1. Slide 1Calorimeter Trigger Upgrade Workshop, 29/11/07Dave.Newbold@cern.ch Track & Calo Trig Simulation D. Newbold, J. Brooke, R. Frazier à Track trigger ‘requirements’ à Track trigger concept n Stacked layers + candidate-driven (‘outside-in’) approach à Trigger simulation à Implications for calo trigger upgrade à Future work… à Goal of giving this talk n The ‘concept’ makes assumptions on the new calo trigger n Are these justified / reasonable / possible? n Who plans to work on track / calo simulation? n How to organise coherent all-detector trigger upgrade simulation?

2. Slide 2Calorimeter Trigger Upgrade Workshop, 29/11/07Dave.Newbold@cern.ch Track Trigger Reqts à Will not rehearse again the motivation for upgrade n Goal: maintain ~LHC trigger performance (thresholds, rates) à Track trigger functions n A stand-alone track trigger is not necessary n For muons: confirmation from tracker of isolated high-pt muon candidates + refinement of p t measurement with extra points For calo: increased rejection of fake e/  objects + refinement of isolation and  jet ID n Rejection of uncorrelated (different primary vertex) double combinations à Constraints Operate within few  s fixed L1 latency - no ‘selective readout’ n Do not add to tracker material / power budget n Reasonable bandwidth for readout / Reasonable processing density n Robust w.r.t background, inefficiency, alignment, etc n Interface to other elements of L1 system Muon / calo trigger upgrade can provide enhanced space / p t resolution?

3. Slide 3Calorimeter Trigger Upgrade Workshop, 29/11/07Dave.Newbold@cern.ch Technical Challenges à On-detector n Raw information from tracker is huge in size n Communication bandwidth likely to drive on-detector power budget n Tracker is a highly integrated electromechanical system Trigger functionality must be integral part of new design from the start à Off-detector: experience from existing L1 trigger n Communications density / data concentration is the key problem n In particular: dealing with overlaps / edges can be very hard n Processing density not a constraint; deep pipelining possible à Heavy on-detector data reduction is required n Communication between tracker layers is probably impractical n Require multiple stand-alone measurements of candidate tracks à Implementation of a track trigger will be challenging! n Focus on reduction of the key parameter: readout bandwidth

4. Slide 4Calorimeter Trigger Upgrade Workshop, 29/11/07Dave.Newbold@cern.ch Trigger Concept: On-Detector à Use ‘stacked layer’ concept: n Presented before in inner pixels context See past talks of J. Jones et al n Use two space-points from sensors separated by some mm Correlate 2D hits for track stub position & slope in  /  Window cut in  excludes low-pt minbias tracks Also cuts digital logic - important  Modularity to match calo trigger towers (0.0875 in  /  ) n One trigger / readout ASIC per TT performs hit correlation n Output 4 candidate stubs ( 3GeV/c) n If >4 candidates, count and flag as possible jet activity Detailed information is not required, since isolation cut already failed

5. Slide 5Calorimeter Trigger Upgrade Workshop, 29/11/07Dave.Newbold@cern.ch Trigger Concept: Layout à Two (or more) specialised trigger planes in ‘outer tracker’ n Low occup. / long lever arm at large r allows coarse-resn sensors Reduced cost and power consumption for sensor + trigger logic n Outer (r=1.2m) plane for direct correlation with muon / calo objects Inner (r=0.6m) plane for b/g rejection, supports  / z measurement n O(250) 2.5Gb/s fibre output – see TWEPP 2007 proceedings for details

6. Slide 6Calorimeter Trigger Upgrade Workshop, 29/11/07Dave.Newbold@cern.ch Trigger Concept: Off-Detector à Divide trigger processing regionally 36 regional subsystems each process a half-detector, 20 degrees in  n Each subsystem process muon, calo and track information n Data-sharing (~25% of data) covers track propagation between regions Simply a duplication of input data - passive optical splitting? à Track finding n Hold track segment info until muon / calo objects are available n Seed track building from candidates; define restricted ROI at inner layer Cuts down enormously on correlation logic n Outer stub eta directions used to identify + request inner layer stubs n A match in eta / phi slope (using beam constraint) flags a physical track NB: No attempt to determine track p t à Output to global trigger n Fixed number of muon / calo / jet candidates with p t / E t, charge, quality n Track-based jet tag for isolation purposes Also possibility of track-count  ID, etc.

