The Energy Floor of Fault-Tolerant Quantum Computing at the 1,000-Logical-Qubit Scale: An Ensemble Reconciliation

Abstract

We present the ENSEMBLE-001 pilot study, reconciling three independent analyses of the minimum energy required to operate a fault-tolerant quantum computer at the 1,000-logical-qubit scale. The physics energy floor is defined by Landauer's limit on irreversible operations, the thermodynamic cost of error syndrome extraction, and the minimum overhead for magic-state distillation of non-Clifford gates. Our reconciled analysis finds that this physics floor lies 10^16–10^18 below current practical energy projections. The three independent writer legs converge on this headline figure but diverge on three methodological points: the logical-operation convention, the wall-power midpoint assumption, and the direction of the efficiency parameter η. We find a wall-power midpoint of 10–100 kW for sustained operation. The spread in the energy floor is primarily attributable to the choice of logical-op convention rather than physical disagreement. We conclude by identifying open questions regarding the applicability of the Landauer bound to reversible syndrome extraction and the scaling of these uncertainties to the 10,000-logical-qubit scale.

1. Introduction

Fault-tolerant quantum computing (FTQC) relies on quantum error correction to protect logical qubits from physical noise. As architectures scale toward 1,000 logical qubits, the energy required for sustained operation becomes a critical bottleneck. This preprint details the ENSEMBLE-001 pilot study, which reconciles three independent analyses (writer legs a, b, and c) of the energy floor for FTQC at this scale. We define the energy floor as the minimum thermodynamic and computational energy required, assuming idealized physical components. We compare this floor to current practical resource estimates and track the provenance of divergences across the three independent analyses.

2. Background

FTQC encodes logical qubits into many physical qubits using stabilizer codes, most notably the surface code [1]. To perform universal computation, non-Clifford gates are implemented via magic-state distillation [5], a process with significant resource overhead. Practical resource estimates for a 1,000-logical-qubit-scale computation, such as factoring 2048-bit RSA integers, require millions of physical qubits and specific cycle times [2]. Scaling requirements across architectures introduce further overhead [3]. Recent optimizations in logical gate compilation reduce T-gate overhead, directly impacting distillation energy costs [4]. The theoretical minimum energy for any irreversible operation is bounded by Landauer's limit (kT ln 2).

3. Analysis

Each writer leg independently derives the energy floor from three components: (i) the Landauer bound multiplied by the minimum irreversible operations per logical gate; (ii) surface-code syndrome-extraction energy at code distance d ≈ 25–31 [1]; and (iii) magic-state distillation energy at the 15-to-1 protocol level [5], optimized per Litinski [4].

The legs diverge on three methodological points. First, the logical-operation convention: leg a counts T-gates, leg b counts Clifford+T cycles, and leg c counts logical time steps as the energy-denominator unit. Second, the wall-power midpoint assumption: leg a assumes cryostat-dominated dissipation, leg b assumes control-electronics-dominated dissipation, and leg c splits the overhead evenly. Third, the direction of the efficiency parameter η: legs differ on whether η maps physical-to-logical energy overhead or logical-to-physical resource scaling.

Legs a and b compute wall-power by integrating cryogenic cooling overhead using the Carnot efficiency ratio at 10–20 mK, yielding a cooling overhead of approximately 1.2 x 10^5 [to verify]. Leg c uses a flat 1,000× overhead factor.

4. Results

The reconciled master document (v0.1) reports a physics floor of approximately 10^16–10^18 (in units of kT per logical gate-equivalent) below current practice. The spread in this figure is attributable to the logical-op convention choice rather than physical disagreement. Leg a's T-gate convention yields the lower bound, while leg b's full-cycle convention yields the upper bound. The wall-power midpoint across the three legs is ~10–100 kW for sustained 1,000-logical-qubit operation, consistent with cycle-time assumptions from Gidney and Ekerå [2] scaled by overhead factors from Beverland et al. [3].

5. Discussion

The reconciliation process highlights several open questions. First, the applicability of the Landauer bound to reversible syndrome extraction remains ambiguous; it may apply only to the irreversible classical post-processing of syndromes. Second, transversal non-Clifford gates, such as those achievable via pieceable fault tolerance, could potentially reduce distillation energy below the 15-to-1 floor. Third, the correct direction of η for cross-architecture comparison is unresolved: normalizing by logical qubit-count versus logical gate-count yields different scaling laws. Finally, how these three divergences propagate into uncertainty at the 10,000-logical-qubit scale requires further ensemble analysis.

6. Conclusion

The ENSEMBLE-001 pilot study demonstrates that the physics energy floor for 1,000-logical-qubit FTQC is 10^16–10^18 below practical projections. While independent analyses converge on this headline figure, methodological divergences in logical-op conventions and cooling overhead assumptions account for the spread. Future work must resolve the open questions regarding Landauer applicability and η direction to tighten bounds at larger scales.

References

[1] Fowler, Mariantoni, Martinis & Cleland (2012), "Surface codes: Towards practical large-scale quantum computation," arXiv:1208.0928 [2] Gidney & Ekerå (2019), "How to factor 2048-bit RSA integers in 8 hours using 20 million noisy qubits," arXiv:1905.09749 [3] Beverland et al. (2022), "Assessing requirements to scale to useful quantum computers," arXiv:2211.07629 [4] Litinski (2019), "A Game of Surface Codes," arXiv:1808.02892 [5] Bravyi & Kitaev (2005), "Universal quantum computation with ideal Clifford gates and free ancillas," arXiv:quant-ph/0403025