310 likes | 404 Views
CMS: GCT2: Concentrator Card Pre-Manufacture Review. 11th August 2006 version 2 (Draft). Last processing stage before calorimetry data sent to Global Trigger 9U VME64x card 2 DPMCs mounts for electron-leaf cards 2x3 Samtec cables to interface to jet-wheel cards
E N D
CMS: GCT2: Concentrator CardPre-Manufacture Review 11th August 2006 version 2 (Draft) CMS: GCT2: Concentrator Card: Pre-Manufacture Review (gregory.iles@cern.ch)
Last processing stage before calorimetry data sent to Global Trigger 9U VME64x card 2 DPMCs mounts for electron-leaf cards 2x3 Samtec cables to interface to jet-wheel cards 1 DPMC mount for Global Trigger interface What is the concentrator ? CMS: GCT2: Concentrator Card: Pre-Manufacture Review (gregory.iles@cern.ch)
Trigger All paths 40 MHz DDR -> 80MHz Electron η+ data from Leaf DPMC Electron data 2 x 160 Single Ended Via J11, J12, J21, J22 1) Iso Elec 2) Non-Iso Elec 3) Energy Sum 4) Jet Counts Electron η- data from Leaf DPMC Electron V4 FPGA Sorted Et and jet count 2 x 50 Diff Pairs Via Samtec J2 & J3 Global Trigger DPMC 7 x dual channel Serdes links 80 Single Ended Jet η+ data from Wheel Card Leaf Leaf 2 x 180 Single Ended Via fully populated 1/2 DPMC Jet V4 FPGA Leaf Jet η- data from Wheel Card Leaf Sorted & unfinished jets 2 x 150 Diff Pairs Via Samtec J1 & J3 5) Forward Jet 6) Central Jet 7) Tau Jet Leaf Leaf CMS: GCT2: Concentrator Card: Pre-Manufacture Review (gregory.iles@cern.ch)
Control & Readout 40 MHz DDR -> 80MHz Slink VME TTCrx Clock Control FMM USB Ethernet 1) Iso Elec 2) Non-Iso Elec 3) Energy Sum 4) Jet Counts Electron η+ data from Leaf DPMC 2 x 40 Single Ended Via J23 Electron η- data from Leaf DPMC Electron V4 FPGA Comm V2 FPGA 2 x 32 Single Ended Jet η+ data from Wheel Card Leaf Jet V4 FPGA Leaf 2 x 34 Diff Pairs Via Samtec J2 V2 driving LVDSEXT Leaf 5) Forward Jet 6) Central Jet 7) Tau Jet Jet η- data from Wheel Card Leaf V2 V4 Leaf 100 100 DCI LVDS requires 62.5mW per pair Leaf CMS: GCT2: Concentrator Card: Pre-Manufacture Review (gregory.iles@cern.ch)
Processing Two Xilinx Virtex4 FPGAs XC4VLX100-FF1513 Must concentrate large amount of data Choose package with most I/O Integrated differential termination makes layout simpler High speed I/O provide reserve capability Communication Xilinx Virtex2 FPGA XC2V3000-BF957 Robust in 3.3V enviroment VME 64x interface Slink TTCrx Ethernet PHY & USB for future Elec FPGA Isolated Electrons Non-Isolated Electrons Energy Sums Jet Counts Jet FPGA Forward Jets Central Jets Tau Jets Implementation CMS: GCT2: Concentrator Card: Pre-Manufacture Review (gregory.iles@cern.ch)
Top Side Switching power supplies J1: VME DPMC: ElecLeaf (and one on other side) Comm JTAG header +5V Elec Comm J2: VME & Slink Clock Distribution Jet DPMC: GlobalTigger CPLD, System & Misc JTAG headers DPMC +3.3V power Trigger data flow Both sides of board Top side only RJ45-LVDS Out (e.g. for TTS) CMS: GCT2: Concentrator Card: Pre-Manufacture Review (gregory.iles@cern.ch)
Bottom Side DPMC: ElecLeaf (and one on other side) Proms Readout path Duplicated Single DPMC: GlobalTigger USB RJ45-LVDS Out (e.g. for TTS) RJ45-LVDS In (e.g. for TTS) Clock distribution Duplicated Single Ethernet CMS: GCT2: Concentrator Card: Pre-Manufacture Review (gregory.iles@cern.ch)
