.webp)
Co-designing physical and QEC architectures means choosing the quantum error-correcting code, the logical operations, and the hardware implementation together — starting from the structure of the target application — rather than optimizing each layer independently. Done well, co-design reduces the resources needed for useful quantum computation by orders of magnitude, moving the roadmap to fault tolerance forward faster than hardware scaling alone.
Why not just design each layer separately?
The traditional approach treats the stack as independent layers: pick an algorithm, compile it to a general-purpose error-correcting code, and run it on whatever hardware exists. Each hand-off leaves enormous efficiency on the table, because generic choices can't exploit the structure that specific applications and specific hardware actually have.
Co-design inverts this. Start from the fundamental ingredients of the target workload — for lattice Hamiltonian simulation, that's lattice translations and problem connectivity — and let those symmetries guide the code design, the compilation, and the physical implementation simultaneously.
What does co-design look like in practice?
QuEra's recent BB* architecture work is a concrete example. The target application is local lattice Hamiltonian simulation — a natural megaquop-scale workload for materials science. Its translation symmetries lead naturally to high-rate bivariate bicycle codes, where the required structured Clifford operations and small-angle magic can be applied in parallel across all logical qubits in a block. On neutral-atom hardware, that same translation structure maps directly onto fast, low-depth, highly parallel acousto-optical deflector shifts — the native way the machine moves groups of atoms.
The result: for prototypical megaquop simulation applications like transverse-field Ising and Fermi-Hubbard dynamics, roughly 100x fewer physical qubits and 1,000x less runtime than the same workloads required only a few years ago.
Other examples run through QuEra's research portfolio. Ultra-high-rate code searches specifically look for codes with hidden symmetries that enable control-efficient syndrome measurements using the hardware's parallel optics. Work with Los Alamos National Lab produced a transversal architecture offering more than 100x speedup over fully fault-tolerant approaches for specialized tasks like Trotterized Hamiltonian simulation. And "tricycle" codes generate magic states directly inside high-rate codes using only low-depth transversal operations.
Why are neutral atoms especially suited to co-design?
Co-design only pays off if the hardware can actually implement the codes it points to. Neutral-atom systems provide flexible, reconfigurable connectivity — atoms move, so any qubit can interact with any other — plus massive parallelism and heterogeneous zones. That flexibility makes rapid architectural exploration viable: different codes can serve different roles (memory, computation, magic state generation) within a single device, and when research uncovers a better code, the architecture can adopt it without redesigning the machine. Hardware with fixed, local connectivity can't follow where co-design leads.
Why does co-design matter for the roadmap?
Because it compresses timelines. Scaling from megaquop to gigaquop systems means overcoming three QEC overheads: space (physical qubits per logical qubit), time (logical operation depth and throughput), and classical decoding. Co-design attacks all three at the architectural level rather than waiting for bigger hardware — which is how QuEra's roadmap reaches 1,000+ logical qubits executing a billion operations in 2028/29, quite a bit earlier than many in the field expected.
