English static mirror for SEO/GEO · AI-assisted translation · Read Chinese original

The 2026 Logical Qubit Race: IBM, Google, and Neutral Atoms (a November 2026 Perspective)

Forum topic · 小凯 · 2026-11-07

Summary

This article surveys the state of the logical-qubit race as of November 2026 across three leading hardware platforms. Google's superconducting chip delivered the decisive 2024 demonstration that enlarging a surface-code lattice (distance 3 to 5 to 7) lowers the logical error rate exponentially, confirming the threshold theorem experimentally; since 2025 it has pushed toward larger codes and logical qubits outliving the best physical qubits. IBM has committed to roughly 200 logical qubits running billion-scale circuits by about 2029, betting on area-efficient qLDPC codes instead of the standard surface code. Neutral atoms hold the record of 48 logical qubits operated from 280 physical qubits (late 2023), with reconfigurable tweezers enabling flexible code choices, though gates are ~100x slower and atoms can be lost. Ion traps lead on gate fidelity (two-qubit gates near 99.9%) but scale slowly due to serial ion chains. The article warns that each platform reports a different scoreboard—logical error rate, qubit count, or fidelity—and the only common currency is runnable circuit size ≈ logical qubits × tolerable gates per qubit. All three camps converge on surface-code-family codes plus magic-state distillation, targeting hundreds of logical qubits within a decade, while genuinely disagreeing only on how the code lattice is physically laid out.

Climbing teams competing for Everest don't settle things by shouting at base camp — they prove themselves at the camps: base camp, Camp I, the assault camp. The altitude at which each team plants its flag is the hard ranking.

Article 056 laid out the ledger (millions of physical qubits, distillation plants, lattice surgery). This article looks at where several climbing teams actually are right now. Three main contenders: superconducting (IBM, Google), neutral atoms, and ion traps — three of the "four boxes" from article 007, now re-ranking themselves on the slope of error correction. (The site already has an article, "IBM's 70 Logical Qubits", covering one battle; this piece is complementary to it.)

Timeliness note: this is a "(November 2026 perspective)" piece — milestones and roadmaps age, principles endure; when the landscape shifts, we update via revision write-backs, per the column's convention.

The Superconducting Team: Passing the "Threshold" Rite of Passage

Google's contribution was decisive: in 2024, its superconducting chip demonstrated that the surface code gets more stable as the code grows — as the lattice went from distance 3 to 5 to 7, the logical error rate fell rather than rose, dropping exponentially along the theoretically predicted slope. This was not "just another better number"; it was the experimentally confirmed threshold theorem (metric #2 from article 056 moving from paper to lab) — the rite of passage for the error-correction route. From 2025 onward, the team has climbed along two directions: larger lattices, and making logical qubits outlive the best physical qubits (direct evidence of compounding).

IBM's contribution is the roadmap and space-saving codes: a public commitment to deliver roughly 200 logical qubits running fault-tolerant circuits of hundreds of millions of gates around 2029 (the "Starling" program). Its bet is not the standard surface code but a new chessboard that saves 50–90% of the real estate (the qLDPC family — in plain language: rearranging the sentries from a square grid into fancy patterns, guarding the same code distance with fewer qubits). Code innovation and hardware competition are two separate fronts — a key variable for the second half of the 2020s.

The Neutral Atom Team: The Chessboard That Moves Itself

The neutral atoms' killer feature is reconfigurability: the qubits are atoms in optical tweezers — move the tweezers and the board rearranges. In late 2023, this camp demonstrated 48 logical qubits operating in formation (supported by 280 physical qubits) — the record for logical qubit count at the time.

Reconfigurability buys freedom of code choice: the surface code is locked into a square grid by the planar fabrication of superconducting chips, but neutral atoms can switch codes and try exotic layouts on the fly — "a chessboard that reshapes itself." The weaknesses are equally clear: slow gates (roughly 100x slower than superconducting) and atom loss (sentries who fall out mid-march need replenishment mechanisms).

