Associative memory computing power and its simulation

An important step in the ATLAS upgrade program is the installation of a tracking processor, the Fast Tracker (FTK), with the goal to identify tracks generated from charged particles originated by the LHC 14 TeV proton-proton collisions. The collisions will generate thousands of hits in each layer of the silicon tracker detector, making the track identification a very challenging computational problem. At the core of the FTK there is the associative memory (AM) system, made with hundreds of AM ASICs chips, specifically designed to allow pattern identification in high density environments at very high speed. This component is able to organize the computational steps of the track identification providing a huge computing power for a specific application. The AM system will in fact be able to reconstruct tracks in 10s of microseconds. Within the FTK team there has also been a constant effort to maintain a detailed emulation of the system, to predict the impact of single component features in the final performance and in the ATLAS data acquisition system. The FTK emulation is however a demanding software, we describe the efforts to have the best performance using commercial computing devices and some ideas for the future evolution.