won. The four leading approaches are superconducting circuits, trapped ions, photonics, and neutral atoms, and each one makes a different trade among gate speed, coherence time, qubit connectivity, and how easily it scales to large numbers. A fifth approach, topological qubits, is a still-unproven pre-commercial bet whose central claim remains scientifically disputed, covered as its own trade-off block below. Those trade-offs are the reason a qubit count from one modality tells you almost nothing when compared against a qubit count from another, because a fast, noisy superconducting qubit and a slow, clean trapped-ion qubit are not the same unit of capability. Understanding the modalities is what lets you read a vendor announcement for what it is, and it is the hardware foundation under the advice in How to Tell Real Quantum Progress From Hype to judge a machine by fidelity and error correction rather than by the headline number.

Source: David P. DiVincenzo, “The Physical Implementation of Quantum Computation,” 2000, arXiv:quant-ph/0002077.

The short version:

• A modality is the physics a qubit is made from. The four leading ones are superconducting circuits, trapped ions, photonics, and neutral atoms, and each has real strengths and real weaknesses.
• Superconducting qubits are fast and lead on raw qubit count, but they need dilution refrigerators near absolute zero and have relatively short coherence times.
• Trapped ions lead on gate fidelity and coherence and offer all-to-all connectivity, but their gates are slower and scaling the traps up is hard.
• Photonic qubits run at or near room temperature and travel naturally through fiber, but single-photon loss and probabilistic operations are stubborn obstacles.
• Neutral-atom systems have reached large qubit arrays without cryogenics and offer flexible connectivity, and they are a fast-rising challenger.
• Topological qubits promise built-in error resistance, but the approach is pre-commercial and its foundational claim is scientifically disputed, so it belongs in the picture as a bet to watch rather than a demonstrated modality.
• Every modality is judged against the same yardstick, the DiVincenzo criteria, and the count that matters for cryptographic risk is logical qubits, not the physical qubits any modality announces.

Think of the modalities like different engines competing to power the first practical aircraft. One design revs fast but overheats and needs constant cooling. Another runs clean and steady but is heavy and slow to build at scale. A third is light and travels easily but keeps losing power along the way. A fourth arrives late but sidesteps the cooling problem entirely. None of them has yet flown the long distance that matters, and you cannot rank them by horsepower alone, because an engine that produces raw power but shakes itself apart is not closer to the goal than a quieter one that holds together. Comparing quantum modalities by qubit count alone makes the same mistake, and the honest comparison weighs speed, stability, and how the whole system behaves.

What is a quantum hardware modality?

A quantum hardware modality is the specific physical system chosen to embody a qubit, meaning the actual two-state quantum object that stores the 0-and-1 superposition and the physical mechanism used to control and read it. Every modality has to represent a qubit somehow, keep it isolated enough to stay quantum, operate on it with gates, and measure it at the end, and the different technologies solve those problems with completely different physics. A superconducting qubit is a tiny circuit cooled until electrical current flows without resistance. A trapped-ion qubit is a single charged atom suspended in a vacuum by electromagnetic fields. A photonic qubit is encoded in a particle of light. A neutral-atom qubit is an uncharged atom held in place by focused laser beams.

Source: David P. DiVincenzo, “The Physical Implementation of Quantum Computation,” 2000, arXiv:quant-ph/0002077.

The choice of modality cascades into everything else about a machine. It sets how fast the gates run, how long a qubit survives before decoherence wrecks it, whether any qubit can interact directly with any other or only with its neighbors, and what it takes to add more qubits without degrading the ones already there. Because those properties differ so sharply, the modalities are not interchangeable, and a lead in one metric on one platform can coexist with a serious weakness in another. That is why the field runs multiple approaches in parallel rather than converging on one winner.

What yardstick measures whether a modality can work?

Every modality is measured against the same checklist, the DiVincenzo criteria, published by David DiVincenzo in 2000 as the minimum requirements a physical system must meet to compute quantumly. They are the reason a serious observer does not ask only “how many qubits” but instead asks whether the whole system clears the bar on every axis. The five core criteria are:

1. A scalable system of well-characterized qubits. The technology must support many qubits whose behavior is precisely known and controllable.
2. The ability to initialize the qubits to a known starting state, such as setting them all to 0 before a computation begins.
3. Long coherence times relative to gate time, so a qubit survives long enough to run many operations before decoherence destroys its state.
4. A universal set of quantum gates, meaning the ability to perform arbitrary single-qubit operations and at least one entangling two-qubit gate.
5. Qubit-specific measurement, the ability to read out any chosen qubit reliably.

Source: David P. DiVincenzo, “The Physical Implementation of Quantum Computation,” 2000, arXiv:quant-ph/0002077.

