Machine Learning with FPGAs for real-time characterization of fast scintillation signals
Calorimeters in high energy physics are widely used to measure the energy position and timing of particles.
The Belle II experiment recently introduced a new feature to their CsI(Tl) crystal calorimeter: Offline pulse shape discrimination which is a new technique to identify strongly interacting particles.
Made possible by the recent advances in machine learning (ML) and the availability of powerful FPGAs we want to push this technique one step further and apply pulse shape discrimination not just offline but in real-time.
In this cross-disciplinary project we will bring together Belle II and the SuperCDMS experiment to solve common challenges towards implementing ML algorithms for waveform characterisation on FPGAs.
Performance will be benchmarked using digitised CsI(Tl) scintillator signals from neutron backgrounds in XFEL and cosmic rays.
In the second phase of this project we want to extend real-time pulse shape discrimination to the scintillation timing frontier using the much faster pure CsI crystals.
The project shall seed fast ML on specialized hardware, pulse shape discrimination in a possible Belle II upgrade with pure CsI crystals, and ultimately 6D (position, energy, time, and pulse shape) reconstruction of calorimeters.