QC Design's new open simulator, Plaquette, unifies all three main approaches to error corrected quantum computing under physics rooted hardware noise, with benchmarks saying standard tools undersize real machines by tenfold.
Most quantum computing roadmaps are sized by simulators that reduce every hardware error to a simple random bit- or phase-flip. A new arXiv preprint from QC Design argues that shortcut is now mis-sizing real-world machines, and the company has released an open tool, Plaquette, that models physics-rooted hardware noise across all three major quantum computing paradigms at once.
Quantum programs don't run directly on physical qubits; they run on logical qubits, clusters of physical qubits wrapped in error-correction codes so a calculation can survive the constant failures of real hardware. The central planning problem is straightforward in form: how many physical qubits does a future machine need to run a useful program? Architecture-level simulators answer it by projecting logical error rates under different code and hardware choices, and the resulting numbers sit behind every public road map from trapped-ion, superconducting, and photonic teams.
The Plaquette paper attacks a methodological shortcut. Most architecture-level noise modeling reduces every error to uniform stochastic Pauli noise (a torrent of random bit- and phase-flips) and assumes that ignoring certain physics, including multi-state leakage (qubits briefly escaping the computational subspace), coherent over-rotations (gates drifting from their target angles), and mode-heating in bosonic modes, adds only a small amount of error to the simulation. That assumption held when machines were small and noisy. As hardware moves into regimes where those channels dominate, the same shortcut now biases the projections that multi-year hardware plans are built on.
Plaquette, QC Design's new open simulator and subject of an arXiv preprint (2607.08767), unifies all three fault-tolerant quantum computing paradigms (circuit-based, measurement-based, and fusion-based quantum computing) under a single platform. Most existing simulators target one paradigm in isolation. Plaquette also swaps Clifford-only stabilizer sampling for a three-sampler stack: standard stabilizer sampling, an XPauli-based leakage solver, and a near-Clifford sampler for circuits just outside the classically tractable regime. Where standard noise is uniform stochastic Pauli, Plaquette bakes leakage, coherent over-rotation, and heating-aware channels into the noise model directly.
The headline result in the paper is an underestimation claim: QC Design's own benchmarks argue that standard stabilizer-based simulators can underestimate logical error rates by more than an order of magnitude on hardware-relevant noise. That is the gap planner teams care about. A tenfold mis-sizing of physical-to-logical qubit ratios changes how many physical qubits a given program needs, which code a team picks (surface code, qLDPC, color code, or something more exotic), and when a given architecture becomes economically buildable.
Plaquette sits between high-level hardware specs and low-level circuit simulators. The company's announcement and its explainer describe it as a compiler front-end and design-automation layer, with automated micro-compilation that the company says lets the simulator handle tens of thousands of physical qubits. Quantum Computing Report frames the same release as architecture-level automation rather than competition for circuit-level tools.
The more-than-order-of-magnitude undercount is the authors' own measurement on their own benchmarks. It is a preprint result, not a peer-reviewed audit. No independent adoption is yet reported in the paper or in QC Design's announcement, which means Plaquette should be read for now as a well-articulated argument about a methodology problem, plus one company's bet on the fix.
What to watch next: whether independent groups reproduce the order-of-magnitude gap using Plaquette or comparable tooling, and whether any architecture roadmap published in the next quarter cites its noise channels. Either would convert the preprint's argument from a vendor assertion into field consensus.