Quantum Technology Roadmap Archive

A technology roadmap connects a desired capability to the scientific and engineering conditions needed to achieve it. An older roadmap is most useful when read as a conditional argument: what had to work, what was already demonstrated, and what remained uncertain? The historical sources below explain three influential ways to frame those questions. They are not predictions of a universal date when quantum technology becomes useful.

Helpful background. Questions for comparing quantum claims distinguishes demonstrated capabilities, conditional estimates and unsupported conclusions. For selected later developments, the 2026 research map was reviewed on 11 October 2026; it is a selective guide, and its two retained computing snapshots stop at 11 August 2026.

A roadmap is a conditional argument

A claim such as “application A becomes feasible with N logical qubits” usually abbreviates several conditions: logical gate error, clock speed, non-Clifford supply, input preparation, verification cost, and a selected classical comparator. A later device reaching the same qubit count may satisfy only one condition.

Read the quantitative definitions as carefully as the headline. “Error per gate” may mean average physical infidelity, a logical Pauli error, or a circuit failure probability. “Rate” may mean attempted runs, accepted states, or verified answers. Comparing different definitions creates a false trend even when every reported number is individually correct.

Historical entry points and what they establish

These three sources ask different questions and remain useful for different reasons.

• Physical requirements for computation. DiVincenzo’s 2000 article organizes requirements such as initialization, controlled operations, readout, and long coherence relative to gate times. It is a framework for evaluating implementations, not a calendar prediction or a complete modern engineering budget.
• Noisy computation as a research regime. Preskill’s 2018 article distinguishes the opportunities and limitations of noisy intermediate-scale devices from the longer-term fault-tolerant objective. Read it for the questions and assumptions posed at that time, without treating the label as a universal size threshold.
• Network capability stages. Wehner, Elkouss, and Hanson’s 2018 roadmap classifies increasingly capable quantum networks. Geographic reach, entanglement delivery, and programmable quantum operations are separate dimensions; satisfying one does not imply the others.

Their value lies in explaining the desired capabilities and their dependencies. A later result should be compared with the original physical statement and its assumptions, rather than a promotional paraphrase or an inferred deadline.

Compare forecasts with a fixed task and a changing model

Suppose an old estimate describes all relevant failures by $L$ logical fault events, each having probability at most $p_L$. The union bound gives

$$P_{\mathrm{fail}}\le Lp_L.$$

If $L=10^9$ and the target is $P_{\mathrm{fail}}\le0.01$, allocating $p_L\le10^{-11}$ is sufficient to certify the target with this union bound. It is not necessary for every possible fault distribution: the bound can be loose. A later experiment showing $p_L=10^{-6}$ does not supply this worst-case certificate merely because it demonstrates a logical qubit; a tighter workload-specific analysis could still establish the target.

Now suppose an algorithmic improvement reduces $L$ to $10^5$. The same sufficient allocation relaxes to $p_L\le10^{-7}$. Progress came partly from changing the workload, not solely from better hardware. Both the algorithm and the error criterion must remain visible in the comparison. No independence of the fault events was used; interpreting average gate infidelity as their probability would require an additional noise model.

The same issue appears in sensing: improved intrinsic sensitivity may be offset by deployment duty cycle, while a better inference model can improve application performance without changing the sensor. In networks, multiplexing can improve bits per second without improving bits per optical mode.

What changed when a forecast changed?

A meaningful retrospective distinguishes several outcomes. A prerequisite can be demonstrated while the application remains conditional. A projected method can fail under realistic noise yet remain correct in its ideal model. A classical algorithm can improve the comparator, shifting a previously favorable crossover. A device result can be replicated but with a revised calibration or uncertainty.

These outcomes should be attached to the original scope. “No useful speedup on this benchmark under these settings” is informative; “quantum computation does not work” is not supported by it. Similarly, a successful prototype does not validate all scaling assumptions.

Publication dates, versions and corrections help distinguish these possibilities. A revised numerical estimate can reflect a better algorithm or calibration model even if the physical device has not changed. A later classical solver can also shift the practical crossover without changing the quantum method. Comparing the same task at the same accuracy reveals which improvement matters.

Physics behind the projections

Use hardware scaling for yield and control constraints, quantum algorithms for access and precision, and quantum internet for delivery and loss assumptions. The 2026 research map connects selected dated developments to more specific research guides, each with its own stated literature cutoff.

Exercises

An older estimate predicts $10^6$ physical qubits for 100 logical qubits. A newer estimate predicts $2\times10^5$ physical qubits for 20 logical qubits. Has physical overhead per logical qubit improved?

Solution

Both ratios are $10^4$ physical qubits per logical qubit. The total changed because the workload size changed. A meaningful overhead comparison would also match logical error, clock rate, ancillas, factories, and operation set.

A roadmap states that a protocol would exceed direct transmission if memory lifetime exceeds $T$ and local gates remain below error $p$. An experiment demonstrates the lifetime requirement alone. Classify the result.

Solution

It demonstrates one prerequisite. The complete performance claim still depends on gate quality and the remaining protocol assumptions. In addition, the memory lifetime must remain adequate during the actual gate and networking schedule; combining results from incompatible operating conditions does not establish the full crossover.

References

• D. P. DiVincenzo, “The physical implementation of quantum computation,” Fortschritte der Physik 48, 771–783 (2000). Author preprint.
• J. Preskill, “Quantum computing in the NISQ era and beyond,” Quantum 2, 79 (2018). doi:10.22331/q-2018-08-06-79.
• S. Wehner, D. Elkouss, and R. Hanson, “Quantum internet: A vision for the road ahead,” Science 362, eaam9288 (2018). doi:10.1126/science.aam9288.