Layers = 14 7 signal 50 ohm single ended track width 100 ohm differential track width/gap 2 ground 5 power Additional layers May need at least one extra power & ground layer ExceptionPCB in UK 20 layers possible Need to look at stack up to verify. Board Specifications (I) CMS: GCT2: Concentrator Card: Pre-Manufacture Review (gregory.iles@cern.ch)
Vias Specs Min drill diameter = 0.3mm Through via layers 1 to 14 depth = 2.45mm aspect ratio = 8.2:1 Blind Top via layers 1 to 8 depth = 1.3mm Blind Bottom via layers 9 to 14 depth = 0.8mm Thickness = 2.45 mm Recommended not to exceed this otherwise Through hole component lead length too short Aspect ratio on vias too large Card edges Board will be milled down to 1.6mm Should this be 2mm ? Board Specifications (II) CMS: GCT2: Concentrator Card: Pre-Manufacture Review (gregory.iles@cern.ch)
Input power From ECAL backplane, but with +5V. 3 connectors (3M:MP2-SP10-51M1) Each connector has 5 pins power, 5 pins ground 6.5A per pin => 32.5A per connector At present 2 of the 3 are used to supply 5V 65A or 325W 1 connector is spare. Power Supply (I) CMS: GCT2: Concentrator Card: Pre-Manufacture Review (gregory.iles@cern.ch)
Clock System Driven by linear regulators 2.5V, 0.5A from LT1763CD QPLL 3.3V, 3A from LT1764E TTCrx, SN65LVDS125, SN65LVDS125 Dual PMC sites Leaf cards & GlobalTrigger card Supplied by either: Datel LSM-10A switcher TPS75933 linear, 3.3V, 7.5A Both situated under PMC site. Not ideal because Height of switcher is 9mm Power dissipation of linear 5A from 5V = 8.5W Could easily add temp sensor to exiting I2C chain. FPGAs and rest of board Main power from 6 Datel LSM-10A switchers 3 of the switchers provide core volatges for each FPGA 2 x 1.2V 1 x 1.5V 3 of the switchers provide I/O power to the FPGAs 2 x 2.5V 1 x 3.3V FPGA Proms are powerwed from LT1763CD, 1.8V, 0.5A Power Supply (II) CMS: GCT2: Concentrator Card: Pre-Manufacture Review (gregory.iles@cern.ch)
JTAG Chain Header Misc TTCrx Slink Ethernet Header System Header CPLD V4 & PROM: Jet V4 & PROM: Elec Voltage Error VME CPLD DPMC: ElecPos Header Comm DPMC: ElecNeg Control WheelPos DPMC: GlobalTrig WheelNeg Red = 3.3V Blue = 2.5V PMC: Spare x8 CMS: GCT2: Concentrator Card: Pre-Manufacture Review (gregory.iles@cern.ch)
Clock Distribution (I) Destinations: 3 x FPGAs 3 x DPMCs 2 x Wheel Cards Destinations: 1 x PMC 1 x two 50 ohm SMB 1 x header 0.1” dual Comm FPGA SN65LVDS125 Cross-Point Switch CLK_DIFF_A (CLK40) CLK_DIFF_B (CLK80) CLK_DIFF_C (CLK80) CLK_DIFF_D (CLK40) CLK160 TTCrx QPLL In: 0 Out: 0 SN65LVDS108 CLK40DES1 CLK80 In: 1 Out: 1 SN65LVDS108 Header Dual 0.1” CLK40 In: 2 Out: 2 SN65LVDS108 Header Dual 0.1” In: 3 Out: 3 SN65LVDS108 CMS: GCT2: Concentrator Card: Pre-Manufacture Review (gregory.iles@cern.ch)