The criteria matter because they expose the trade-offs. A modality can be excellent on scalability and weak on coherence, or superb on coherence and weak on gate speed, and the DiVincenzo checklist is what keeps you from mistaking a strength on one axis for readiness across all of them. Modern practice adds a further, harder requirement on top of the original five: for a cryptographically relevant machine, the qubits must be clean enough to sit below the error-correction threshold, because without that no modality can build the logical qubits a real attack needs.

How do the four leading modalities compare?

The four leading modalities each win on some axes and lose on others, and laying them side by side shows why no single one is obviously ahead. The comparison below reflects the general engineering trade-offs rather than any single machine, and the specific device figures that anchor it are cited in the sections that follow.

Modality	Cooling	Typical gate speed	Connectivity	Standout strength	Main weakness
Superconducting	Dilution refrigerator, near absolute zero	Fast (nanoseconds)	Nearest-neighbor on a chip	Highest physical qubit counts, mature fabrication	Short coherence, heavy cryogenics
Trapped ion	Room-temperature vacuum, laser cooling	Slow (microseconds)	All-to-all within a trap	Highest gate fidelity, long coherence	Slower gates, hard to scale trap size
Photonic	Room temperature or modest cooling	Very fast (light speed)	Reconfigurable via optics	No cryogenics, natural for networking	Photon loss, probabilistic gates
Neutral atom	Room-temperature vacuum, laser trapping	Moderate	Flexible, reconfigurable	Large arrays without cryogenics	Younger platform, atom-loss control
Topological (disputed)	Cryogenic	Unproven at scale	Unproven at scale	Promised hardware-level error resistance	Pre-commercial, and the underlying physics is contested

The single most useful thing to read off this table is that the axes trade against each other. Superconducting circuits buy speed and qubit count at the price of cryogenics and short coherence. Trapped ions buy fidelity and coherence at the price of gate speed and scaling difficulty. Photonics buys room-temperature operation and networking at the price of loss. Neutral atoms buy large arrays without cryogenics at the price of being a younger, less proven platform. A machine that leads on one column can trail on another, which is exactly why cross-modality qubit-count comparisons mislead.

What are the trade-offs for superconducting qubits?

Superconducting qubits are the modality with the highest physical qubit counts and the fastest gates, which is why most of the record-setting headlines come from them, and they pay for that with short coherence and demanding cryogenics. A superconducting qubit is a small circuit that must be cooled in a dilution refrigerator to a few thousandths of a degree above absolute zero, where it behaves as a clean two-state quantum system controlled by microwave pulses. IBM’s Condor chip, unveiled in December 2023, reached 1,121 superconducting qubits, the largest single processor of its kind at the time, and Google’s Willow chip, announced in December 2024, carries 105 qubits with average coherence times approaching 100 microseconds.

Sources: IEEE Spectrum, “IBM’s Condor Quantum Computer Has Over 1,000 Qubits,” spectrum.ieee.org; Google Quantum AI, “Meet Willow, our state-of-the-art quantum chip,” December 9 2024, blog.google.

The trade-off worth reading carefully is that a large superconducting qubit count is width, not depth. The coherence time near 100 microseconds means a qubit survives only a few thousand operations before decoherence corrupts it, so the raw count says little about whether the machine can run a long computation. The strongest evidence that superconducting hardware is maturing is not the qubit record but the 2024 demonstration of a below-threshold logical qubit on this modality, where adding physical qubits lowered the error rate, which is the property a code-breaking machine ultimately needs.

Source: Rajeev Acharya et al. (Google Quantum AI), “Quantum error correction below the surface code threshold,” Nature 638, 920-926, 2025, arXiv:2408.13687.

What are the trade-offs for trapped-ion qubits?

Trapped-ion qubits lead the field on gate fidelity and coherence and offer all-to-all connectivity, and they pay for that with slower gates and a hard scaling problem. A trapped-ion qubit is a single charged atom held in place by electromagnetic fields in a vacuum and manipulated with laser pulses. Because the ions share a common vibrational mode, any ion in a trap can be entangled with any other, which is genuine all-to-all connectivity rather than the nearest-neighbor layout a superconducting chip is stuck with, and that connectivity makes many algorithms cheaper to run. Trapped-ion systems have consistently posted the highest two-qubit gate fidelities of any modality, which puts them furthest below the error-correction threshold per operation.

The cost is speed and scale. Entangling gates on ions run in microseconds rather than the nanoseconds of superconducting gates, so a trapped-ion machine does fewer operations per second, and packing more ions into a single trap gets harder as the trap grows because the shared vibrational modes become more crowded and delicate. The leading engineering answer is to connect many small traps rather than build one enormous one, but networking traps together cleanly is itself an unsolved-at-scale problem. The result is a modality that is arguably the cleanest per qubit and among the hardest to grow to the millions of physical qubits a cryptographic attack requires.

