Using AI to capture the invisible in dark matter detection
Los Angeles CA (SPX) Jul 30, 2024 -
In the underground tunnel of the world's largest particle collider, the Large Hadron Collider (LHC), beams of protons collide at nearly the speed of light, creating conditions reminiscent of the Big Bang. These collisions, occurring 350 feet beneath the France-Switzerland border, produce subatomic debris that could hold clues to the universe's "missing matter."Duke University physicist Ashutosh Kotwal is on the hunt for dark matter, an elusive substance five times more abundant than ordinary matter but still shrouded in mystery. Utilizing a novel design detailed in the May 3 issue of Scientific Reports, Kotwal hopes to capture evidence of dark matter using artificial intelligence.
Ordinary matter, the visible substance forming people and planets, is only a fraction of the cosmos. Dark matter, invisible and detectable mainly through its gravitational influence on stars and galaxies, remains largely unknown. Scientists at the LHC are striving to change this by using detectors that function like giant 3D digital cameras, continuously capturing the particle sprays resulting from proton-proton collisions.
Typically, only ordinary particles activate the detector sensors. However, scientists hypothesize that dark matter might manifest as heavy charged particles that travel a short distance from the collision point and then decay into dark matter particles, leaving no trace. This disappearance would create a distinctive "disappearing track" in the detector's inner layers.
To identify these tracks, rapid detection is crucial. The LHC's detectors generate about 40 million particle snapshots per second, an overwhelming amount of data to process. Kotwal likens the challenge to finding a needle in a haystack, noting, "Most of these images don't have the special signatures we're looking for. Maybe one in a million is one that we want to save."
Researchers have only a few millionths of a second to determine the significance of each collision and decide whether to store it for further analysis. "To do that in real time, and for months on end, would require an image recognition technique that can run at least 100 times faster than anything particle physicists have ever been able to do," Kotwal said.
Kotwal's proposed solution is a "track trigger," a swift algorithm designed to identify and highlight these fleeting tracks amid tens of thousands of concurrent data points. This system divides the task of analyzing each image across numerous AI engines operating simultaneously on a silicon chip, processing images in less than 250 nanoseconds and filtering out irrelevant data.
This approach, initially described in two papers from 2020 and 2021, was demonstrated in the May issue of Scientific Reports to operate effectively on a silicon chip. Kotwal and his undergraduate student team plan to develop a prototype by next summer, with the full device, comprising approximately 2000 chips, expected to be ready for installation at the LHC detectors within three to four years.
As the LHC's performance increases, producing more particles, Kotwal's device could be vital in ensuring that dark matter detection does not go unnoticed. "Our job is to ensure that if dark matter production is happening, then our technology is up to snuff to catch it in the act," Kotwal said.