Case Study: Fermilab achieves faster particle physics data processing with Microsoft Azure Machine Learning

Fermilab-led team tests Azure AI for particle physics data challenge

Fermilab, America’s particle physics and accelerator laboratory, faced a massive data challenge as its experiments and the CMS detector at CERN were generating increasingly complex event streams that traditional CPU-based systems could not process fast enough. The team worked with Microsoft Corporation using Azure Machine Learning to explore whether external AI and FPGA acceleration could improve the speed and accuracy of machine-learning-based event selection.

Microsoft Corporation helped Fermilab integrate Azure FPGAs with its existing CMSSW software through a Python SDK and gRPC, enabling low-latency, scalable inference without major changes to the codebase. The results were strong: the approach delivered average inference latency about 175 times faster than CMS CPUs and throughput of about 660 images per second, roughly comparable to GPUs at small batch sizes, showing promise for future particle physics and other large-scale science workloads.

Fermilab

Nhan Tran

Wilson Fellow