Bicycle Architecture Compiler

Tour de Gross arXiv:2506.03094 presents the Bicycle Architecture - a modular architecture for fault-tolerant quantum computation based on bivariate bicycle codes. The analysis features:

• A universal Instruction Set Archirecture (ISA) of logical gates in bivariate bicyle codes
• Circuit-level noise benchmarks of all the ISA instructions
• An end-to-end compilation scheme from Pauli-Based Computation (PBC) to the bicycle ISA

This compiler prototype implements and benchmarks this end-to-end compilation scheme so that it may inform future Qiskit development for fault-tolerant architectures. Some example use cases are:

• Random PBC circuits as a form of worst-case Clifford+T quantum circuits. See ./notebooks/random_circuits.ipynb
• User-specified PBC circuits supporting arbitrary-angle rotations, see ./notebooks/custom_circuits.ipynb

Installation

This software is tested on some Linux and macOS platforms.

Rust

We recommend using rustup to install a rust toolchain. Then run

shell> cargo build --release

Binaries are built and placed in ./target/release/. In the documentation we use these binaries without including their (relative) path. Alternatively, you can use cargo run --release --package <packageName> -- <arguments>.

Gridsynth

The compiler depends on the gridsynth binary for synthesizing angles other than $\pm\pi/4$. Gridsynth can be installed by following the installation instructions. Please ensure gridsynth is available on your path.

To test your installation, the following command should succeed

shell> gridsynth -p --epsilon 1e-4 0.1

and print something like (the exact output may differ)

HTSHTSHTSHTSHTHTSHTSHTSHTSHTHTHTSHTHTSHTHTSHTSHTSHTHTSHTSHTHTHTSHTSHTHTHTSHTHTHTHTSHTSHTS

Optional dependencies

To run the notebooks, we require the packages numpy, matplotlib, and jupyter. These can be installed via

shell> pip install -r optional_dependencies.txt

Furthermore, the following commandline applications are helpful and may be required for some functionality:

• jq - commandline JSON processor, or parsing JSON to a newline-delimited input.
• GNU parallel in order to run ./scripts/run_random_numerics.sh in parallel.

Usage

This Rust workspace consists of many Rust crates (packages) that can be used either as library or, in many cases, also as a binary. The contents of the repository as summarized as follows.

pbc-compiler/
├── analysis/                      # Python modules for postprocessing and steering
├── scripts/                       # Scripts for generating data (has some overlap with analysis)
├── notebooks/                     # Notebook for running and plotting a random circuit experiment.
├── data/                          # Cached measurement tables, and random circuit data
└── crates/
    ├── bicycle_common/            # Common definitions. Bicycle instructions.
    ├── bicycle_benchmark/         # Random generation of PBC circuits
    ├── bicycle_cliffords/         # Clifford gate implementation via search
    ├── bicycle_compiler/          # PBC to bicycle circuit compiler
    ├── bicycle_numerics/          # Additive noise estimates and stats collection
    └── bicycle_random_numerics/   # Benchmarking via random PBC circuits

Each crate has more info in their respective READMEs.

Many binary crates can be used via their compiled binaries (obtained by cargo build or cargo run --package <package>). For an example workflow that generates benchmarks see scripts/README.md, and more advanced examples are illustrated by Jupyter notebooks in ./notebooks/.

Testing

To test in release mode, run

shell> cargo test --release

The script ./scripts/local_QA.sh runs quality assurance tests locally. This includes test, rustfmt, and clippy.