At present everything driven by 40MHz & 80MHz QPLL clock, including VME. Comm FPGA cannot configure the clock without leaving itself vunerable. The QPLL has searche mode, which may force DCMs to unlock. Solution: Plan to hard code QPLL and Cross-Point Switch configuration. No DCMs Backup: Use 40MHz utility clock for VME. Then bridge to QPLL clock domain. Clock traces are at present not length matched Suggest that at least the 6 traces to the Comm, Jet and Elec FPGA are length matched to within an inch. Would need an extra power/gnd layer Extra layer would also help routing around Samtec bolt holes. An extra layer would allow length matching up onto the leaf cards. Clock Distribution (II) CMS: GCT2: Concentrator Card: Pre-Manufacture Review (gregory.iles@cern.ch)
Interfaces (not VME) • RJ45 • Require an RJ45 with LVDS outputs for Trigger Throttle System • Added extra capacity • 2 x RJ45-LVDS outputs • 2 x RJ45-LVDS inputs • USB & Ethernet • USB uses Cypress CY7C68001 • Wrong connector at present (Type A rather than Type B) • Ethernet Uses Intel LXT971A • Both used on Source/IDAQ cards • Can benefit from experience at Imperial College • Both request careful layout. Not yet checked • Slink • Will use ECAL transition card and DCI CMS: GCT2: Concentrator Card: Pre-Manufacture Review (gregory.iles@cern.ch)
Recommended changes Replace sense resistor with fuses on switchers. What about linear and leaf supplies? Can 30A flow through antipads Replace caps with fuses for +5V supply Match length clock traces Move clock outputs (simple) Change USB type A to type B Mounting holes for Samtec Decoupling caps at connectors p3v3, p2v5a, p2v5b, p5v, p1v2b and 3.3V for all DPMC sites CMS: GCT2: Concentrator Card: Pre-Manufacture Review (gregory.iles@cern.ch)
No change recommended Comm FPGA JTAG Hdr in awkward location Switcher 9mm high. Caps on leaf at least 1mm. Max clearance = 15mm. Depnds on leaf Linear dissipating 8.5W in confined space Power supply power up rules for Xilinx devices CMS: GCT2: Concentrator Card: Pre-Manufacture Review (gregory.iles@cern.ch)
Susceptible to power supply noise SerDes power will be provided by local linear regulators Mounted on DPMC If design revision necessary cost and turnaround time should be substantially less. Clock distribution similar to that of Concentrator Card. Cross-point switches receives clk40 and clk80 from concentrator and an exetrnal source Distributed by four ICS83948I_147 ICs Clk 80 sent to SerDes chips Clk40/80 sent to CPLDs Global Trigger Card (I) CMS: GCT2: Concentrator Card: Pre-Manufacture Review (gregory.iles@cern.ch)
Global Trigger Card (II) DPMC XC2C256 FT256 DS92LV16 Tx or Rx 42 Cable 0 DS92LV16 Tx or Rx Self contained power from linear regulators LT1764E / 3.3V LT1763CD / 2.5V LT1763CD / 1.8V Instantiate 8 times - 7 for transmit - 1 for receive Provides 2.5V and 3.3V logic level conversion. Also allows us to test DS92LV16 serdes units in loop back mode, albeit one at a time, should we encounter problems. CMS: GCT2: Concentrator Card: Pre-Manufacture Review (gregory.iles@cern.ch)
Testing • Connectivity test • Can test connectivity of ~80% of board with either JTAG or custom firmware. • Samtec connections • Loopback with production cables • PMC sites • DPMC test board (Matt Stettler) • FPGA-FPGA connnections • Insitu tests • VME & Slink etc are probably best tested by final or test firmware • E.g. for VME by writing/reading register many times • Alternative is a dedicated JTAG loopback system. • Time consuming to construct CMS: GCT2: Concentrator Card: Pre-Manufacture Review (gregory.iles@cern.ch)
