First question: is it even a chip?
Most of this learning ladder is about superconducting qubits, where the qubit really is a feature etched into metal on a wafer. But once you step outside that world, the word chip starts to slip. Some platforms still build everything with lithography on a single die. Others use a lithographed chip only to make the trap or the cage, while the qubit itself is a free-floating atom or ion held above the surface by fields. Drawing that line clearly is the most useful thing you can do before comparing anything else.
On the truly-lithographic side sit gate-defined quantum dots (a single electron's spin trapped under metal gates in silicon), integrated photonic circuits (qubits made of light steered through waveguides etched in glass or silicon), and NV centres (a single atomic defect inside a diamond crystal). On the not-quite-a-chip side sit trapped ions and neutral-atom arrays: both use a fabricated chip or device to make the trap, but the qubit is an isolated atom floating in vacuum. That distinction drives almost everything that follows.
Is the qubit IN the chip, or held ABOVE it?
platform the qubit is... lithographic chip?
---------------- ----------------------- ------------------
superconducting a circuit on the wafer yes (qubit = chip)
spin (quantum dot) an electron under gates yes (qubit in Si)
photonic light in a waveguide yes (qubit in glass/Si)
NV centre a defect inside diamond yes (qubit in crystal)
trapped ion a free ion in vacuum trap is a chip;
qubit is NOT
neutral atom a free atom in vacuum trap optics + chip;
qubit is NOT
Rule of thumb: 'fabricated qubit' vs 'fabricated cage
for a natural qubit' is the deepest split in the field.A fair scorecard, axis by axis
There is no single 'best' number, so a fair comparison needs several axes at once. Four matter most for chip people. Coherence: how long the qubit remembers its state. Connectivity: how easily one qubit can talk to another. Manufacturability: can you make many of them, identical, with a known process? And maturity: how big and how reliable are today's real devices. Every platform trades these against each other; nobody wins on all four.
Scorecard (broad, honest, qualitative -- NOT a ranking): platform coherence connectivity manufactur. maturity ------------- --------- ------------ ----------- -------- superconduct. medium near-neighbor good (litho) high spin / dot medium near-neighbor excellent(a) medium trapped ion very long all-to-all(b) modest med-high photonic n/a (flies) link-by-link good (litho) medium neutral atom long reconfigurable modest rising NV / diamond long (warm) very local hard niche (a) spin dots reuse the silicon transistor toolchain -- in principle. (b) within one trap zone; scaling beyond it is the open problem. 'n/a (flies)': a photon has no idle coherence -- it is measured, not stored, so the usual coherence axis does not apply.
Read that table sideways and the bets become clear. Trapped ions and neutral atoms buy spectacular coherence and lovely connectivity — atoms are perfect copies of each other and can be shuffled or linked over long range — but pay for it with lasers, vacuum, and slow, hard scaling. Spin dots and photonics buy manufacturability by borrowing the silicon and photonics fabs the world already runs, but they fight tighter coherence, fussy uniformity, or the awkward fact that photons refuse to sit still and wait.
The shared headaches: yield and crowding
Two problems cut across every fabricated platform, and they are exactly the ones that bite superconducting chips too. The first is yield: if each qubit has some chance of coming out broken, the chance that all of them work falls off a cliff as the count grows. The second is frequency crowding: when qubits are tuned to specific frequencies and made by an imperfect process, two neighbours sometimes land on the same frequency by accident — a collision — and can no longer be addressed independently. Both get worse, not better, as you scale up.
Yield: chance the WHOLE chip works (each qubit good w.p. y)
whole-chip yield = y ^ N (N = number of qubits)
y = 0.99 per qubit:
N=10 -> 0.99^10 ~ 90% of chips fully work
N=100 -> 0.99^100 ~ 37%
N=1000 -> 0.99^1000~ 0.004% (basically none)
y = 0.999 per qubit:
N=1000 -> 0.999^1000 ~ 37%
Lesson: going from 99% to 99.9% per qubit is the
difference between 'nothing works' and 'a third work'
at the thousand-qubit scale. Per-qubit quality is
everything once N is large.Frequency crowding: collisions in a row of fixed-freq qubits
target frequencies (GHz), as designed:
Q1 Q2 Q3 Q4 Q5
5.00 5.10 5.20 5.30 5.40 <- nicely spaced
what fabrication actually delivers (each drifts a bit):
Q1 Q2 Q3 Q4 Q5
5.02 5.09 5.19 5.19 5.41
^^^^^^^^^^
Q3 and Q4 collided!
-> they overlap, cross-drive each other, and cannot
be controlled independently. One bad qubit can
spoil its neighbor too.
More qubits packed into the same frequency band
=> collisions become almost unavoidable. This is why
spacing, tunability, and tight fabrication all matter.Here the atom platforms have a quiet structural advantage worth being honest about: because every atom is a perfect copy of every other atom, they have no parameter variability and no frequency crowding from fabrication at all. Their scaling problems are different — delivering laser light to thousands of sites, shuffling atoms without losing them, building the vacuum — but the yield-and-crowding trap that haunts lithographic qubits simply does not apply to a natural atom. That is a real, often underappreciated point in their favour.
A shared part, and why nobody has won
One component shows up across several of these worlds: the single-photon source. Photonic qubits need one to emit qubits one photon at a time; trapped ions, neutral atoms, and NV centres all want one to convert a stationary qubit into a flying photon so two distant modules can be linked. A good single-photon source must spit out exactly one photon on demand, in a clean, identical state every time — and building one that is bright, pure, and indistinguishable enough remains genuinely hard. It is a shared bottleneck, not a solved part, and progress on it lifts several platforms at once.
- Resist 'winner' talk. When you read that one platform is 'ahead', ask on which axis — coherence, count, connectivity, or fidelity — because a lead on one axis routinely comes with a deficit on another.
- Separate the qubit from the wiring. Many headlines are about qubit counts, but the unsolved engineering — getting signals in and out without cooking the chip or crowding the spectrum — is shared by almost everyone.
- Watch shared components like single-photon sources. A breakthrough there helps photonics, ions, atoms, and NV at the same time — so it can quietly reshuffle the whole comparison.
So why has no modality won? Because the four axes pull against each other, and no platform yet does well on all of them at a useful size. Today's devices everywhere are small and noisy: tens to a few hundred working qubits, all firmly in the noisy, pre-error-correction era, none of them anywhere near replacing a classical computer. The honest summary is that this is a portfolio, not a race with a clear leader — and the deciding factor will likely be which platform's engineering bottlenecks give way first, not which has the prettiest single-qubit number today.