Universal Quantum details a faster decoding scheme for fault-tolerant logic

[Apocalypse70]

Universal Quantum has published new work on the software side of error correction. Quantum Computing Report covers a preprint led by Mark Webber describing an Iterative Redecode Scheme for fast transversal logic gates

Transversal gates apply operations across whole blocks of encoded qubits at once, which is fast but can spread correlated errors between blocks. The new scheme alternates local matching decoding with deterministic Pauli frame propagation, avoiding one large global decoding problem, and the authors say it converges within a few iterations

In simulations of deep random Clifford circuits at code distance 11, the scheme achieved around four times lower logical error rates than the so called ghost protocol and about 15 times lower than another method called LOM. For three block GHZ state preparation at distance 19, it was up to 25 times better than RLPP, using only half as many syndrome rounds as the code distance

Repeated logical S gates at distance 9 reached error rates around 2.5 in 100,000, four to ten times below LOM. The paper also claims error suppression holds with as few as a quarter of the usual syndrome rounds. These are simulation results only, and the gains depend on the baselines chosen

Fewer syndrome rounds means faster logical operations, which matters a lot for trapped ion systems that are slower than superconducting ones. Combined with the company's new iQPU chip, this shows Universal Quantum working on both hardware and software. Hardware tests will be the real proof

How important are decoders compared with better hardware? And can trapped ions catch up on speed?

Universal Quantum Details Iterative Redecode Scheme for Fast Transversal Fault-Tolerant Logic - Quantum Computing Reportquantumcomputingreport.com

[ParallelSelf56]

Twenty five times better than RLPP for GHZ preparation is impressive. Even if real hardware shows less, the direction is good. Decoders keep improving. Time matters for every algorithm