Schedule & Status • PCB manufacture & assembly in September • Manufacture 3 PCBs • Assemble 1 • At present 1 month behind schedule • Also need GT DPMC card layout and manufacture CMS: GCT2: Concentrator Card: Pre-Manufacture Review (gregory.iles@cern.ch)
Appendix Following slides list the signal counts and how they were obtained CMS: GCT2: Concentrator Card: Pre-Manufacture Review (gregory.iles@cern.ch)
Jet interface Arrives from 2 x wheel cards via high speed Samtec cable assemblies 240 LVDS signals from each Wheel card (40 MHz DDR -> 80Mhz) 200 for trigger path 34 for control & readout 2 for clk 4 for jtag Electron interface Arrives from 2 x leaf dual PMC cards mounted on concentrator card Only 206 out of 360 I/O used) 160 for trigger path 40 for control & readout 2 for clk 4 for jtag Incoming data CMS: GCT2: Concentrator Card: Pre-Manufacture Review (gregory.iles@cern.ch)
Global Trigger interface Tranmit to Global Trigger on 7 cables 2 unidirectional SerDes channels per cable Each channel driven by NatSemi DS92LV16 Takes 16 bit parallel data at 80MHz. Transmits at 1.44 Gb/s Loopback testing possible with 1 extra cable Mounted on Dual PMC All 360 I/O connected Allows relatively fast & cheap modifications if problems exist with high speed serial links Slink to DAQ Signals connect to VME J2 Uses ECAL transition card to host SLINK transmitter card Outgoing data CMS: GCT2: Concentrator Card: Pre-Manufacture Review (gregory.iles@cern.ch)
Jet Trigger: Guide How to understand the next slide 6 x Leaf (3 per Wheel) 2 x Wheel U2 1 x JetFinder & Cntrl Jet 6 clustered jets (12) & Ht(13) H: ~3x160, R: 3x85 What does this mean ? All numbers in “bits” assuming 80 MHz data transmission on single-ended & differential pairs 6 clustered jets (12) & Ht(13) = 6 clustered jets of 12 bits each and 13bits for Ht per leaf card H: ~3x160 = “Have” aprrox 160 bits from each of the 3 leaf cards R: 3x85 = “Require” 85 bits from each of the 3 leaf cards (6x12+13) Spare capacity on bus Bus at limit, although running at double capacity (160MHz data) should be possible in future CMS: GCT2: Concentrator Card: Pre-Manufacture Review (gregory.iles@cern.ch)
Jet Trigger 6 x Leaf (3 per Wheel) 2 x Wheel 1 x Concentrator Et(13), Exy(34) Ht(13) H: 2x80, R: 2x60 Ctnrl (64), Et(13), Exy(34), Ht(13) H: ~3x160, R: 3x124 1) Iso Elec 2) Non-Iso Elec 3) Energy Sum 4) LoopBack cable U2 1 x JetFinder & Cntrl Energy Electron SerDes pair (40) H: 180, R: 160 6 clustered jets (12+2 spare) JetCount(60) H: ~3x160, R: 3x144 H: 160, R: 0 Et(13), Exy(34), Ht(13) Double Compare 3x3 requires 2x3 pre clusters (18) H: 400, R: 168 GT H: 160, R: 0 H: 2x20, R: 2x0 SerDes pair (40) H: 180, R: 160 U1 2 x JetFinder Jet Jet 5) Forward Jet 6) Central Jet 7) Tau Jet 8) JetCounts 12 clustered jets (12+2 spare) JetCount(60) H: ~3x320, R: 3x228 12 sorted jets (14 + 2 spare) JetCnt(60) H: 2x280, R: 2x192 Next Leaf Leave at least 2bits per bus for BC0 (equivalent to 1 signal at 80MHz) Double Compare 3x3 requires 2x3 pre clusters (18) H: 120, R: 108 CMS: GCT2: Concentrator Card: Pre-Manufacture Review (gregory.iles@cern.ch)
