Beyond the big four lies a scrappier crowd of qubits, each betting everything on one breakthrough the front-runners cannot reach: a spin in silicon you could mass-produce on a chip line, a flaw in diamond born to network, and a Majorana quasiparticle that would protect itself in hardware — if it turns out to exist at all. A survey of the contenders still in the race.
Part of the How to Build a Quantum Computer series.
The four previous posts each followed one road to its horizon: a superconducting circuit faking an atom, a real atom grabbed by its charge, a thousand neutral atoms pinched in light, and a photon that refuses to sit still. Four industrial-scale highways, four well-funded bets. But the field guide to building a qubit does not stop at four pages. Off the main routes runs a scrappier, more fragmented crowd of platforms, and they do not pretend to be all-rounders. Each one stubbornly attacks a single breakthrough the front-runners cannot pull off — and bets the whole platform on it. This final deep-dive brings the contenders on stage: spins in silicon you could in principle stamp out on a chip line, color centers in diamond born to network, a topological qubit that would protect itself at the level of physics, and a long tail of honorable mentions. One of them might be the future. One of them might not even exist.
It is worth asking, before the tour, why anyone bothers. The big four already span fast-versus-precise, few-versus-many, compute-versus-network. What is left to want?
The honest answer is that every one of the big four pays a tax the others cannot waive. Superconducting and photonic machines cannot be mass-produced the way classical chips are — superconducting qubits are hand-tuned snowflakes, and a useful photonic computer needs millions of near-perfect components. Most platforms cannot network without coaxing a photon out of matter, and superconducting qubits cannot do even that without a lossy microwave-to-optical transducer. And every platform, without exception, is fighting the same brutal arithmetic of quantum error correction: thousands of noisy physical qubits spent to buy one good logical one.
The contenders in this post are, each, a wager against exactly one of those taxes. Semiconductor spin qubits bet that the answer to “you can’t mass-produce them” is use the existing semiconductor industry. Color centers in diamond bet that the answer to “you can’t network” is a qubit that emits a photon and runs at room temperature. Topological qubits make the most radical bet of all — that the answer to “error correction is ruinously expensive” is to build the protection into the hardware so there is far less to correct. The pattern that emerges is the cleanest in the whole series: the team that scales best and the team that networks best are almost never the same team. Let us meet them.
The first contender hides its qubit inside the most studied material on Earth. The qubit is the spin of a single electron confined in a gate-defined quantum dot — a tiny puddle of one electron, pinched out of a two-dimensional electron gas by voltages on metal gates patterned over a silicon/silicon-germanium heterostructure. Spin-up is $\vert 0\rangle$, spin-down is $\vert 1\rangle$. (A close cousin, the route Bruce Kane proposed in 1998, instead stores the qubit in the nuclear spin of a single phosphorus donor atom implanted in silicon
The blueprint is older than most of the hardware. In 1998 Daniel Loss and David DiVincenzo wrote down how to compute with these spins
whose strength $J(t)$ you switch on and off with a gate voltage. Pulse $J$ for the right duration and you get a $\sqrt{\text{SWAP}}$ — a genuine entangling gate — in tens of nanoseconds. The whole logical apparatus, in other words, is voltages on gates, which is precisely what a semiconductor foundry already knows how to make.
Reading the answer out is the subtle part, because a lone electron spin is far too faint to detect directly. The trick is to convert spin into charge: Pauli spin blockade (or energy-selective tunneling) makes the electron’s ability to hop to a neighbor depend on its spin, and a nearby charge sensor — a single-electron transistor or sensor dot, increasingly read by fast radio-frequency reflectometry — registers that hop in a single shot. Spin in, charge out, photon never required.
Why silicon in particular, and why the strange superscript? Because the enemy of a spin qubit is a bath of other spins. Natural silicon contains about 4.7% of the isotope $^{29}\mathrm{Si}$, whose nuclear spin-$\tfrac12$ flickers a random magnetic field at the qubit and dephases it. Strip those out — enrich the crystal to nearly pure spin-zero $^{28}\mathrm{Si}$ — and you give the electron a “semiconductor vacuum” to live in, pushing coherence into the millisecond range and letting fidelities climb. Every high-fidelity result below was achieved in purified $^{28}\mathrm{Si}$; the isotopic purification is not a detail, it is the enabling trick.
