Speaker
Description
Fast and efficient processing of data from the tracking detector of the ATLAS experiment is required for the high-luminosity program. The tracking detector is equipped with semiconductor sensors with high-segmentation of about 50 by 50 microns. A charged particle crossing a sensor ionizes a few pixels along its trajectory. Our firmware processes images from the sensors to calculate coordinates of particle crossings. The firmware groups pixels with non-negligible ionization into clusters and then estimates coordinates of the particle crossing by calculating cluster’s centroid. The firmware is programmed into FPGAs that run as CPU co-processors. The FPGAs will receive collision data from the experiment at the rate of about 1 MHz. We benchmark the performance of the clustering firmware interfaces at Vitis kernel and compare it to a clustering algorithm for CPUs. The FPGAs in AMD Alveo U250 cards demonstrate faster processing and energy savings in comparison to the CPUs.
