Exploring the Synchronization Requirements of Fault-Tolerant Quantum Computers

Quantum error correction (QEC) codes use multiple physical qubits to create a logical qubit that detects and corrects errors caused by noisy environments. During this process, the QEC cycle goes through different phases. These phases must be synchronized to support multi-qubit logical operations. Even in an ideal quantum computer where operational latencies are uniform, logical qubits on the surface code could go out of sync from one another due to different reasons: (i) dropouts due to qubit/coupler defects could change code cycle times, (ii) code switching between the surface code and other codes that can create magic states more efficiently, (iii) code switching between the surface code and other codes that have higher encoding rates and hence make better quantum memories, and (iv) speculative use of leakage reduction circuits. To synchronize logical qubits, the leading logical qubit must pause syndrome generation to absorb the synchronization slack to match the phases of two logical qubits. During this time, QEC cannot track errors that may have occurred during the synchronization period, leading to uncorrectable errors.

We propose active synchronization to reduce the impact of synchronization and enable lower logical error rates. In active synchronization, the synchronization slack is absorbed by distributing it between multiple code cycles, rather than making the leading logical qubit wait for the one lagging at the end. Active synchronization breaks the total slack into smaller chunks - it slows the leading logical qubit gradually. Using active synchronization improves the logical error rate by up to 2.4x with a circuit-level noise model. Furthermore, because active synchronization incurs fewer errors, the average decoder latency is reduced by up to 2.2x.