A modular path to 10,000 logical qubits

Our architecture group has posted a preprint describing a route to 10,000 logical qubits built from superconducting modules connected by cryogenic photonic links. The design uses surface codes at distance 21 and a distributed decoder that keeps pace with a 1 MHz syndrome cycle.

Readers who follow quantum architecture will recognize the genre: a stack of assumptions, each defensible in isolation, multiplied together into an optimistic machine. We have tried to be explicit about that. The paper includes a table of the thermal, optical, and classical-compute budgets, and a short section on the three assumptions we think are most likely to fail.

The first is that photonic link heat can be held below 1.5 mW per channel at scale. The second is that classical decoding can be distributed without introducing a synchronization bottleneck. The third is that module yield will improve faster than module cost. If any of these is wrong by a large factor, the design needs to change.

We publish preprints because the architecture is more useful as a target for criticism than as a press release. The manuscript is open for comment, and the simulation code is in the lab's public repository.