And the headline bet is written into the qubit’s birthplace: this thing can be built on the same 300 mm CMOS wafer lines that print the chip in your phone. Intel made the point concrete in 2023 with “Tunnel Falls,” a 12-qubit silicon spin chip fabricated in its commercial 300 mm fab using extreme-ultraviolet lithography — each qubit roughly 50 nm across, thousands of times smaller than a transmon in linear size, and millions of times smaller in footprint
The performance has finally caught up to the dream. For years two-qubit gates were stuck below 99% — too slow, too noisy to error-correct. Then, in early 2022, the dam broke: three independent groups vaulted the ~99% surface-code threshold within months of one another
There is even a counterintuitive perk: because a spin qubit is so small and so robust, it can run hot. Where a transmon demands roughly 10 mK, silicon spin qubits have executed a universal gate set above 1 kelvin
So what is the catch? Two of them. First, it is still tiny. The biggest working devices hold a dozen-ish qubits — Intel’s 12, SQC’s 11 — and charge noise plus uneven yield still means each dot is calibrated more or less by hand, the same snowflake problem superconducting qubits have. And the charge noise is not an incidental nuisance; it is the bill for the platform’s own strengths. The very electrical and spin–orbit coupling that lets you drive gates fast with EDSR and run the qubit hot is also the channel through which a stray electric field reaches in and dephases the spin — fast control and good isolation are pulling on the same rope. Second, it does not shine. A confined electron spin is not optically active; it emits no usable photon, so linking two silicon chips over fiber needs the same lossy microwave-to-optical transduction that haunts the superconducting platform. Silicon spin is the overachieving cousin who could in principle be mass-produced — but who is, for now, terrible at long-distance calls.
The next contender is the most beautiful idea in the room — and the most contested claim in this entire series. Read it with your skeptic’s hat on; that is the only sober way through.
The idea first. Every qubit so far stores information somewhere — in a circuit, an atom, a photon — and “somewhere” is exactly where noise can find it. A topological qubit would store information nowhere in particular. The vehicle is a Majorana zero mode (MZM): a peculiar excitation, predicted to live at the ends of a one-dimensional topological superconductor, that is its own antiparticle — half of an electron, in a sense. Alexei Kitaev showed in 2001 that a simple model wire can host exactly two such modes, one at each end
which is either empty or filled — and that occupation, the shared fermion parity, is the qubit. (Strictly, a usable logical qubit needs four Majoranas — two pairs sharing a fixed total parity, because total fermion parity is conserved — but this two-mode picture captures the non-locality and is the right intuition for the idea.) The information is not at the left end or the right end; it is in the correlation between two modes that can sit microns apart. A local perturbation — a stray charge, a phonon, a wandering field — touches only one Majorana at a time, and a single Majorana carries no information at all. To corrupt the qubit you would have to act on both ends at once, which local noise cannot coordinate. The energy splitting between the two parity states is exponentially suppressed with the wire length, $\sim e^{-L/\xi}$, so the qubit is protected by geometry. That is topological protection: error resistance built into the hardware, before any error-correcting code is layered on top. If it worked, the hardware itself would be suppressing errors for free.
And the computation is meant to be just as robust as the memory. You operate on the qubit not by pulsing fields but by braiding — physically moving the Majoranas around one another. These excitations are non-Abelian anyons, which means that swapping two of them transforms the parity state in a way that depends only on which worldlines crossed, not on how fast or how precisely you moved them; the gate, like the memory, is protected by topology. The real limit is that braiding alone is not enough: it generates only the Clifford gates, the subset a classical computer can efficiently simulate. To reach universal quantum computation you still need one non-topological ingredient — a “magic state” distilled at real overhead to supply the missing T-gate. Topology, in other words, would shrink the error-correction bill dramatically, but not abolish it.
Now the reality, stated plainly: no topological qubit has been convincingly demonstrated, and the leading claims are actively disputed. The history is a cautionary tale, and it must be told in full.
In 2018, a Delft–Microsoft team published a Nature paper claiming a quantized $2e^2/h$ conductance plateau as a smoking-gun signature of a Majorana zero mode
Microsoft has continued, and made two further claims that remain contested. In 2023 its team reported devices “passing the topological gap protocol,” a multi-step transport test meant to certify a topological phase
So weigh it. The promise is extraordinary: a qubit that corrects itself in hardware. The evidence, to date, is not there: no accepted topological qubit with a measured gate fidelity or coherence time, a retracted flagship paper, and live, published disputes over the most recent claims. It is the field’s highest-risk, highest-reward bet, and the only intellectually honest verdict in 2026 is not proven — a card still face-down.
The third contender is the solid-state mirror of the photonic platform — and the near-opposite of silicon spin: bad at mass production, a natural at networking. Its qubit is the electron spin of a nitrogen-vacancy (NV) center — a point defect where a nitrogen atom sits next to a missing carbon in the diamond lattice. The defect’s ground state is a spin triplet; pick two of its sublevels, say $m_s=0$ as $\vert 0\rangle$ and $m_s=-1$ as $\vert 1\rangle$, and you have a spin qubit you can drive with microwaves. The magic is the readout. Shine green laser light on an NV and it fluoresces — but the $m_s=0$ state glows noticeably brighter than $m_s=\pm1$, so counting photons tells you the spin state. That spin-dependent fluorescence lets you initialize, manipulate, and read out the qubit optically, at room temperature — and in isotopically purified diamond these spins reach a Hahn-echo coherence time $T_2 \approx 1.8$ ms, among the longest room-temperature electron-spin coherences in a solid
But the reason the NV center earns its place in this researcher’s series is networking, and here it is a genuine star. An NV electron spin can emit a photon entangled with its own state — a spin–photon entanglement. Send that photon down a fiber, interfere it with a photon from a distant NV, and a successful detection swaps the entanglement onto the two faraway spins: you have entangled two nodes that never touched. This is not hand-waving. In 2015 a Delft team led by Ronald Hanson used entangled NV spins 1.3 km apart to perform the first loophole-free Bell test, closing the locality and detection loopholes simultaneously and settling an 80-year-old argument about quantum nonlocality
The costs are the mirror image of the strengths. Scaling is brutal: defects nucleate at essentially random positions in the diamond, so assembling many into a large, precisely aligned array is a fabrication nightmare — there is no lithographic “place an NV here” the way there is for a transistor. Photon collection is poor: only a few percent of an NV’s emission lands in the coherent zero-phonon line (the sharp, usable part of the spectrum), and it comes out at 637 nm, well off the telecom band, so long hauls need wavelength conversion (the SiV-in-a-cavity approach is partly a fix for this). And network-grade coherence and optics generally want cooling to a few kelvin. None of this makes diamond a front-runner for large-scale computing — but as a network node and sensor, it pushes the solid state to its ceiling. Diamonds, it turns out, are a qubit’s best friend: a flaw you talk to with green light, that moonlights as a magnetometer, and that can fire its state down a fiber to a node across the city.
