Google Quantum AI preprint reports 10x qubit overhead cut with modular error-correcting codes
Google Quantum AI researchers, led by Oscar Higgott, describe modular hyperbolic surface and color codes built by wiring planar modules together with sparse boundary connections. The work is an arXiv preprint (2610.03682) and has not yet been peer reviewed. Circuit-level simulations assumed a 0.1% physical gate error rate (SI1000 noise model) and inter-module gates ten times noisier at 1%, and they show a 10-fold reduction in physical qubit overhead relative to the surface code. Extrapolating to larger systems, the authors project the reduction rising to 15-fold, and they report modular codes with encoding rate k/n = 1/16 and distance d ≥ 22, which they say exceeds the [[288, 12, 18]] two-gross bivariate bicycle code. The design uses a syndrome extraction system about 4.5 times smaller than the code itself, measures high-weight inter-module stabilizers once every three error correction rounds, and implements logic through fault-tolerant walking circuits using code automorphisms.