Transcript transparencies

Level 0 Decision Unit
Interface
7 May 2003
R. Cornat, J. Laubser J. Lecoq, R. Lefèvre, P. Perret
LPC Clermont
L0DU Architecture
7
8
2
3
: 20 words of 32 bits
At 40 MHz it
receives in parallel:
CALO:
e,,0L,0G,h,ET,Nspd
Muons:
8 cand. 
Vertex:
pile-up

640 bits @ 40MHz
OUTPUT:
90 bytes@1 MHz
RS, L1, DAQ (common L1)
Which kind of links:
copper/optical
Our philosophie was to use the simplest, the more robust and
the cheapest solution:
(ethernet cables with RJ45 connectors)
LVDS @ 40 MHz … but too many links (160) !
2 other solutions:
 serial LVDS
 optical links
(serialisation   4 extra cycles to add in the L0 latency)
Serial LVDS
LVDS 21:3 (DS90CR215/6)
4 pairs/package  2 packages/candidates  40 cables
Size: package: 10x15 mm2
connectors: 1/package RJ45: 10x15 mm2
Cost: 40 CHF/word - For L0DU (5 units) 1300 CHF
 Possibility to use LVDS 48:8 (DS90CR483/4)
9 pairs/pack. could be 8 pairs/pack. 1pack/cand. 40 cables
Size: pack: 20x20 mm2
Cost: similar
Connectors:
If needed smaller connectors can be used
16 RJ45 connectors (8 words) fit on the front of a 6U board
It exists more compact connectors (more expensive):
z-pack connectors 8 cm for 8 words
Cables:
Shielded cables and differential pairs are used
Distance has to be lower than  20 meters
(tested by Orsay up to 22m @60 MHz)
Optical link
Serialisers and deserialisers: TLK2501 (used by L0μ and common L1)
16 bits @ 80 MHz
Optical transeivers: HFBR-5921L Transmitter/Receiver
 1 pack/ candidate, 1fiber/candidate
+ 1 crystall oscillator + 1 PLL+VCXO per board
Size: Ser/Des: 15x15 mm2 Trans: 15.3x49 mm2
connector:
Cost: 310 CHF/word - For L0DU (5 units) 15200 CHF
+ test bench + lab equipment!
 Possibility to use ribbon cables and a 12 channel transceiver
(L0μ solution): cost similar
Comments
Noise and EMC aspects:
LVDS links are fully differential and then completly immune
against common mode voltage fluctuation (for ex. ground)
provided they will be compatible with the power supply of the
chips … which is MANDATORY in any case: The ECS ethernet
links, for instance, have the same needs!
Differential links are immune against all common mode noise
Optical links are more sensitive to jitter (which is only due to
effect of the noise on the digital threshold …)
RJ45 connectors are more robust than optical connectors
A 80 MHz clock is needed for optical link
Optical cost development is higher
Optical can go to longuer distance (>20 m if needed)
Conclusion
 Copper and optical solution can be used
(Both will be used and will work for other applications in LHCb)
 Optical solution is more expensive
(We agree that this is the overall cost which has to be minimised!)
 We don’t want to force people to use LVDS
 If every body prefer optical we will use it
IF we have the money for the extra cost: to
built L0DU, its test bench and specific equipment
 Next meeting in Clermont: L0DU workshop
May 15 ?