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

A Thousand Physical Qubits per Logical Qubit: The Bill for Quantum Error Correction

Forum topic · 小凯 · 2026-11-07

Summary

This article, part of a Chinese quantum computing series ('From Coins to Error Correction', entry 058), explains the economics of quantum error correction overhead. Below the surface-code threshold, each increase of code distance d by 2 divides the logical error rate by a constant factor (exponential suppression), but each step costs qubits: reaching 10⁻⁹–10⁻¹² logical error rates requires d≈25–30, meaning 2d² ≈ 1,000–2,000 physical qubits per logical qubit. The author emphasizes this figure is a snapshot of current hardware, not a permanent number. A second major cost is magic state distillation factories: a typical distillation plant occupies on the order of a thousand physical qubits while supplying only a few logical qubits' T-gate needs. The break-even condition is physical, not economic: error correction pays off only when the physical error rate is below threshold, which is why the whole industry's 'below threshold' milestone matters. The post also outlines three cost-reduction routes: qLDPC codes (saving 50–90% of qubits), faster real-time decoders, and hybrid usage of bare physical qubits in shallow circuit segments. It closes with three takeaways and a preview of a follow-up on magic state distillation.

> Prerequisites: entries 056 (error metrics and overhead), 057 (race and roadmaps), 055 (surface-code checkerboard), 023 (yeast and counting yeast). > Next up: entry 059, "The T-Gate Refinery: Magic State Distillation."

A Thousand Physical Qubits per Logical Qubit: The Bill for Error Correction

Shipbuilding is a strange business: before the first ship is built, the dry dock must be paid for — dry docks, gantry cranes, workshops, all fixed investments that burn money even when no ship is under construction. Without a dock, an ocean-going vessel is impossible; with one, every ship carries its share of the cost.

Error correction is the same business. Entry 056's checkup report gave three metrics; this post spreads that bill into economics: why does one logical qubit need a thousand physical qubits? How much land does the distillation factory occupy? And the most important question — when is error correction cheaper than no error correction? (The break-even point.)

Where "a thousand" comes from: the price of exponential suppression

First, restate entry 056's account as a curve. With a fixed physical error rate \(ε_p\) (say 10⁻³) and a threshold of 10⁻², the relation between checkerboard size d and logical error rate shows exponential suppression: each increase of d by 2 divides the logical error rate by roughly a fixed factor (the factor depends on how far ε_p sits below threshold).

This is both good news and bad news: suppression works (the exponential curve is real), but every level of suppression is paid for in qubits — d=25–30 is needed to reach 10⁻⁹–10⁻¹² (the precision for codebreaking), and 2d² means one to two thousand physical qubits supporting a single logical qubit. The "a thousand" in the title is the coordinate on this exponential curve corresponding to "today's hardware"; each tenfold improvement in \(ε_p\) (down to 10⁻⁴) shifts the coordinate far to the left — "a thousand" is not an eternal number, it is the current hardware's price quote (the qLDPC codes of entry 057 can save a large additional chunk).

The factory is bigger than the shop: the distillation plant's footprint

The other half of the qubit bill is the factory. Half of the hook planted in entry 023 is cashed in here (entry 059 covers the process): the "pure yeast states" needed for logical T gates must be refined in a distillation plant — and the plant itself is a huge qubit consumer: in a typical configuration, a single distillation plant occupies on the order of a thousand physical qubits, with output sufficient to feed only a few logical qubits' appetite for yeast.

The shipyard metaphor holds precisely here: shops (logical qubits) can be small and numerous, but the factory (distillation) must be large and centralized — and its capacity is back-derived from the order book (Shor's tens-of-billions of yeast states, entry 038). This is also the hidden line in every roadmap from entry 057: a big fraction of the total qubit count is not "for computing" but "for supporting the computation" — read any roadmap with the "factory footprint" pulled out separately.

Break-even: when does error correction "pay for itself"

The final step of the economics. The no-correction "running bare" plan: one physical qubit endures alone, lifetime about a thousand gates (entry 051). The correction plan: two thousand qubits plus a distillation plant plus a time tax, in exchange for a logical qubit with an unbounded lifetime (entry 055's compounding).

The break-even criterion is physical, not market-based: the physical error rate must be below threshold. Below threshold (entry 056, metric two), the bigger the checkerboard the more stable it gets, and the shape of the compounding curve guarantees "you inevitably win" — this is the entire economic basis for the world betting on error correction: not faith, but the shape of a curve. Above threshold, nothing works. Today's industry-wide keyword "below threshold", translated into business language, is: past the break-even point.

And after the break-even point there is a second economic hurdle: total-budget thinking. Running Shor requires tens of billions of gates, and the total failure probability must be below one — the budget must be split across every layer: the logical layer, the distillation layer, the measurement layer, decoding latency... no layer may overspend. This is the hardware version of entry 048's "end-to-end accounting": the algorithm's cost, the correction cost, the distillation cost — all aligned on one total ledger.

After reading the bill, read the "money-saving guide"

With the economics done, here are three ongoing "cost-cutting routes" (viewpoint: November 2026):

One, change the code family (entry 057's qLDPC): rearrange the sentinels from a square grid into fancier patterns, saving half to ninety percent of qubits at the same code distance — the biggest cost variable. Two, change the decoder: faster real-time decoding (including machine-learning-assisted) — decoding latency is the biggest part of the time tax, and saving time is saving lifetime. Three, mixed staffing: not every qubit in the fleet must serve as a "fully protected logical qubit" — use bare physical qubits in shallow circuit segments and bring in error correction only for critical segments (the main topic of entry 060, "survival guide for the transition zone").

Tearing down the analogy

Where the shipyard analogy borrowed well — and where it doesn't:

  • Disanalogy one: a dry dock is a fixed investment usable for decades once built; quantum error correction is a continuously burning service — sentinels patrol every microsecond. "Install it and you're done" does not exist; the bill is a flow meter, not a one-time receipt.
  • Disanalogy two: a business break-even point depends on market demand (whether tickets sell); the break-even point for error correction depends on a physical constant (the threshold) — markets can wait, thresholds only respect hardware quality. The analogy borrows "the dock precedes the ship", not "demand decides".

Luggage to take home

1. "A thousand" is a hardware quote: the current coordinate on the exponential suppression curve (d≈25–30 → 2d²≈1,000–2,000); each tenfold hardware improvement shifts the quote left, and qLDPC saves another half to ninety percent — overhead is a variable, compounding is the constant. 2. The factory is bigger than the shop: distillation plants occupy thousand-scale qubit footprints while feeding only a few logical qubits — on roadmaps, qubits "supporting the computation" outnumber those "doing the computation" (entry 059 tours the plant). 3. The break-even point is a physical constant: below threshold, the compounding curve guarantees error correction wins — the economic basis for the whole industry's bet; total-budget thinking (accounting layer by layer) is the discipline of engineering implementation, the hardware version of entry 048's "end-to-end".

The hook for the next stop: the most expensive line on the bill is the distillation plant — so let's tour the plant: why is the T gate so expensive? How is "impure yeast" refined step by step into "qualified yeast"? What purity counts as passing? Entry 059, "The T-Gate Refinery: Magic State Distillation."

*Tags: quantum computing, from-coins-to-error-correction, error-correction-overhead, break-even.*

Tags

#quantum-computing#quantum-error-correction#surface-codes#qldpc#magic-state-distillation#error-correction-overhead#threshold

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/178636892