SVT specifications
(Marvin Johnson, 7 March 1997
SOME DESIGN CONSIDERATIONS FOR THE SILICON VERTEX TRIGGER
(Marvin Johnson)
Parameters defined by other parts of the DØ system:
- Input data obtained from the silicon by splitting the optical cable.
This means that :
- Data format is that of the SVX II chip plus port card number.
- Data input rate is 53 Mhz
- Serial format requires 850 nm receiver and HP Glink chip.
- Average readout time is around 7 micro seconds
- Level 1 input from fiber trigger is the same as the muon format
- 6 tracks per sector
- Momentum ordered with highest first
- Fiber number of the outer inner barrel
- Sign of the track
- If L2 fiber data is needed, it will only be available by splitting the
optical signal; conditions as in item 1 above.
- System will be located on the second floor of MCH (no other space)
- Each trigger crate will be a geographic sector and will need to
conform to DØ specifications for geographic sectors.
This includes communication with the trigger framework for trigger
information.
- Data transfer to L2 will be by a specification that is now being finalized.
- Data transfer to L3 will be via a VBD.
- Total processing time must be less than 50 microseconds
including data readout from SVX II and formatting for L2 Global.
- Although not absolutely required, each trigger board should process
a multiple of 4 readout fibers. The port card and VRB both are 4 fiber
modules.
- Modules must support VME 32.
Some design suggestions (not required)
- The optical receiver design is done for the readout board so it could
be copied.
- preliminary design indicates that cluster finding can be done at 53 Mhz
with an FPGA. This would eliminate storing all the data and then processing
it. (see DØNote 3168)
- it appears to be possible to do all the track fitting in commercial
DSP boards.
- Selecting data in a fiber road can be done with one FPGA for 4 fibers.
- VME 64 can do 30 or more Mbytes/second so it is a good candidate for a bus
Some questions:
- How to arrange fiber trigger signals relative to Si signals?
Silicon is 6 fold symmetric. CFT trigger divided into 80 sectors. The fibers
determine the roads so one needs to include all the silicon under a group of
fiber roads.
Would like to have Si trigger divided into sectors
so that sector fits into one crate; at 4 fibers per chip, estimate that about
1 sixth of the total fit into one crate , i.e. fiber trigger has 14 sectors per silicon
sector so maximum number of trigger tracks is 96.
-
How many tracks need to be processed per 4 fiber group?
(assume 16 maximum, based on study by Jerry Blazey)
- How long does the fitting algorithm take in a fast DSP?
- How many hits are in each road?
This sets the speed required for VME transfers if one goes in this direction.