Northwestern and Fermilab Leverage Underground NEXUS Data for NVIDIA Ising AI Benchmark
Summarize this article with:
Northwestern and Fermilab Leverage Underground NEXUS Data for NVIDIA Ising AI Benchmark Northwestern University and Fermilab have partnered to release a high-dimensional dataset generated at the Northwestern EXperimental Underground Site (NEXUS) to train and validate the NVIDIA Ising open model family. Located 107 meters beneath the surface at Fermilab, the NEXUS facility provides a low-background environment essential for isolating superconducting quantum devices from cosmic radiation. The data, derived from a month-long measurement campaign, includes the first reported observations of correlated charge jumps across multiple qubits in a controlled underground setting. This information is now globally accessible via the American Science Cloud (AmSC), providing a standardized dataset for AI-driven quantum research. The NEXUS dataset serves as a primary test case for NVIDIA Ising Calibration, a vision-language model (VLM) designed to automate the tuning of quantum processors. Unlike traditional numerical summaries, the VLM analyzes 2D experimental plots—such as sinusoidal patterns that shift during charge jump events—to diagnose hardware performance. By combining real NEXUS experimental data with synthetic examples, the researchers established a benchmark that evaluates the model on six diagnostic tasks: technical description, outcome classification, physical interpretation, fit quality assessment, parameter extraction, and stability judgment. This benchmark is intended to create a community standard for how vision models interpret quantum experimental results. To facilitate the deployment of these tools, the collaboration utilizes Fermilab’s centralized GPU infrastructure. NVIDIA Ising is hosted on a high-performance cluster and registered with a lab-wide model router, allowing researchers to apply the NVQCA agent to experimental data through a single endpoint. This infrastructure supports real-time identification of charge jumps an d quasiparticle dynamics, addressing a major bottleneck in quantum processor calibration. Future efforts under the SQMS and CosmiQ programs are expected to expand this framework by incorporating additional quantum sensing data from Fermilab’s QUIET and LOUD testbeds into the open-access cloud. For technical details regarding the NVIDIA Ising benchmark and NEXUS data integration, consult the Northwestern University announcement here. Further context on the correlated charge noise study and the resulting Nature Communications paper is available via the Fermilab newsroom here. April 14, 2026 Mohamed Abdel-Kareem2026-04-14T21:21:48-07:00 Leave A Comment Cancel replyComment Type in the text displayed above Δ This site uses Akismet to reduce spam. Learn how your comment data is processed.