CMS ESR December 2002 1
HTR Status and Clocking IssuesD. Baden, T. Grassi
http://www.physics.umd.edu/hep/esr_dec_2002.pdf
HCAL Trigger ReadoutHCAL Trigger ReadoutHCAL Trigger ReadoutHCAL Trigger Readout
CMS ESR December 2002 2
Shield Wall
SBS
HPD
FE MODULE
12 HTRs perReadout Crate,2 DCC
FRONT-ENDRBXReadout Box (On detector)
READ-OUT CrateTrigger Primitives
Fibers at 1.6 Gb/s3 QIE-channels per fiber
QIE
QIE
QIE
QIE
QIE
QIE
CC
A
GOL
DCC
TTC
GOL
CC
A
HTR
HTR
CAL
REGIONAL
TRIGGER
32 bits@ 40 MHz
16 bits@ 80 MHz
CC
A
S-Link: 64 bits @ 25 MHz
Rack CPU
FE/DAQ ElectronicsFE/DAQ ElectronicsFE/DAQ ElectronicsFE/DAQ Electronics
CLK
HTR
CMS ESR December 2002 3
HTR Principal FunctionsHTR Principal FunctionsHTR Principal FunctionsHTR Principal Functions
1. Receive front-end data for physics running
• Synchronize optical links
• Data validation and linearization
• Form TPG’s and transmit to Level 1 at 40 MHz
• Pipeline data, wait for Level 1 accept• Upon receiving L1A:
• Zero suppress, format, transmit to the concentrator (no filtering)
• Handle DAQ synchronization issues (if any)
2. Calibration processing and buffering of:
• Radioactive source calibration data
• Laser/LED calibration data
3. Support a VME data spy monitoring
CMS ESR December 2002 4
Readout VME CrateReadout VME CrateReadout VME CrateReadout VME Crate
“BIT3” board – Slow monitoring over VME– Commercial VME/PCI Interface to CPU
FanOut board– Takes TTC stream in– Clone and Fanout timing signals
HTR (HCAL Trigger and Readout) board– Spy output over VME– FE-Fiber input– TPG output (SLBs) to CRT– DAQ/TP Data output to DCC
DCC (Data Concentrator Card) board– Input from HTRs– Spy output– Output to DAQ
DCC
VME CRATE
20m Copper1.2 Gb/s
DAQ
Calorimeter Regional Trigger
BIT3
Fiber1.6 Gb/s
FanOut
HTR
Front End Electronics
HTR
DCC
HTR
HTR
...
TTC fiber
CMS ESR December 2002 5
• Board organized around 2 identical sets of circuitry:
• Optical inputs• 1.6 GHz, 8B/10B frames, 3ch/link• Dual LC detectors and drivers
• TI TLK2501 Deserializers• Crystal RefClk• TTC 80MHz backup
• Xilinx Virtex FPGA XCV1000E• 24 channels each
• TPG signals• Sent to SLB over backplane,
LVDS• SLBs mounted 6 to a transition
board• Level 1 accept output to DCC
• LVDS output
• VME• Altera FPGA and firmware
OLD DESIGN
““Old” HTR Design Old” HTR Design (Summer 2002)(Summer 2002)
““Old” HTR Design Old” HTR Design (Summer 2002)(Summer 2002)
CMS ESR December 2002 6
HTR Functional ExperienceHTR Functional ExperienceHTR Functional ExperienceHTR Functional Experience
• What was tested:• VME fully tested and working
• Some changes necessary to conform to CMS VME standards• Optical links and synchronization
• No indication of any problems. Big success here – was a real worry
• LVDS to DCC• Tested, working (Will change cable/connector to Cat 6/RJ45)
• Fanout of timing signals on two Cat5 cables• Plan to change to a single Cat6 or Cat7 cable (very low cross-talk)
• Firmware – full tests of:• Pipeline and L1A triggering capability• In-line histogramming for source calibration
• TTCrx• Not working at all (4 bad on 4 tested).
