A logical qubit is not a component you buy. It is a budget you negotiate with noise, and the exchange rate is set by how much the decoder knows about each failure. Erasure conversion changes that exchange rate by telling the decoder where an error happened, at the cost of extra hardware on every physical qubit. The trade is rarely stated in the headline threshold number, which is why two groups can quote the same fidelity and arrive at opposite machine sizes.
Under erasure-converted qubit arrays the dominant failure mode is converted into a located loss event: the syndrome still has to be processed, but the decoder is told which physical qubit dropped out of the computation. Under conventional depolarizing-noise qubits the decoder sees an unlocated Pauli error and has to infer both the position and the type from the syndrome graph. That inference is the expensive part, because the number of candidate explanations grows with the code distance. A located error removes one whole branch of the search, which is why threshold estimates move so sharply with a modest amount of conversion hardware.
The resource that decides machine size is not gate fidelity alone but how much the decoder still has to guess about each failure.
Part One: What is actually being compared
To compare erasure-converted qubit arrays with conventional depolarizing-noise qubits honestly you first have to name the quantity being optimised. In fault-tolerant quantum error correction it is tempting to use the headline specification, but the binding quantity is the one that appears in the bill after a full year of operation: the sum of the committed cost, the per-unit cost and the cost of the occasions when the system behaves badly. Once that is written down, most of the specification-sheet argument simply disappears.
The cost structure behind erasure-converted qubit arrays is dominated by a fixed commitment that must be paid before the first unit of value appears, while conventional depolarizing-noise qubits converts that commitment into a variable charge on every unit. That difference is invisible in a per-unit comparison and decisive in a budget, because a fixed cost rewards scale and a variable cost punishes it. Whichever side your volume forecast lands on determines the answer long before the specification sheet does.
Part Two: The arithmetic that decides it
Take a surface-code patch targeting a logical error rate near one in a million. With unlocated errors at a physical gate error of about 0.1 percent, published overhead estimates sit in the range of one to several thousand physical qubits per logical qubit depending on connectivity and leakage assumptions. If roughly half the dominant errors can be flagged as erasures, the effective threshold roughly doubles and the required distance falls by a few code steps, which compounds into a reduction of perhaps 40 to 60 percent in the physical-qubit count. The saving is not linear in the converted fraction; it is convex in the decoder's remaining uncertainty.
The second-order consequence is organisational. Choosing erasure-converted qubit arrays creates a dependency on a capability that must be staffed, monitored and kept current; choosing conventional depolarizing-noise qubits keeps the capability outside the team and turns it into a vendor relationship. Neither is free, and the one that looks cheaper in a spreadsheet is frequently the one that adds a standing meeting to the calendar.
Part Three: Where the simple model breaks
The catch is that conversion is not free information. It needs an auxiliary level, a readout channel, and a per-cycle classification step that consumes time inside the error-correction cycle. If that classification adds even a few hundred nanoseconds to a cycle that was budgeted at a microsecond, the slower cycle gives physical noise more time to accumulate and eats part of the threshold gain. Leakage into non-computational states, which conversion hardware is supposed to detect, also produces false flags when the discriminator drifts, and a false located error is worse than an unknown one because the decoder trusts it.
A located error collapses the decoder's search space, which is why a modest converted fraction can remove roughly half the physical-qubit budget instead of a proportional slice.
Part Four: The cost that arrives later
There is a second bill that arrives at scale. Every converted qubit needs its own drive and readout line, and cryogenic wiring is priced per line, not per qubit. A design that cuts the qubit count by half but raises the line count by a third can end up more expensive in the dilution refrigerator, in the room-temperature electronics, and in the calibration labour that keeps thousands of discriminators aligned. Control-plane cost scales with the number of channels, while the decoder benefit scales with the logical error rate, so the two curves cross at a machine size that is easy to miss in a small demonstration.
Expect a split by scale rather than by industry. Large operators can amortise the fixed side of erasure-converted qubit arrays and will move first; smaller ones keep conventional depolarizing-noise qubits because their utilisation is spiky and their tolerance for operational complexity is low. The interesting competition happens in the middle, where a managed service can make either option look like the other.
Part Five: What this changes outside the engineering team
Outside the laboratory the consequence is about who gets access to fault tolerance. A machine that needs half the physical qubits is a machine that can be built with half the cryostat volume, which is the difference between a national facility and something a university or a cloud provider could replicate. That changes procurement, staffing and the kinds of experiments that get scheduled. It also changes the timeline for useful chemistry and materials workloads, because those are gated by logical-qubit count rather than by raw gate speed, and a threshold improvement is the shortest route to more logical qubits on the same hardware.
For a team choosing a roadmap, the practical question is not which threshold is higher in a paper. It is how many control channels, how many calibration hours and how much cycle time the improvement costs at the machine size they can actually build. Those three numbers decide whether erasure conversion is an optimisation or a new architecture.
A market sizing sanity check
A reasonable sizing: if fault-tolerant machines reach even a few hundred deployed logical qubits by 2032, and if each logical qubit carries on the order of a thousand physical channels with associated cryogenic and room-temperature control electronics, the addressable control-and-decoder hardware sits somewhere in the low single-digit billions of dollars annually, with a plausible band of about $1.5B to $4B depending on how many independent machines are built. The lower bound comes from a handful of flagship systems; the upper bound assumes cloud-style replication across regions. The dominant uncertainty is not qubit physics but how many organisations decide they need their own machine rather than buying time on someone else's.
All figures here are an author's synthesis of published ranges and a stated model, not a measurement of a shipped system. The point of the arithmetic is to locate the crossover and expose which assumption drives it; a vendor quote will differ, and the direction of that difference is more informative than the final number.
The comparison that matters
The comparison that matters is between uncertainty removed and complexity added. Erasure conversion wins when the decoder is the bottleneck and the control plane still has headroom; standard decoding wins when the cycle budget is already tight or when wiring density is the binding constraint. Neither result is a property of the qubit alone, and a roadmap that quotes only the threshold has quietly dropped half the equation.
The honest caveat is that this comparison is sensitive to one assumption: that the workload keeps its current shape. Shift the mix toward the case where conventional depolarizing-noise qubits is strong and the result changes without any technology changing at all. That is why the durable output of an evaluation is not a verdict but the threshold at which the verdict flips.
Choose the architecture by counting control channels and cycle time at your target machine size, not by comparing threshold numbers quoted on small devices.
SOL: HCqPmPJM3NVtS6YrmVaTZYnuP8XZ2PFgQZKT94VK1LkS
Creating this takes real effort — thank you for the support.
Top comments (0)