News

Fermilab, IBM and Nvidia plan a modular quantum computing platform

The collaboration will bring quantum processors and accelerated classical computing together for scientific workloads, alongside new work on GPU-based error correction.
Inside a dilution refrigerator at the London Centre for Nanotechnology.

Fermilab, IBM and Nvidia plan to pair IBM Nighthawk quantum processors with Nvidia DGX B200 systems at Fermilab for experiments in high-energy and nuclear physics. The installation is designed to grow beyond a single quantum processor, allowing researchers to investigate how several such processors can share work with the classical computers alongside them.

Error correction gives the classical machines a job each time the quantum hardware returns a set of error measurements. The decoder has to infer which faults could have produced those measurements, and several possible corrections can fit the same observations. Searching more of them takes time. Anna Grassellino, Fermilab’s chief technology officer, identified three jobs for accelerated computing in the collaboration: “design algorithms, study errors and coordinate increasingly complex quantum systems”.

Nvidia has released its TCBP-JOBS decoder through CUDA-Q QEC 0.9.0. The method combines trajectory-calibrated belief propagation, which learns fault probabilities, with a jittered ordered-statistics beam search that uses GPU parallelism to explore candidate corrections.

The associated research evaluates several code families, including surface, colour and bivariate bicycle codes, across memory and surgery circuits. Its largest reported speedup, around 13,000-fold, is a comparison with an unsparsified long-beam Tesseract configuration in a bivariate bicycle decoding case.

Across the reported comparisons, decoder speed and logical-error performance vary with the code and noise model. The search has to explore candidate corrections quickly enough to be useful while retaining enough information to select a reliable answer, so the paper reports those outcomes together.

Alongside decoding, the collaboration addresses the practical problem of allocating quantum and classical resources. Nvidia has described work to bring quantum processors into Slurm scheduling through established quantum resource-management interfaces, so a job can request the required resources through a familiar computing environment.

A job that needs both a quantum processor and a GPU cluster has to secure time on both. Handling those requests through Slurm would let a computing centre apply its allocation policies to the combined experiment.

The partners have released the decoder in CUDA-Q QEC 0.9.0 ahead of the planned Fermilab installation.

Sources: Nvidia: Fermilab and IBM collaboration announcement; Fermilab: planned modular co-design platform; Nvidia Research: TCBP-JOBS decoding and noise learning; Nvidia: quantum resources and Slurm scheduling.

Continue reading