Source: David P. DiVincenzo, “The Physical Implementation of Quantum Computation,” 2000, arXiv:quant-ph/0002077.

What are the trade-offs for photonic and neutral-atom qubits?

Photonic and neutral-atom qubits are the two approaches that sidestep the cryogenics that dominate superconducting machines, and each brings a different bet about how to scale. A photonic qubit is encoded in a particle of light, so it runs at or near room temperature, moves naturally through optical fiber, and is a natural fit for networking quantum computers together over distance. Its stubborn obstacles are that photons are easily lost, which corrupts a computation, and that the two-photon gates at the heart of the modality are probabilistic rather than deterministic, so the hardware has to work hard to make operations succeed reliably. Photon loss and detection efficiency remain the central engineering challenges for the approach.

Neutral-atom qubits are uncharged atoms held in a grid by focused laser beams, and they combine several attractive properties: they operate without dilution refrigerators, they can be arranged into large and reconfigurable arrays, and the atoms are naturally identical, which helps uniformity. The approach has moved quickly from a research curiosity toward large arrays, with neutral-atom systems reaching well over a thousand physical qubits, and its flexible, rearrangeable connectivity is a real advantage for error-correction layouts. Its main weaknesses are that it is a younger platform with less operational maturity than superconducting or trapped-ion systems, and that controlling atom loss and readout across a large array is an active problem.

Because the specific fidelity and qubit-array figures for photonic and neutral-atom vendors move quickly and are often reported by the vendors themselves rather than in peer-reviewed form, treat any single announced number as provisional, and prefer peer-reviewed measurements over press releases whenever a specific figure has to carry weight.

What are the trade-offs for topological qubits, and why are they disputed?

Topological qubits are a fifth, pre-commercial modality whose promise is a qubit with error resistance built into the physics itself, and whose central claim remains scientifically contested, so the honest treatment is a careful, even-handed one rather than a cheer or a dunk. The idea is to store quantum information in a topological property of the system, spread out in a way that local noise can’t easily corrupt, which would make the qubit far more stable than the conventional approaches and could reduce the enormous error-correction overhead a logical qubit otherwise needs. That is the appeal. The difficulty is that building one depends on physics that has not yet been unambiguously demonstrated.

In February 2025, Microsoft unveiled a chip it called Majorana 1 and announced it as the world’s first quantum processor powered by topological qubits. The supporting peer-reviewed evidence, a paper in Nature, is real engineering work: it demonstrated a single-shot measurement of fermion parity in an indium-arsenide and aluminum device with a 1% assignment error. The distance between that result and the announcement is where the calibration sits.

Read carefully, the Nature paper is more modest than the press. The paper’s own text notes that its measurements do not, by themselves, determine whether the low-energy states detected are topological, and Nature’s editors attached a note stating that the results do not represent evidence for the presence of Majorana zero modes in the reported devices. The independent physicist Scott Aaronson landed on the honest middle, that the claim of a topological qubit had at that point been neither accepted nor rejected by the field. That caution has history behind it: a 2018 Nature paper from a Microsoft-affiliated group claiming quantized Majorana conductance was retracted in 2021 for insufficient scientific rigour after outside physicists found problems in the data.

The dispute has stayed live and civil. Henry Legg, a physicist at St Andrews, published a formal challenge arguing the group had not demonstrated the basic physics needed for even a single topological qubit, and Microsoft published a rebuttal. Independent replication is what will settle it, which is exactly why topological qubits belong in this map as a bet to watch rather than a demonstrated modality: the promised advantage is real if the physics holds, and whether it holds is still an open question. The worked example of reading this kind of claim without a cheer or a dunk is in How to Tell Real Quantum Progress From Hype.

Sources: Morteza Aghaee et al. (Microsoft Quantum), “Interferometric single-shot parity measurement in InAs–Al hybrid devices,” Nature 638, 651-655, 2025, nature.com; Scott Aaronson, “FAQ on Microsoft’s topological qubit thing,” Shtetl-Optimized, February 20 2025 (quoting the paper’s own text and the Nature editorial note), scottaaronson.blog; Retraction Note, “Quantized Majorana conductance,” Nature 591, E30, 2021 (retracting H. Zhang et al., Nature 556, 74-79, 2018, for insufficient scientific rigour), nature.com; Adrian Cho, “Debate erupts around Microsoft’s blockbuster quantum computing claims,” Science, 2025, science.org.

Why don’t cross-modality qubit counts compare?