Beyond the three main contenders runs a tail of platforms too early, too specialized, or too historic to headline — but worth knowing, if only to read the field’s news correctly.
Trapped electrons and electrons on solids. Why trap a whole atom when the qubit lives on one electron? A small community traps bare electrons — historically floating on liquid helium, and more recently on a film of frozen neon. In 2022 a Chicago-led group integrated a single electron on solid neon into a superconducting-circuit (cQED) architecture and ran it as a charge qubit with surprisingly long coherence
Molecular qubits. Individual ultracold polar molecules, held in optical tweezers like neutral atoms but with a far richer internal structure (rotational, vibrational, and spin states), have been proposed as qubits for years. In 2023 a Princeton group achieved on-demand entanglement of two individually controlled molecules in a tweezer array, using their dipole–dipole interaction to make a Bell pair
Rare-earth ions in crystals. The tail’s most network-relevant member. Single rare-earth ions — erbium or europium doped into a host crystal such as yttrium orthosilicate — pair optical addressability with exceptionally long-lived nuclear-spin memories. Erbium’s appeal is specific and large: it emits directly in the ~1.5 μm telecom band that fiber networks are built around, sidestepping the very wavelength-conversion problem that dogs the NV’s 637 nm light. Single erbium ions have now been detected and coherently controlled
Nuclear-spin ensembles (NMR). The oldest qubit platform of all is also the most cautionary. Liquid-state nuclear magnetic resonance — radio pulses driving the nuclear spins of atoms in a molecule — ran the first quantum algorithms: Grover search in the late 1990s and, famously, the first physical run of Shor’s algorithm, factoring 15 with a seven-spin molecule in 2001
One honest caution to close the tail, because it trips up almost every popular ranking: do not confuse qubit counts across architectures. D-Wave’s Advantage2 carries more than 4,400 superconducting flux qubits — far more than any gate-model machine
Line the contenders up and the survey’s thesis snaps into focus: each bets on a different breakthrough, patching exactly where the front-runners fall short — silicon spin on mass production, color centers on networking, topological on hardware-level error correction. The map and table below spell out the trades; what unites the three is their shared soft spot, not their strengths. All are small-scale today — roughly a dozen working qubits each — with younger, more fragmented communities than the big four.
| Silicon spin | Topological (Majorana) | Color center (NV/SiV) | |
|---|---|---|---|
| Qubit | spin of one electron in a quantum dot | non-local parity of two Majorana modes | electron spin of a diamond defect |
| Headline bet | mass-produce on CMOS lines | hardware-level error protection | room-temperature network node |
| Best fidelity | up to 99.99% single-qubit (donor, 2025) | none demonstrated (disputed) | high, but few qubits |
| Coherence | ms (electron), s (nuclear) | unmeasured | ms at room temp; ~minute memory |
| Operating temp | ~0.1 K, even >1 K (“hot”) | ~10 mK (and unproven) | room temp; few K for networking |
| As a network node | needs transduction | n/a (not built) | natural and proven |
| Status | real, ~12 qubits | not proven | real, networking-focused |
And so the account closes, five builds deep. There is no universally best qubit — only different trades, each platform with its one gift and its matching tax. Superconducting is fast; trapped ions are precise; neutral atoms scale; photons fly; and the solid-state contenders each chase a single breakthrough none of the others can reach. The lesson that matters most for a quantum network is hiding in that sentence: the team that scales best (silicon spin) and the team that networks best (color centers) are not the same team — and no single platform yet has both. So the likeliest future is not one qubit seizing the crown, but a quantum network stitching many together — a superconducting or atomic processor here, a diamond node there, photons carrying entanglement between them, each doing the one job it does best. Which is, when you think about it, exactly the bet a quantum-network researcher is paid to make. The interesting question was never which qubit wins. It is how you wire them together.
More in this series — How to Build a Quantum Computer: Superconducting · Trapped ion · Neutral atom · Photonic · Other platforms