13-Feb-2004 HCAL TriDAS 1 HCAL Tridas SLHC Drew Baden University of Maryland February 2004
13-Feb-2004 HCAL TriDAS 2 Parameters Just to be sure I’m on the right page... –SLHC Bx = 80MHz –Level 1 will run at 80MHz Vitesse links upgraded TPG data to Level 1 at 2x current speed Same 100kHz L1A rate Questions –What will be the L1 latency? Same latency in time? –We like it but we’d like it to increase! Same latency in clock ticks? –Disaster –Any anticipated changes to TTC? –What will HCAL build anyway?
13-Feb-2004 HCAL TriDAS 3 HCAL in SLHC – FE and Clocking With current technology: –HCAL HB/HE Front-end needs no changes, so… –HCAL could keep same FE boards and same GOL 1.6Gbaud links We already integrate over ~2 buckets for Bx = 40MHz In HTR, associate energy with beam crossing number for TPG and DAQ paths: –Use a simple peak finding algorithm (“up/down” over 3 successive buckets) –More on this later… –HF FE might need some changes (signal is too fast for 40MHz front-end…) HCAL synchronization/clocking scheme –Current clocking scheme: Very versatile, will accommodate current 40MHz running –See Chris Tully’s talk…
13-Feb-2004 HCAL TriDAS 4 Timing signals - Overview Fanout Card (in GLOBAL mode) Fanout Card (in CRATE mode) Fanout Card (in CRATE mode) Fanout Card (in CRATE mode) Unique board for HCAL and possibly ECAL One board per TPG crate HTRHTR HTRHTR HTRHTR DCCDCC … … Rack-to-Rack CAT 7 This scheme allows 18 TPG crates. Is it enough ? Depends also on ECAL. Otherwise need one more layer. Low-skew clock distribution tree dedicated to the Trigger Primitives output (SLB) HTRHTR HTRHTR HTRHTR DCCDCC HTRHTR HTRHTR HTRHTR DCCDCC VME : get histograms results and adjust timing
13-Feb-2004 HCAL TriDAS 5 Fanout board (2 operating modes: Global or Crate) TTC fiber Clk80 Input from GLOBAL Fanout 18 Outputs 40MHz RX_CLK = 40MHz RX_BC0 INT_BC0 RX_CLK = 40MHz RX_BC0 QPLL EXT 80MHz QPLL can run stand-alone TTCrx EXT_BC0 FPGA Delay
13-Feb-2004 HCAL TriDAS 6 Complete path of a 40MHz RX_CLK TTC fiber CLK40_Des1 FPGA CAT7 (RX_CLK, RX_BC0) TTCrx QPLL FPGA TTCrx QPLL Fanout board in Crate-mode Fanout board in Global-mode 3.3V CMOS Path is 3.3V differential PECL unless otherwise stated. Path of RX_BC0 is similar but comes from the FPGA rather the QPLL In the Global-mode card, do not mount the buffers for CLK80 and TTC CAT7 (RX_CLK, RX_BC0, TTC, CLK80) HTR SLB Max skew on HTR traces is 0.7 ns. Spec is: Skew < 12 ns across HCAL and ECAL
13-Feb-2004 HCAL TriDAS 7 HCAL TPG – Current 40MHz Scheme SLB TPG40 SLB TPGBC0 TTC TTCrx Crate80 Serial Optical Data Ref Clk Deserializers (8) 20 Recovered Clk TPG Path SYS40 Clk TTC Broadcast Async Fifo PLL TTC40 x2 XILINX LC Fiber Data Princeton Fanout Card (1/VME crate) SYS80 Clk Princeton Fanout Card (GLOBAL) HCALHCAL
13-Feb-2004 HCAL TriDAS 8 HCAL in SLHC – HTR/SLB Changes to HTR for 80MHz running –HB/HE tower energy takes ~50ns to be collected For 80MHz, adopt a more complicated peak finding algorithm –Use weighted filter to associate energy with crossing Need some simulation effort here to study how best to do this –Will require more cycles inside FPGA Current measurements: 12 ticks 2x will probably not be enough This is where an increase in Latency TIME is critical Will current Xilinx Virtex2 3000s 80MHz? –Probably yes – currently only 50% logic used, 80MHz is not overly fast Can we send data to SLBs at twice the rate? –Probably yes – HTR layout was done by hand with a lot of care –Need to test this on production boards sometime this year
13-Feb-2004 HCAL TriDAS 9 HCAL in SLHC – HTR/SLB (cont) Can we use the same HTRs for 80MHz running? Probably yes. Depends on results of: –Simulation/testbeam data analysis to study 12.5ns filtering schemes –Results of tests on Rev4 HTRs (appearing now) on SLB connectivity –Results of efforts to constrain phase differences for sychronization See Chris Tully’s talk… Note: increase in L1 latency will be necessary
13-Feb-2004 HCAL TriDAS 10 Improvements to Level 1: Assumption: Level 1 will need 2xTPG transmission rate If so, we will need to rebuild the SLBs Consider whether an increase of latency can be put to good use here… –Preprocess TPG before sending to Level 1? –Extend “real” jet triggers to forward region (HF) could be done soon as at test Perhaps finer granularity, jet isolation…ask Tully! –Consider a joint HCAL/ECAL solution….