Cross-modality qubit counts do not compare because a qubit is defined by its physics, and the qubits differ in exactly the ways that decide capability. A thousand fast, noisy superconducting qubits and a hundred slow, clean trapped-ion qubits are counting different things, and neither number on its own tells you how deep a computation the machine can finish. The metric that actually gates a cryptographic attack is logical qubits, which fold gate fidelity, coherence, and connectivity into a single measure of reliable computing power, and building even one logical qubit demands clearing the error-correction threshold that a raw count says nothing about.

This is why the honest way to read a modality announcement is to ask the same questions across all of them. What is the two-qubit gate fidelity, and does it sit below the error-correction threshold? Has the machine demonstrated a logical qubit whose error rate improves with more physical qubits? How deep a computation can it sustain before decoherence wins? Those questions apply identically to a superconducting chip, an ion trap, a photonic processor, and a neutral-atom array, and they cut through the temptation to rank machines by the one number that is easiest to put in a press release. The full board-grade version of this reading lives in How to Tell Real Quantum Progress From Hype.

Common misconceptions

• “The modality with the most qubits is the closest to breaking encryption.” Qubit count is width, and it is not comparable across modalities. A modality with fewer, cleaner qubits can be closer to the logical qubits a cryptographic attack needs than one with a larger, noisier count.
• “One modality has clearly won.” None has. Superconducting circuits lead on count and speed, trapped ions on fidelity and connectivity, photonics on room-temperature operation and networking, and neutral atoms on large arrays without cryogenics, so the field runs all four in parallel.
• “All quantum computers need to be cooled to near absolute zero.” Superconducting machines do, but trapped-ion, photonic, and neutral-atom systems operate at room temperature or with far less cooling. Cryogenics is a property of one modality, not of quantum computing itself.
• “A qubit is a qubit, so the counts are directly comparable.” A qubit is defined by its physics, and superconducting, ion, photonic, and neutral-atom qubits differ in speed, coherence, and connectivity. The comparable figure is logical qubits, not physical ones.
• “Trapped ions are behind because they have fewer qubits.” Trapped ions lead on gate fidelity and coherence and offer all-to-all connectivity, which are the axes that matter for building logical qubits. Their smaller counts reflect a harder scaling problem, not lower quality.
• “Microsoft’s topological qubit means the modality is proven.” The February 2025 Majorana 1 announcement rests on a Nature paper whose own text, and the journal’s editorial note, say the results do not demonstrate the Majorana zero modes a topological qubit needs, and the claim is scientifically disputed. It is a pre-commercial bet awaiting independent replication.

Questions people ask

What are the main types of quantum computers? The four leading modalities are superconducting circuits, trapped ions, photonics, and neutral atoms. Each builds a qubit from different physics and trades gate speed, coherence, connectivity, and scalability differently, which is why no single one has clearly won (arXiv:quant-ph/0002077).

Which modality is closest to breaking cryptography? None is close, and the question is better asked in logical qubits than physical ones. The modality that first builds thousands of clean logical qubits below the error-correction threshold gets there, and as of 2026 all four are far from that, with superconducting hardware having demonstrated a single below-threshold logical qubit (arXiv:2408.13687).

Why do superconducting machines have the most qubits? Because the fabrication is mature and the gates are fast, so vendors can pack many qubits on a chip, with IBM’s Condor reaching 1,121 (spectrum.ieee.org). The price is short coherence, near 100 microseconds on Google’s Willow (blog.google), and heavy cryogenics, so the large count is width rather than depth.

Are trapped-ion computers better than superconducting ones? Neither is simply better. Trapped ions lead on gate fidelity, coherence, and all-to-all connectivity, while superconducting circuits lead on qubit count and gate speed. Which matters more depends on the goal, and both are being pursued seriously toward a cryptographic-scale machine.

Do any quantum computers run at room temperature? Photonic and neutral-atom systems operate at room temperature or with far less cooling than superconducting machines, which need dilution refrigerators near absolute zero. Room-temperature operation is a modality-specific advantage, though photonics still fights photon loss and neutral atoms are a younger platform.

How should I read a vendor’s qubit-count announcement? As a physical-qubit figure specific to that modality, not a cross-comparable measure of capability. Ask for the two-qubit gate fidelity, whether the machine sits below the error-correction threshold, and whether it has demonstrated a logical qubit, because those decide whether the count can ever become computing power (How to Tell Real Quantum Progress From Hype).

What are topological qubits, and are they real? Topological qubits are a fifth, pre-commercial approach that would store information in a way the physics itself protects from noise, which is why the promise is so attractive. Whether one has actually been built is disputed: Microsoft’s February 2025 Majorana 1 announcement rests on a Nature paper whose own text and editorial note say it does not demonstrate the Majorana zero modes required, so it awaits independent replication (nature.com, scottaaronson.blog).

Last verified 2026-07-12 · Maintained by Addie LaMarr, LaMarr Labs.