ORNL’s NEUROPix Project Gets DOE Funds to Put AI Inside Particle Detectors

TL;DR: ORNL’s NEUROPix project has secured a three‑year DOE grant to embed brain‑inspired AI chips directly in particle detectors, aiming to filter collision data in real time. The approach could cut the data deluge from modern accelerators by processing signals at the source.

Context Modern particle accelerators generate more data than can be written to disk, forcing scientists to discard potentially valuable information. Oak Ridge National Laboratory physicist Mathieu Benoit noted that the data rate now exceeds storage capacity. To address this, researchers are moving intelligence closer to the detector so that only the most relevant signals are kept.

Key Facts The Department of Energy’s High Energy Physics program awarded ORNL a three‑year grant for the NEUROPix initiative. The team will use spiking neural networks—a neuromorphic computing method modeled on the human brain—to process particle‑collision data in real time at the detector. Benoit said the goal is to sort or compress data quickly while preserving the information that matters most.

What It Means By embedding AI chips in the detector itself, the project aims to reduce latency and storage demands, enabling faster identification of rare particle interactions. Over the next three years, the ORNL group will integrate the neuromorphic chips into prototype pixel modules and evaluate their performance in test beams. Success could lead to broader adoption of on‑detector AI in future high‑energy physics experiments.