13-Feb-2004 HCAL TriDAS 11 Current HCAL/ECAL System –ECAL: Data sent to Trigger Concentrator Card (TCC) using fiber G-Link TCC sends data to RCT via SLBs and copper cables running at 1.2Gpbs Data is buffered in the Front End –HCAL: Data buffered in the HTR Sent to the RCT via SLBs, just like ECAL ECAL FE TCC L1 Trigger DAQ DCC HCAL FE HTR DCC Data Flow L1A Flow TPG Flow Data Flow L1A Flow TPG Flow Level 1 Pipeline
13-Feb-2004 HCAL TriDAS 12 First part of Upgrade Focus on ECAL and HCAL commonality via SLB Proposal: –New board: Calorimeter Synch Board (CSB) (and L1 preprocessor…) –Replace SLBs with ~passive high rate transmitter cards (G-link?) –Send raw HCAL data from HTR to new board No need for compression inside HTR – frees up BRAM and logic Accept fiber TPG data from HCAL (and ECAL)? Implement SLB functionality on a single card Send data from CSB to RCT via mezzanine –Could use Vitesse copper for compatibility with current system –Could use Fiber serial output Would need few of these (compared to SLBs) Could accept both ECAL and HCAL data on same board…food for thought
13-Feb-2004 HCAL TriDAS 13 CSB Schematic Fiber input –Gigabit or G-link G-link accommodates ECAL Channel count –At least 48 for HCAL –Enough channels to do summing if used by ECAL Output –Via mezzanine sites Can build small vitesse cards to use with current RCT Can use Fiber for possible future upgrades to RCT Use FPGA with built-in serializer –Current ones have too much latency – investigate? FPGA(s) FIBERFIBER FiberFiber CUCU or RJ45 ethernet
13-Feb-2004 HCAL TriDAS 14 CSB 9U board with fiber inputs FPGAs to implement SLB functionality –Histogramming, synchronization, etc Mezzanine for output to RCT –Can make small vitesse transmitters, or replace with faster optical in the future Added bonus… –Do some preprocessing to offload some of the stuff RCT does now –Save a few clock ticks in latency due to integration –Offload some of the HTR work – might make it able to run at 80MHz? –Level 1 “pre-processing” (depends on ECAL…more in subsequent slides) –Possibility of adding EMF, isolation, etc. to TPG Level 1 preprocessing… –Virtex 2 Pro’s have built-in Motorola processors Can do monitoring, etc. Ethernet access via front-panel –DAQ path for the TPG path –Implement jet triggers in forward direction?
13-Feb-2004 HCAL TriDAS 15 Who? Maryland (Baden etal) University of Virginia (Hirosky) Princeton (Tully, Marlow) –Prototype card for HO Trigger Fiber input Copper inputs Virtex2Pro built in serdes+PPC computer
13-Feb-2004 HCAL TriDAS 16 Fun Can we think up a new Level 1 architecture? –CMS pipeline contains a certain number of events –Current trigger is a pipelined processor Level 1 chews on all “n” events in the pipeline simultaneously –Could we instead consider building a “farm” of “n” processors to chew on 1 event at a time, at least for the calorimeter trigger?
13-Feb-2004 HCAL TriDAS 17 “Asynchronous Level 1” HCAL ECAL PreL1 RCT processors Here we need to put the L1 answers back into a synchronous stream….