The Ion Trap Team: A Small Elite Class

Ion traps have the highest gate fidelity in the field (two-qubit gates at the 99.9% level), and their logic experiments are straight-A results: in 2024 they set the benchmark record for logical gate fidelity. But their formation is serial — ions lined up in a single chain, with newcomers forced to queue — making scaling an inherently slow variable. On the roadmap its role is the "high-fidelity validation rig": new codes and new processes get verified in a small, elite system before being handed to a large formation.

The Scoreboard Trap: How to Compare Fairly

Article 009 covered how hard it is to judge quantum advantage; the same applies here — each team uses a different scoreboard: superconducting reports logical error rate, neutral atoms report logical qubit count, ion traps report gate fidelity. Cross-team ranking must return to the only hard currency:

> Runnable circuit size ≈ number of logical qubits × gates tolerated per qubit (= 1/logical error rate)

Anyone reporting only one metric is playing to their strengths. The reader's discipline: first ask which scoreboard is being used, then look at the number — and this applies to press releases too.

Shared Destination, Real Divergence

The consensus is remarkably uniform: use the surface code or its space-saving relatives, supply the yeast with distillation plants (the two taxes from article 056), and push for "hundreds of logical qubits running deep circuits" within ten years. The real disagreement is only one layer deep: how to lay out the board — planar lithography (superconducting, fast but fixed), movable atoms (neutral, slow but free), serial matrices (ion traps, precise but hard to scale).

One more calibration of article 039's "two calendars" (November 2026 view): the 2029–2033 roadmap target of "hundreds of logical qubits" is an engineering commitment, not a mathematical theorem — per article 056's ledger, these machines remain far from the thousands of logical qubits needed to "break RSA," but may be close to "running economically valuable quantum simulations." The next five years of this race determine the slope of the green zone's expansion.

Taking the Analogy Apart

Where the mountain-camp metaphor borrowed:

  • Disanalogy #1: Everest has one standard route and altitude is an absolute scale; the error-correction race is a two-dimensional choice surface of "hardware × code," with scoreboards defined by each team (article 056's pre-tax/post-tax question applies again). "Who leads?" must first be answered with "leads on which metric?"
  • Disanalogy #2: climbing teams don't share strength; the three hardware routes transfuse each other (superconducting's readout techniques, neutral atoms' reconfiguration ideas, ion traps' fidelity records are all being absorbed by the others). This is "co-opetition," not zero-sum.

Luggage to Take Home

1. Three teams, three cards: superconducting (decisive threshold experiment + 2029 roadmap + space-saving new codes), neutral atoms (48 logical qubits + reconfigurable board), ion traps (highest fidelity + slow scaling) — each picks codes that suit its own hardware. 2. The scoreboard trap: logical qubit count, error rate, and fidelity are not interchangeable; the hard currency is "runnable circuit size = qubit count × gate tolerance." 3. Shared destination, divergent paths: code families and distillation schemes are converging; "how to lay the board" is the real route dispute. The 2029–2033 "hundreds of logical qubits" is an engineering commitment — per the article 056 ledger, far from breaking ciphers, possibly near economical simulation.

Hook for the next stop: the race is lively but the ledger is quiet — turning article 056's ledger into an "economics": why does one logical qubit need a thousand physical ones? How is a distillation plant's throughput accounted? When does "correcting errors" become cheaper than not correcting (the break-even point)? Coming in article 058: *One Logical Qubit, a Thousand Physical Qubits: The Error-Correction Bill*.

Tags

#quantum-computing#logical-qubits#error-correction#surface-code#qldpc#superconducting-qubits#neutral-atoms#ion-traps

This page is an English static mirror generated for search and AI citation. It may be a full translation or structured summary of the Chinese original. Canonical interactive discussion lives on the Chinese page: https://zhichai.net/topic/178636891