Jet data sent to ‘”Jet’” FPGA Top 4 rank of central, forward and tau jets. Hence 12 objects 12 sorted jets (min 14 bits each) 5 bits phi 3 bits eta (no need for sign) 6 bits rank 9 unsorted jets (min 14 bits each) 18 phi regions -> max 9 jets 1 bit phi (each jet covers 2 phi) 0 bits eta (events in the middle) 10 bits Et 1 bit tau veto 2 bits spare Total = 147 signals @ 80MHz 294 bits Available = 150 signals @ 80MHz Jet data sent to “Elec” FPGA Et-total (13 bits) 12+1 bits mag + overflow Et-missing (26 bits) x & y components 12+1 bits mag + overflow Jet counts (36 bits) 6 jet count regions each 5 bit Alternative more flexible system of 12 jet count regions of 3 bits each (not TDR) Ht (13 bits) 12+1 bits mag + overflow Total = 38 + 7 signals @ 80MHz 75 + 13 bits Available = 40 + 10 signals @ 80MHz Signal: From single Jet-Wheel CMS: GCT2: Concentrator Card: Pre-Manufacture Review (gregory.iles@cern.ch)
Electron data sent to ‘”Elec” FPGA Top 4 rank of isolated and non-isolated electrons (min 14 bits each) 8 electron objects (14 bits) 5 bits phi 3 bits eta (no need for sign) 6 bits rank To reduce the latency the FPGAs on the electron leaf card will not share data. Hence each FPGA will send 8 electron objects to the concentrator (i.e. concentrator receives 16 electron objects from each leaf card) Total = 112 signals @ 80MHz 224 bits Available = 160 signals @ 80MHz Signal: From single Elec-Leaf CMS: GCT2: Concentrator Card: Pre-Manufacture Review (gregory.iles@cern.ch)
Data transmitted between V4 FPGAs The 9 unsorted jets on the boundary between the two wheels are turned into clusters in the “Jet” FPGA These jets will contribute to Ht and the jet-counts being summed in the “Elec” FPGA Ht (13 bits) 12+1 bits mag + overflow Jet counts (60) 12 types each 5 bits Total = 37 signals @ 80MHz 73 bits Available = 80 signals @ 80MHz Signal: Between Jet/Elec FPGAs CMS: GCT2: Concentrator Card: Pre-Manufacture Review (gregory.iles@cern.ch)
Readout Max slink sustained rate = 200 MB/s Assume no source generates more than 100MB/s 10bits @ 80MHz Control Serial VME 2bits L1A & BC0 2bits Serial Fast Commands from TTC B channel (e.g. resync) 1bit AsyncReset 1bit Serial FastFeedback 1bit Total Required = 17 signals Minimum available = 32 signals Signal: Control & Readout CMS: GCT2: Concentrator Card: Pre-Manufacture Review (gregory.iles@cern.ch)
Global Trigger interface GT receives 7 cables (2 unidirectional SerDes channels per cable) Each channel driven by NatSemi DS92LV16 Takes 16 bit parallel data at 80MHz. Adds 2 bits. Transmits at 1.44 Gb/s Require 2 bits for powerdown/sync 252 signals (7 x 2 x 18) Require 1 cable for loopback testing Generates 16 bits parallel data Require 4 bits for lock, refclk, powerdown and recovered clk 40 signals (1 x 2 x 20) NatSemi chips do not have JTAG Could use local loopback to test data lines only. Require 2 bits for outenable and local loopback on all chips 44 signals ((14 x 2) + 16 Mounted on dual PMC All 380 I/O connected, Require at least 292 signals, perhaps 336 Slink Signals connect to VME J2 and hence to ECAL transition card Signal: GT interface CMS: GCT2: Concentrator Card: Pre-Manufacture Review (gregory.iles@cern.ch)