• What was not tested: Anything to do with TPG
CMS ESR December 2002 7
HTR Board ExperienceHTR Board ExperienceHTR Board ExperienceHTR Board Experience
• Produced ~12 boards• Several bare boards were delivered warped
• Many opens under FPGA after assembly (~9 boards)
• Some fixed after reflow (a few)
• Some worse after reflow (shorts)
• X-rayed a few boards, sometimes inconclusive
• Some opens on VME side
• Non BGA FPGA, indicates bad vias
• Few other various open circuits
• Finally got ~8 boards to “work”• Questionable reliability
CMS ESR December 2002 8
ModificationsModificationsModificationsModifications• Change board from using white-tin to gold traces
• This process was sold to us by the board maker. Our mistake.• Used only for very high volume, cost competitive products, very
difficult and expensive to control. • Gold is flatter and not very much more expensive (~$50/board),
better for FPGAs
• Change assembly house• Insufficient Quality Control on current assembler – they are fired.• We visited 2 high-end assemblers
• Modern Machines • Step up and step down oven temp control.• In-line X-ray for BGA QC• Manufacturability Review
• Add stiffeners to HTRS• Flexability of 9U VME boards was underestimated
• Worry: fine-line BGA (FBGA) can pop connections
CMS ESR December 2002 9
Modifications (cont)Modifications (cont)Modifications (cont)Modifications (cont)
• Change from FBGA to BGA• FBGA 1.0mm pitch, change to BGA 1.27mm pitch
• No additional expense, plenty of available real estate, no need to push
• We are just being very conservative here
• JTAG capabilities added• Will help with debugging
• By making these changes…• We have profited from the summer
• We have reduced our production risk considerably
CMS ESR December 2002 10
HTR Design ChangesHTR Design ChangesHTR Design ChangesHTR Design Changes• SLB transition board issues:
• Worries about so many LVDS signals over backplane for old design• Routing is too complicated
• Many signals going to same backplane location• Requires multi-layer routing with many vias
• TPG cables very thick• Mechanical issues are very worrisome
• SLB changes needed (e.g. height reduced after ECAL redesign…)• Solution: move SLB’s to HTR motherboard
• Benefits:• Mechanically attach SLB’s to HTR front panel for mechanical stability• Eases routing requirements, reduces board and assembly risks, cheaper too
• Change from Xilinx VirtexE to Virtex2• More resources, block ram, hardware multipliers• Big cost reduction (save $180k)• More modern chip for long-term maintenance
• Clock synchronization• Decouple “80MHz” crystal from FPGA system clock• Will allow us to use crystal to maintain synchronization of serdes
• This gives us 2 solutions for our “40ps” jitter requirement issue
CMS ESR December 2002 11
P1
to D
CC
New HTR Conceptual Design New HTR Conceptual Design New HTR Conceptual Design New HTR Conceptual Design
P2
LVDS
to L
evel
1 C
al T
rig
ger
LVDS
SLB
SLB
SLB
SLB
SLB
SLB
FPGAXilinxXC2V
LCLC
TITI
TI
TI
TI
TITI TI
LCLC
FPGAXilinxXC2V
LCLC
TITI
TI
TI
TI
TITI TI
LCLC
VMEFPGA
Fibers
No
P3!
8-way8-way
fro
m
Fan
ou
t
RJ45TTC
CMS ESR December 2002 12
Clocking SchematicClocking SchematicClocking SchematicClocking Schematic
TTCrxTTC
80 MHz LVPECL Crystal 1 to 8
Fanout
1 to 8 Fanout
Single width VME
BC0
80MHz
40MHz
SLB
SLB
SLB
SLB
SLB
SLB
TI(16)
FPGA
BC0BC0
40MHz 1 to 8 Fanout
80MHz
TTC mezzTTC TTC broadcast bus
Cat 6/7 quad cable (allows PECL)
TTC Fanout Board
QPLL
• Start with Fanout card• TTCrx Maryland mezzanine card or CERN TTCrm daughterboard• QPLL• Fanout on Cat6/7 quad twisted pair TTC, BC0, 40MHz, 80MHz
• In HTR:• Send TTC signal to TTCrx mezzanine board, access to all TTC signals• Send 80MHz clean clock (cleaned by QPLL) to mux
• Select 80MHz clean clock OR crystal to TI deserializers
80 MHz
40 MHz
CMS ESR December 2002 13
HCAL TRIDas Clock SchemeHCAL TRIDas Clock SchemeHCAL TRIDas Clock SchemeHCAL TRIDas Clock Scheme
TTCrx QPLL
(‘CC’ means Clean Clock)
Cat6/7 RJ45
RJ45 TTCMezz
TTC
SLB
Xilinx
TTC broadcast, L1A, BCR, EVR, CLK40
Fanout Card
4 twisted pair…
TTC BC0 CC40CC80
HTR Board
CC40
CC80
BC0
CMS ESR December 2002 14
Fanout – HTR schemeFanout – HTR schemeFanout – HTR schemeFanout – HTR scheme
HTR
TTC fiber
TTC LVDS
CLK803.3V-PECL
RX_BC0 LVDS
Cat6Eor Cat7cable
8 clks to TLKs
DS90LV001
Q1Q2Q3Q4Q5Q6Q7Q8
MC100LVE310 3.3V PECL
CLK403.3V-PECL
LVDSFanoutx 8
PCK953LVPECL-to-LVTTLFanout(top layer)
PCK953LVPECL-to-LVTTLFanout(top layer)
8 clks to TLKs + TPs
To 6SLBs
Diff. to 2Xilinx+ termin.
Diff. to 6 SLBsSingle-end to 2 xilinx
TTC daughter
card
ININ_b
Brdcst<7:2>,BrcstStr, L1A,BCntResto xilinxand SLBs
CLK80 LVDS
FanoutBoard
Low-jitterFanout x 15
O/E
Brdcst<7:2>,BrcstStr BC0
Fanoutbuffer
TTCTTC
TTC
FPGA
Fanout x 15
Brdcst<7:2>,BrcstStr,BCntRes,L1A
CMOSLVDSor diffPECL
……..……..
