Triggering systems for high-granularity detectors in high-energy and high-luminosity environments – HiGranTS
Real-time image processing for the future high-energy particle detectors
Triggering systems for high-granularity detectors in high-energy and high-luminosity environments
Developing real-time algorithms for the processing of 3-dimensional images of high-energy particles collisions
New detector technologies are being developed for future particle colliders, and in particular for the upgraded version of the Large Hadron Collider (LHC), called the High-Luminosity LHC. Among these new detectors, a particularly innovative type of detectors are called high-granular calorimeters. These detectors are recording images of particle collisions in three dimensions containing several millions of pixels at a rate of 40 million images per second. Given the high rate and large size of these images, a real-time filtering and selection of interesting collision images is needed. The goal of this project was to develop algorithms capable of performing this task, and to investigate the utilization of artificial intelligence algorithms in this context.<br />The main challenges of the project were the large amount of data the developed algorithms have to process and the fact that this processing needs to be done in real-time. It implied significant efforts on studying and understanding the impact of data compression and algorithm size reduction on the quality of the collision images selection. It resulted in the development of efficient algorithms needing very few processing resources.
The systems on which those real-time processing algorithms run are not made of standard computers, but of custom electronics chips. Their development therefore requires very specific tools and expertise in order to map them onto this hardware. A combination of such tools has been used to develop these algorithms. In particular, recent tools that automate the development process, called high-level synthesis tools, have reduced the development time significantly. In parallel a dedicated simulation of the same algorithms has also been written to study their performance.
The simulation of the algorithms has been run on the worldwide computing grid used by the LHC experiments, and GPUs have also been used to train artificial intelligence algorithms. The implementation of the algorithms have finally been tested on dedicated electronics chips.
During this project it has been shown that complex algorithms, including artificial intelligence algorithms, could be run for the real-time filtering of collision events. But it was found that they require specific compressions and formatting of their input data coming from the detectors. They also require to be carefully optimized in order to reduce their size.
This project allowed to strengthen existing collaborations with several institutes around the world. These collaborations will continue in the future.
While the focus so far has been put on the exploration of new ideas, we are now in a position where we can work on the construction of an actual system for future highly-granular calorimeters.
In the context of this project, four presentations have been made in different national and international conferences, dedicated to particle physics detector development, to electronics for high-energy physics and to artificial intelligence for high-energy physics. Associated with these communications in conferences, two written articles have been published in peer-reviewed journals.
New calorimetry techniques developed for future high-energy and high-luminosity accelerators are providing more granular detectors that give access to a 3D view of particle showers. The amount of data produced by such detectors is enormous and raises new challenges for the trigger systems.
In addition, the information from the inner trackers will be included in the first level (L1) of these systems, providing the possibility to develop so-called Particle Flow algorithms already at the electronics level of the trigger.
The objective of this project is to develop and implement innovative event reconstruction techniques for L1 trigger systems, based on highly granular calorimeters coupled with trackers. This relies on the resolution of technological obstacles in several points of the trigger chain in order to ensure that the best trigger decisions are made.
Major discoveries in high-energy physics always relied on the development of innovative detectors and data acquisition techniques. These technologies had also numerous applications in various domains (e.g. health, energy). New calorimetry techniques developed for future high-energy and high-luminosity accelerators are providing more granular detectors that give access to a 3D view of particle showers. This is in particular the case for the calorimeters which are being developed for the very high luminosities at the LHC (HL-LHC). The fine segmentation of these calorimeters is a powerful tool to reconstruct very busy collision events produced in such colliders, made of the products of more than a hundred proton-proton interactions (pile-up). But the amount of data produced by such detectors is enormous and raises new challenges for the trigger systems that need to transfer and process these data in the most effective way. The architecture of these systems and the algorithmic techniques implemented need to be completely redesigned to make use of this unprecedented data flow.
In addition, more global pictures of the collision events are also necessary at trigger level to maintain an efficient selection of interesting physics events at high luminosities. This is why the information from the inner trackers will be included in the level-1 (L1) trigger systems of the ATLAS and CMS experiments for the HL-LHC, providing the possibility to develop so-called Particle Flow algorithms already at the electronics level of the trigger system. The topologies of interesting collision events need to be identified rapidly despite the extremely harsh environment induced by the pile-up of more than a hundred of collisions. There are currently no algorithms that can identify electrons, photons, tau leptons and hadron jets in 3D calorimeters and trackers, within the time window available in L1 trigger systems.
The objective of this project is to develop and implement innovative event reconstruction techniques for L1 trigger systems, based on highly granular calorimeters coupled with trackers. These techniques will allow to make use of the full potential of the upgraded detectors for the HL-LHC, such as the CMS new highly granular endcap calorimeters (HGCal) and track trigger. This relies on the resolution of technological obstacles in several points of the trigger chain in order to ensure that the best trigger decisions are made.
Monsieur Jean-Baptiste Sauvan (Laboratoire Leprince-Ringuet)
The author of this summary is the project coordinator, who is responsible for the content of this summary. The ANR declines any responsibility as for its contents.
LLR Laboratoire Leprince-Ringuet
Help of the ANR 295,777 euros
Beginning and duration of the scientific project: October 2018 - 48 Months