CircLS: Compiling Lattice Surgery to Physical Circuits with Dynamic Allocation
TL;DR
CircLS is a full-stack lattice surgery compiler that allocates data patches dynamically: each patch is allocated at its first use and freed at its last use, and the freed tiles can be reused as ancilla paths.
Abstract
In fault-tolerant quantum computing, lattice surgery (LS) is one of the leading ways to realize logical operations, and the Pauli product measurement (PPM) is the basic instruction of LS-based computing. Compilers that work on the PPM sequence, however, stay at the logical level rather than the physical circuit level. This is because lowering PPMs from the logical level to physical circuits is complicated. CircLS is a full-stack lattice surgery compiler: it compiles a quantum program through its PPM sequence to a Stim circuit, lowering every PPM through linear-time stabilizer construction rules. On this basis, we develop a compiler that allocates data patches dynamically: each patch is allocated at its first use and freed at its last use, and the freed tiles can be reused as ancilla paths. Against the two baselines, CircLS reduces the allocated spacetime volume by 34% and 27% and the logical error rate (LER) by 40% and 48%, and it compiles programs that they cannot. CircLS is open source at https://github.com/John-YuehanZhang/CircLS