64k networked multi-threaded processors and their real-time application in high energy physics

Particle physics experiments create large data streams at high rates ranging from kHz to MHz. In a single event the number of created particles can easily exceed 20.000. The architecture of high resolution tracking detectors does not allow to handle the event data stream exceeding 10 TByte/s. Since only some rare scenarios are interesting a selection process increases the efficiency by identifying relevant events which are processed afterwards. This trigger has to be fast enough to avoid loss of data. In case of the ALICE experiment at CERN the trigger is created by analyzing data of the transition radiation detector where about 16.000 charged particles cross six independent layers. Nearly 1.2 million analog data channels are digitized at 10 MHz by 10 bit ADCs within 2 mu s. On this data stream of 13 TByte/s a trigger decision has to be made within 6 mu s. (5 refs).