Roadmaps to Fault Tolerance

From theorem to engineering plan

The threshold theorem says fault tolerance is possible below $p_{\text{th}}$. A roadmap is the engineering claim that a given hardware platform can actually get there — and stay there while scaling to millions of qubits. Every serious roadmap is organized around the same three milestones, in order of difficulty.

The three milestones every roadmap shares

1. Below threshold. Demonstrate physical error rates $p < p_{\text{th}}$ for all operations — one- and two-qubit gates, measurement, reset — under realistic, simultaneous operation.
2. Break-even / suppression with $d$. Show that a logical qubit, encoded at distance $d$, outlives its constituent physical qubits, and that increasing $d$ keeps lowering the logical error rate. This last point — error rate dropping as the code grows — is the true signature of being below threshold.
3. A scalable logical qubit and logical gates. Run many rounds of error correction in real time (the decoder must keep up with the syndrome stream), then perform fault-tolerant logical operations including the expensive non-Clifford $T$ via magic states.

The pivotal recent result is the experimental demonstration that a surface-code logical qubit's error rate decreases as the distance grows from $d=3$ to $d=5$ to $d=7$ — direct evidence of operating below threshold, milestone 2 achieved on hardware.

Platform trade-offs

Different physical systems hit these milestones with different strengths and weaknesses:

• Superconducting qubits. Fast gates (tens of ns) and a natural 2D layout suited to the surface code, but fixed nearest-neighbor connectivity and the need for cryogenic control wiring. The platform where below-threshold surface-code operation was first shown.
• Trapped ions. Very high gate and measurement fidelities and all-to-all connectivity within a trap, but slower gates and the engineering challenge of shuttling ions between traps to scale beyond a single register.
• Neutral atoms. Reconfigurable arrays with movable qubits give flexible, even long-range, connectivity that suits high-rate qLDPC codes; challenges include atom loss and measurement speed.
• Photonics. Room-temperature operation and natural networking, built on measurement-based or fusion-based schemes; the central difficulty is deterministic entangling operations and photon loss.

What separates a roadmap from a wish

A credible roadmap commits to numbers on three axes simultaneously: the physical error rate (must fall further below threshold to shrink $d$), the qubit count (must reach the $\sim 2d^2$ per logical qubit times the algorithm's logical width, plus magic-state factories), and the classical control (the decoder must process syndromes faster than they arrive, or the logical error rate floors out regardless of $d$). Progress on any one axis is necessary but not sufficient; fault tolerance demands all three at once.

What to remember

Roadmaps translate the threshold theorem into milestones — below threshold, suppression that improves with $d$, then scalable logical operations — and platforms differ mainly in how they reach each one. The recent demonstration of logical error rates that fall as the surface-code distance grows marks the field's crossing from "possible in theory" to "happening in the lab," with the remaining work being scale, speed, and the non-Clifford resources of the final lessons.

Sign in on the full site to ask questions and join the discussion.