Neural Networks in HEP triggers

Paper: 312
Session: F (talk)
Speaker: Nobrega, Ricardo, LIP, Lisbon
Keywords: neural networks, trigger algorithms, trigger systems

Neural Networks in HEP triggers

R. Nobrega
LIP/Lisbon, CERN


Neural Networks (NN) have been widely used as classifiers in physics
analysis, but comparatively few applications in on-line triggers for
high energy physics (HEP) experiments exist. I shall describe two
examples of NN in HEP triggers: a Ring Imaging Cherenkov (RICH)
trigger, proposed for the SQUASH experiment, and an Electron/Photon
trigger at LHC.

In spite of depending on quite different detectors (a gaseous RICH
with a photomultiplier matrix readout in SQUASH versus a crystal
calorimeter in CMS), the trigger algorithms for both projects share a
very similar implementation, thanks to the L-Neuro chips that emulate
the NN.

Prototype boards of the CMS 1st Level Trigger were tested during the
summer of '96. These boards were developed by LIP-Lisbon and
LPNHE-Palaiseau, based on L-Neuro2 chips, and the tests took place at
CERN. A matrix of 7x7 PbWO_4 crystals was irradiated with an electron
beam, and successful operation of the boards at realistic repetition
rates of 40MHz was achieved.

Coping with the huge background rejection ratios present at LHC
implies the use of multi-level triggers. Therefore, a study of a
NN-based 2nd level calorimeter trigger was performed. This trigger has
access to the full calorimeter granularity, reducing by one order
of magnitude the trigger rates after the 1st level trigger.