7. Slide 7Calorimeter Trigger Upgrade Workshop, 29/11/07Dave.Newbold@cern.ch Processing Topology à Generic FPGA-based board for all subsystem functions

8. Slide 8Calorimeter Trigger Upgrade Workshop, 29/11/07Dave.Newbold@cern.ch Track Trigger Simulation à Need to confront the various ideas with simulation n Highly topical for evaluation of tracker design ‘straw men’ à Some immediate questions for simulation n What are the occupancies for the various tracker layouts? n How many trigger layers? Where? n How to tune the stacked layers? (spacing versus granularity) n Number of candidates per module? Precision of primitives? n How much compression is possible? What scheme to use? n Optimal stub-finding algorithm (given power constraints, etc)? How much can be analogue / digital? n Detailed strategy for correlation with calo / muon trigger? n Sensitivity to alignment, noise, inefficiency, dead channels, etc n Eventually: simulate performance on benchmark channels à This is a great deal of work, phase space is large n Need to focus quickly on most promising ideas + pursue n There is 100% interplay with the overall tracker simulation effort

9. Slide 9Calorimeter Trigger Upgrade Workshop, 29/11/07Dave.Newbold@cern.ch Simulation Tools à Requirements on simulation tools n Clean interface with tracker simulation Discussing with tracker upgrade sim group; trig sim cannot use tracker digis! n Clean interface with calo / muon trigger simulation Evolution of existing emulator? Or completely new code? n Modular – digitisation; trig prim generation; off-detector algorithms n Flexible – assume no particular geometry, granularity, sensor type, etc à Work on a first toy prototype now starting n First pass at tracker hits -> ‘L1TrackHits’ (barrel only at present) No bit-level simulation, but resolutions can be smeared n ‘Perfect’ window based stub-finder, no attempt to emulate digital logic n Can be driven from four-vectors (incl. pileup) or full simulation à First goal: proof-of-principle for ‘perfect stubs’ electron track correlator n Learn lessons for construction of real framework n Work with tracker upgrade sim group to understand technical issues

10. Slide 10Calorimeter Trigger Upgrade Workshop, 29/11/07Dave.Newbold@cern.ch Toy Simulation à Interface to tracker simulation n Watershed is effectively the correlator part of module readout chip n Assume this corresponds to a physically distinct readout path n All trigger simulation takes place in idealised coordinate system n We do assume tracker physical modularity matching trigger towers

11. Slide 11Calorimeter Trigger Upgrade Workshop, 29/11/07Dave.Newbold@cern.ch Existence Proof à 500k gen. level minbias events - barrel only à Av. ~6k hits per BX in 60cm layer from in-time events alone à OOT hit occupancy requires further work

12. Slide 12Calorimeter Trigger Upgrade Workshop, 29/11/07Dave.Newbold@cern.ch Calo Trig I/F - Reqts à Proposed interplay of track and calo n Calo operates as stand-alone trigger Provides fixed number of candidates to track correlator n Correlator searches for matching tracks Begin with calo candidate, propagate inwards through tracker layers n Augments “e cand” with charge, track-based position, z-vertex posn. Augments “  cand” with track based isolation flag Augments  -jet candidates with track count confirmation n Correlator forwards augmented candidates to GT à Requirements n Precision: need at least tower-level position information Can we make use of ‘strip’-level position in EE? May help with propagation n Latency: must leave time for track correlator to operate The correlator latency requirement need not be large Probably dominated by ‘lookup time’ for stub lists in inner layer(s) Preparation of stub lists in the shadow of calo / muon processing

13. Slide 13Calorimeter Trigger Upgrade Workshop, 29/11/07Dave.Newbold@cern.ch Summary & Future Work à Track trigger concepts exist – at ever-increasing levels of detail à Track trigger simulation n Essential as input into tracker design; first toy tools being constructed now n ~1.5FTE in UK, hope to go to ~5 by late 2008 (See UK R&D proposal) DN, J. Brooke, R. Frazier, G. Heath, A. Rose + new postdocs / grad stdts. The proposed tracker-trigger watershed is the correlator ASIC n Bit-level simulation is probably not required at this stage (but soon…) à Calo / track interface n Track trigger relies on calo (and muons) for operation n The hardware is likely to be co-located in regional processors n Need to reflect this in a single, consistent, simulation package à Future work n Continue proof-of-principle simulation work - few weeks timescale n Work closely with tracker upgrade group n Put together a single coherent simulation of upgraded trigger Need to discuss between track trigger proponents, calo and muon experts

14. Slide 14Calorimeter Trigger Upgrade Workshop, 29/11/07Dave.Newbold@cern.ch Backup

15. Slide 15Calorimeter Trigger Upgrade Workshop, 29/11/07Dave.Newbold@cern.ch On-Detector System Parameters à Design for track acceptance for p t > 4GeV/c Largely above minbias spectrum, still good acceptance for jet /  tracks Track curvature lies within ±1TT in  ; limits data-sharing requirements n Sensor doublet must cover all physical high-p t track trajectories I.e. physical overlap of sensors à Hit resolution requirements Resolution in  dictated by practical limit of layer spacing Spacing upper limit from accidental coincidence reduction - to be tuned At 10mm spacing, resolution > 0.5mm n Resolution in z dictated by slope-matching + vertex z-resolution At 10mm spacing, resolution > 2mm is adequate (c.f. ~ 70mm) n Use full-precision 2D layers, also functioning as stereo layers? à Readout requirements n (4b+4b) position, (4b+4b) slope (with offset) will allow stub matching n Implies ~1200 x 10Gb/s readout fibres n Can reduce further (factor of ~10?) using on-detector data compression Zero-suppression; p t sorting; variable-sized payload (asynchronous links)