……..……..
15 connectorson bottomlayer ?
15 Cables &Connectorstbd
……..……..
NB100LVEP221is LVDS compatible
TTCrx
(or daughter card)
QPLL
AN1568/D Fig 11Onsemi.com
RJ45
~FifteenRJ45connectors
PECLfanout
e.g. DS90LV110
..
..
2 TestPoints forCLK40andBC0
CLK40 LVDS
PECLfanout
..
..
..
..
80.0789 MHz3.3V crystalDiff. PECL
MC100LVEL37
CKCKCK/2CK/2
..
..
..
..
9U Front-panel space = 325 mm ; => space per connector ~ 21.5 mm
Notes: SLBs require fanout of CLK40, BC0. FE-link possibly requires CLK80. PECL fanout was tested in TB2002. One Cat6E cable (low x-talk) replaces the 2 Cat5 cables used in TB2002. TTC and BC0 remain LVDS as in Weiming’s board.HTR needs Broadcast bus, BCntRes and L1A: from TTCrx if we get it to work, otherwise we have to fan them out.
LVDS
Tullio Grassi <[email protected]>
CMS ESR December 2002 15
TTCrx Mezzanine cardTTCrx Mezzanine cardTTCrx Mezzanine cardTTCrx Mezzanine card
• Very simple card:• 2 PMC connectors
• TTCrx chip
• TTC signal driver on motherboard
• Will be sent out for prototype next week
• Used by HTR, DCC, and Fanout cards
CMS ESR December 2002 16
TTC Distribution – Fanout CardTTC Distribution – Fanout CardTTC Distribution – Fanout CardTTC Distribution – Fanout Card
• Currently HCAL has 6 TTC partitions:• Each partition requires TTCvi and TTCex
• Each HCAL VME crate will have a single TTCrx receiving data directly from TTCex in a single VME card (Fanout Card)
• Fanout TTC signal to HTR mezzanine card with TTCrx chip
• Use quad twisted pair CAT6/7 cable allows PECL fanout
• TTC raw, BC0, 40MHz clean, 80MHz clean fanout
• Cost savings and simplification
• TTC monitoring by Fanout card over VME
• Count resets, etc…
CMS ESR December 2002 17
Random Latency IssueRandom Latency IssueRandom Latency IssueRandom Latency Issue
•Texas Instruments TLK2501 Serdes• Run with 80MHz frame clock – 20 bits/frame, 1.6GHz bit clock
• 625ps bit time
• Latency from data sheet:
• ~20ns variation (overall latency between 47 and 67ns)
• Fiber to fiber alignment could cross a 40MHz bucket boundary.
•How to fix? • SLB “knows” this latency – we will read it out after each reset
• HCAL LED fast rise time
• Can pulse during abort gap and align channels
• Requires LED pulsing alignment
• FE will send BC0 signal on all fibers
•Will measure this alleged latency with new HTR boards
CMS ESR December 2002 18
TPG AlignmentTPG AlignmentTPG AlignmentTPG Alignment
• TPG alignment performed in SLB• Necessary: All HTRs will send common BC0 to SLB’s
within each of 16 VME crates
• Calibration procedure to be performed for crate-crate alignment
• Initial alignment with LEDs, laser, etc.
• Final alignment with LHC first beam data
• Use “1-bucket” running to check everything
• This will ensure successful alignment
CMS ESR December 2002 19
DAQ AlignmentDAQ AlignmentDAQ AlignmentDAQ Alignment
• DAQ data must also be aligned• Must know L1A bucket for zero suppression
• Only if we will do ZSP on 1 or 2 HCAL channels centered on L1A bucket
• If ZSP done with sum over 5 channels, then this alignment is not critical
• Solution: discussed in previous slide• Read from SLB• FE sending known ID after with fixed offset relative to
BC0 during abort gap• Comparison of the two for error checking
• DAQ check on BC0 in DCC for alignment• Will send BC0, BCN, and EVN with the data to DAQ
CMS ESR December 2002 20
MISC ErrorsMISC ErrorsMISC ErrorsMISC Errors
• What happens if DCC finds mismatch in EVN?• DCC will then issue resynch request to aTTS system
• Details not yet defined but is fully programmable
• Fiber Link/synchronization errors (GOL/TI)• Work out protocols to inform DCC
• Reset requests to aTTS as well
• FE Clock/GOL PLL link errors• If GOL loses synch, then transmitter will send out IDLE
characters
• IDLE characters are illegal in a pipelined system!
• HTR will trap on IDLE as a signal that FE/GOL is having trouble
CMS ESR December 2002 21
ScheduleScheduleScheduleSchedule
O N D J F M A M J J A S O N D J F M A M J J A S O N D J F M A M J J A S O N DFirmwareBoard layoutFab/assembly 20 boards will be built but not assem bled Pre-production HTR boardCheckoutBoard layout if needed
Fab/assembly if needed Production prototypeCheckoutProductionTestbeam ?Vertical Slice ?
2002 2003 2004 2005